100% found this document useful (1 vote)

339 views259 pages

Python Ceng240

Uploaded by

Gözdenur Özdemir

Available Formats

Download as PDF, TXT or read online on Scribd

Download as pdf or txt

100% found this document useful (1 vote)

339 views259 pages

Python Ceng240

Uploaded by

Gözdenur Özdemir

Available Formats

Download as PDF, TXT or read online on Scribd

Download as pdf or txt

You are on page 1/ 259

Programming with Python for Engineers

Release 1.0

Sinan Kalkan, Onur Tolga Sehitoglu, Gokturk Ucoluk

Jun 21, 2021

Contents

Preface 1

1 Basic Computer Organization 9

1.1 The von Neumann Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 The Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 The CPU . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4 The Fetch-Decode-Execute Cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5 The Stored Program Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.6 Pros and Cons of the von Neuman Architecture . . . . . . . . . . . . . . . . . . . . . . . . 17
1.7 Peripherals of a computer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.8 The running of a computer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.9 Important Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.10 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2 A Broad Look at Programming and Programming Languages 23

2.1 How do we solve problems with programs? . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3 Data Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4 The World of Programming Languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5 Introducing Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.6 Important Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.7 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.8 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3 Representation of Data 39
3.1 Representing integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.2 Representing real numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.3 Numbers in Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.4 Representing text . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.5 Containers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.6 Representing truth values (Booleans) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.7 Important Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.8 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

4 Dive into Python 57

4.1 Basic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.1.1 Numbers in Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.1.2 Boolean Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.2 Container data (str, tuple, list, dict, set) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

i
4.2.1 Accessing elements in sequential containers . . . . . . . . . . . . . . . . . . . . . 62
4.2.2 Useful operations common to containers . . . . . . . . . . . . . . . . . . . . . . . 63
4.2.3 String . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.2.4 List and tuple . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2.5 Dictionary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.2.6 Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.3 Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.3.1 Arithmetic, Logic, Container and Comparison Operations . . . . . . . . . . . . . . 75
4.3.2 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.3.3 Evaluating Expressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.3.4 Implicit and Explicit Type Conversion . . . . . . . . . . . . . . . . . . . . . . . . 78
4.4 Basic Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.4.1 Assignment Statement and Variables . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.4.2 Variables & Aliasing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4.3 Naming variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.4.4 Other Basic Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.5 Compound Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.6 Basic actions for interacting with the environment . . . . . . . . . . . . . . . . . . . . . . 85
4.6.1 Actions for input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.6.2 Actions for output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.7 Actions that are ignored . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.7.1 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.7.2 Pass statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.8 Actions and data packaged in libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.9 Providing your actions to the interpreter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.9.1 Directly interacting with the interpreter . . . . . . . . . . . . . . . . . . . . . . . 89
4.9.2 Writing actions in a file (script) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.9.3 Writing your actions as libraries (modules) . . . . . . . . . . . . . . . . . . . . . 90
4.10 Important Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.11 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.12 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

5 Conditional and Repetitive Execution 93

5.1 Conditional execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.1.1 if statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.1.2 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.1.3 Nested if statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.1.4 Practice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.1.5 Conditional expression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.2 Repetitive execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.2.1 while statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.2.2 Examples with while statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.2.3 for statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.2.4 Examples with for statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.2.5 continue and break statements . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.2.6 Set and list comprehension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.3 Important Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.4 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

6 Functions 113

ii
6.1 Why define functions? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.2 Defining functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.3 Passing parameters to functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.3.1 Default Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.4 Scope of variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.5 Higher-order functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.6 Functions in programming vs. functions in Mathematics . . . . . . . . . . . . . . . . . . . 119
6.7 Recursion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
6.8 Function Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.9 Programming Style . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6.10 Important Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.11 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.12 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

7 A Gentle Introduction to Object-Oriented Programming 135

7.1 Properties of Object-Oriented Programming . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.1.1 Encapsulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
7.1.2 Inheritance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
7.1.3 Polymorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.2 Basic OOP in Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.2.1 The Class Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
7.2.2 Special Methods/Operator Overloading . . . . . . . . . . . . . . . . . . . . . . . 143
7.2.3 Example 1: Counter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.2.4 Example 2: Rational Number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
7.2.5 Inheritance with Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
7.2.6 Interactive Example: A Simple Shape Drawing Program . . . . . . . . . . . . . . 147
7.2.7 Useful Short Notes on Python’s OOP . . . . . . . . . . . . . . . . . . . . . . . . 148
7.3 Widely-used Member Functions of Containers . . . . . . . . . . . . . . . . . . . . . . . . 149
7.4 Important Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
7.5 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
7.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

8 File Handling 155

8.1 First Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
8.2 Files and Sequential Access . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
8.3 Data Conversion and Parsing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
8.4 Accessing Text Files Line by Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
8.5 Termination of Input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
8.6 Example: Processing CSV Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
8.7 Formatting Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
8.8 Binary Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
8.9 Note on Files, Directory Organization and Paths . . . . . . . . . . . . . . . . . . . . . . . 169
8.10 List of File Class Member Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
8.11 Important Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
8.12 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
8.13 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171

9 Error Handling and Debugging 173

9.1 Types of Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
9.1.1 Syntax Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
9.1.2 Type errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

iii
9.1.3 Run-Time Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
9.1.4 Logical Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
9.2 How to Work with Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
9.2.1 Program with Care . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
9.2.2 Place Controls in Your Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
9.2.3 Handle Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
9.2.4 Write verification code and raise exceptions . . . . . . . . . . . . . . . . . . . . . 182
9.2.5 Debug Your Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
9.2.6 Write Test Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
9.3 Debugging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
9.3.1 Debugging Using Debugging Outputs . . . . . . . . . . . . . . . . . . . . . . . . 184
9.3.2 Handle the Exception to Get More Information . . . . . . . . . . . . . . . . . . . 188
9.3.3 Use Python Debugger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
9.4 Important Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
9.5 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192

10 Scientific and Engineering Libraries 193

10.1 Numerical Computing with NumPy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
10.1.1 Arrays and Their Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . 193
10.1.2 Working with Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
10.1.3 Linear Algebra with NumPy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
10.1.4 Why Use NumPy? Efficiency Benefits . . . . . . . . . . . . . . . . . . . . . . . . 204
10.2 Scientific Computing with SciPy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
10.3 Data handling & analysis with Pandas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
10.3.1 Supported File Formats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
10.3.2 Data Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
10.3.3 Accessing Data with DataFrames . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
10.3.4 Modifying Data with DataFrames . . . . . . . . . . . . . . . . . . . . . . . . . . 210
10.3.5 Analyzing Data with DataFrames . . . . . . . . . . . . . . . . . . . . . . . . . . 211
10.3.6 Presenting Data in DataFrames . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
10.4 Plotting data with Matplotlib . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
10.4.1 Parts of a Figure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
10.4.2 Preparing your Data for Plotting . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
10.4.3 Drawing Single Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
10.4.4 Drawing Multiple Plots in a Figure . . . . . . . . . . . . . . . . . . . . . . . . . . 217
10.4.5 Changing elements of a plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
10.5 Important Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
10.6 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
10.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220

11 An Application: Approximation and Optimization 221

11.1 Approximating Functions with Taylor Series . . . . . . . . . . . . . . . . . . . . . . . . . 221
11.1.1 Taylor Series Example in Python . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
11.2 Finding the Roots of a Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
11.2.1 Newton’s Method for Finding the Roots . . . . . . . . . . . . . . . . . . . . . . . 225
11.2.2 Misc Details on Newton’s Method for the Curious . . . . . . . . . . . . . . . . . . 226
11.2.3 Newton’s Method in Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
11.2.4 Newton’s Method in SciPy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
11.3 Finding a Minimum of Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
11.3.1 Newton’s Method for Finding the Minimum of a Function . . . . . . . . . . . . . 231
11.3.2 Misc Details for the Curious . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232

iv
11.3.3 Newton’s Method in Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
11.3.4 Newton’s Method for Finding Minima in SciPy . . . . . . . . . . . . . . . . . . . 234
11.4 Important Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
11.5 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
11.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235

12 An Application: Solving a Simple Regression Problem 237

12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
12.1.1 Why is regression important? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
12.1.2 The form of the function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
12.2 Least-Squares Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
12.3 Linear Regression with SciPy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
12.3.1 Create Artificial Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
12.3.2 Download and Visualize Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
12.3.3 Fit a Linear Function with SciPy . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
12.3.4 Analyze the Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
12.4 Non-linear Regression with SciPy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
12.4.1 Create Artificial Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
12.4.2 Download and Visualize Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
12.4.3 Fitting a Non-linear Function with SciPy . . . . . . . . . . . . . . . . . . . . . . . 248
12.4.4 Analyzing the Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
12.5 Important Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
12.6 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
12.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250

v
vi
Preface

i. About this book

a. Target audience of the book

This book is intended to be an accompanying textbook for teaching programming to science and engineering
students with no prior programming expertise. This endeavour requires a delicate balance between provid-
ing details on computers & programming in a complete manner and the programming needs of science and
engineering disciplines. With the hopes of providing a suitable balance, the book uses Python as the pro-
gramming language, since it is easy to learn and program. Moreover, for keeping the balance, the book is
formed of three parts:
• Part I: The Basics of Computers and Computing: The book starts with what computation is, intro-
duces both the present-day hardware and software infrastructure on which programming is performed
and introduces the spectrum of programming languages.
• Part II: Programming with Python: The second part starts with the basic building blocks of Python
programming and continues with providing the ground formation for solving a problem in to Python.
Since almost all science and engineering libraries in Python are written with an object-oriented ap-
proach, a gentle introduction to this concept is also provided in this part.
• Part III: Using Python for Science and Engineering Problems: The last part of the book is dedicated
to practical and powerful tools that are widely used by various science and engineering disciplines.
These tools provide functionalities for reading and writing data from/to files, working with data (using
e.g. algebraic, numerical or statistical computations) and plotting data. These tools are then utilized
in example problems and applications at the end of the book.

1
b. How to use the book

This is an ‘interactive’ book with a rather ‘minimalist’ approach: Some details or specialized subjects are
not emphasized and instead, direct interaction with examples and problems are encouraged. Therefore,
rather than being a ‘complete reference manual’, this book is a ‘first things first’ and ‘hands on’ book. The
pointers to skipped details will be provided by links in the book. Bearing this in mind, the reader is strongly
encouraged to read and interact all contents of the book thoroughly.
The book’s interactivity is thanks to Jupyter notebook1 . Therefore, the book differs from a conventional
book by providing some dynamic content. This content can appear in audio-visual form as well as some
applets (small applications) embedded in the book. It is also possible that the book asks the the reader to
complete/write a piece of Python program, run it, and inspect the result, from time to time. The reader is
encouraged to complete these minor tasks. Such tasks and interactions are of great assistance in gaining
acquaintance with Python and building up a self-confidence in solving problems with Python.
Thanks to Jupyter notebook running solutions on the Internet (e.g. Google Colab2 , Jupyter Notebook
Viewer3 ), there is absolutely no need to install any application on the computer. You can directly down-
load and run the notebook on Colab or Notebook Viewer. Though, since it is faster and it provides better
virtual machines, the links to all Jupyter notebooks will be served on Colab.

ii. What is computing?

Computing is the process of inferring data from data. What is going to be inferred is defined as the task. The
original data is called the input (data) and the inferred one is the output (data).
Let us look at some examples:
• Multiplying two numbers, X and Y , and subtracting 1 from the multiplication is a task. The two
numbers X and Y are the input and the result of X × Y − 1 is the output
• Recognizing the faces in a digital picture is a task. Here the input is the color values (3 integers) for
each point (pixel) of the picture. The output is, as you would expect, the pixel positions that belong to
faces. In other words, the output can be a set of numbers.
• The board instance of a chess game, as input, where black has made the last move. The task is to
predict the best move for white. The best move is the output.
• The input is a set of three-tuples which look like <Age_of_death, Height, Gender>. The task, an
optimization problem in essence, is to find out the curve (i.e. the function) that goes through these
tuples in a 3D dimensional space spanned by Age, Height and Gender. As you have guessed already,
the output is the parameters defining the function and an error describing how well the curve goes
through the tuples.
• The input is a sentence to a chatbot. The task is to determine the sentence (the output) that best follows
the input sentence in a conversation.
These examples suggest that computing can involve different types of data, either as input or output: Num-
bers, images, sets, or sentences. Although this variety might appear intimidating at first, we will see that,
by using some ‘solution building blocks’, we can do computations and solve various problems with such a
wide spectrum of data.
1
https://jupyter.org
2
https://colab.research.google.com/
3
https://nbviewer.jupyter.org/

2 Contents
iii. Are all ‘computing machinery’ alike?

Certainly not! This is a common mistake a layman does. There are diverse architectures based on totally
different physical phenomena that can compute. A good example is the brain of living beings, which rely on
completely different mechanisms compared to the micro processors sitting in our laptops, desktops, mobile
phones and calculators.
The building blocks of a brain is the neuron, a cell that has several input channels, called dendrites and a
single output channel, the axon, which can branch like a tree (see Fig. 1).

Fig. 1: Our brains are composed of simple processing units, called neurons. Neurons receive signals (infor-
mation) from other neurons, process those signals and produce an output signal to other neurons. [Drawing
by BruceBlaus - Own work, CC BY 3.0, https://commons.wikimedia.org/w/index.php?curid=28761830]

The branches of an axon, each carrying the same information, connect to other neurons’ dendrites (Fig.
2). The connection with another neuron is called the synapse. What travels through the synapse are called
neurotransmitters. Without going into details, one can simplify the action of neurotransmitters as messengers
that cause an excitation or inhibition on the receiving end. In other words, the neurotransmitters, through a
chemical process along the axon, are released into the synapse as the ‘output’ of the neuron, they ‘interact’
with the dendrite (i.e. the ‘input’) of another neuron and potentially lead to an excitation or an inhibition.

Contents 3
Fig. 2: Neurons ‘communicate’ with each other by transmitting neurotransmitters via synapses. [Drawing
by user:Looie496 created file, US National Institutes of Health, National Institute on Aging created origi-
nal - http://www.nia.nih.gov/alzheimers/publication/alzheimers-disease-unraveling-mystery/preface, Public
Domain, https://commons.wikimedia.org/w/index.php?curid=8882110]

Interestingly enough, the synapse is like a valve, which reduces the neurotransmitters’ flow. We will come

4 Contents
to this in a second. Now, all the neurotransmitters flown in through the input channels (dendrites) have an
accumulative effect on the (receiving) neuron. The neuron emits a neurotransmitter burst through its axon.
This emission is not a ‘what-comes-in-goes-out’ type. It is more like the curve in Fig. 3.

Fig. 3: When a neuron receives ‘sufficient’ amount of signals, i.e. stimulated, it emits neurotransmitters on
its axon, i.e. it fires. [Plot by Original by en:User:Chris 73, updated by en:User:Diberri, converted to SVG
by tiZom - Own work, CC BY-SA 3.0, https://commons.wikimedia.org/w/index.php?curid=2241513]

The throughput of the synapse is something that may vary with time. Most synapses have the ability to
ease the flow over time if the neurotransmitter amount that entered the synapse was constantly high. High
activity widens the synaptic connection. The reverse also happens: Less activity over time narrows the
synaptic connection.
Some neurons are specialized in creating neurotransmitter emission under certain physical effects. Retina
neurons, for example, create neurotransmitters if light falls on them. Some, on the other hand, create physical
effects, like creating an electric potential that will activate a muscle cell. These specialized neurons feed the
huge neural net, the brain, with inputs and receive outputs from it.
The human brain, containing about 1011 such neurons with each neuron being connected to 1000-5000 other
neurons by the mechanism explained above, is a very unique computing ‘machine’ that inspires computa-
tional sciences.
A short video on synaptic communication between neurons
The brain never stops processing information and the functioning of each neuron is only based on signals
(the neurotransmitters) it receives through its connections (dendrites). There is no common synchronization
timing device for the computation: i.e. each neuron behaves on its own and functions in parallel.
An interesting phenomenon of the brain is that the information and the processing are distributed. Thanks
to this feature, when a couple of neurons die (which actually happens each day) no information is lost com-

Contents 5
pletely.
On the contrary to the brain, which uses varying amounts of chemicals (neurotransmitters), the microproces-
sor based computational machinery uses the existence and absence of an electric potential. The information
is stored very locally. The microprocessor consists of subunits but they are extremely specialized in function
and far less in number compared to 1011 all alike neurons. In the brain, changes take place at a pace of 50
Hz maximum, whereas this pace is 109 Hz in a microprocessor.
In Chapter 1, we will take a closer look at the microprocessor machinery which is used by today’s computers.
Just to make a note, there are man-made computing architectures other than the microprocessor. A few to
mention would be the ‘analog computer’, the ‘quantum computer’ and the ‘connection machine’.

iv. What is a ‘computer’?

As you have already noticed, the word ‘computer’ is used in more than one context.
1. The broader context: Any physical entity that can do ‘computation’.
2. The most common context: An electronic device that has a ‘microprocessor’ in it.
From now on, ‘computer’ will refer to the second meaning, namely a device that has a ‘microproces-
sor’.
A computer…
• is based on binary (0/1) representations such that all inputs are converted to 0s and 1s and all outputs are
converted from 0/1 representations to a desired form, mostly a human-readable one. The processing
takes places on 0s and 1s, where 0 has the meaning of ‘no electric potential’ (no voltage, no signal)
and 1 has the meaning of ’some fixed electric potential (usually 5 Volts, a signal).
• consists of two clearly distinct entities: The Central Processing Unit (CPU), also known as the mi-
croprocessor (�P), and a Memory. In addition to these, the computer is connected to or incorporates
other electronic units, mainly for input-output, known as ‘peripherals’.
• performs a ‘task’ by executing a sequence of instructions, called a ‘program’.
• is deterministic. That means if a ‘task’ is performed under the same conditions, it will produce al-
ways the same result. It is possible to include randomization in this process only by making use of a
peripheral that provides electronically random inputs.

v. What is programming?

The CPU (the microprocessor - �P) is able to perform several types of actions:
• Arithmetic operations on binary numbers that represent (encode) integers or decimal numbers with
fractional part.
• Operations on binary representations (like shifting of digits to the left or right; inverting 0s and 1s).
• Transferring to/from memory.
• Comparing numbers (e.g. whether a number n1 larger than n2 ) and performing alternative actions
based on such comparisons.
• Communicating with the peripherals.

6 Contents
• Alternating the course of the actions.
Each such unit action is recognized by the CPU as an instruction. In more technically terms, tasks are solved
by a CPU by executing a sequence of instructions. Such sequences of instructions are called machine codes.
Constructing machine codes for a CPU is called ‘machine code programming’.
But, programming has a broader meaning:
a series of steps to be carried out or goals to be accomplished.
And, as far as computer programming is concerned, we would certainly like these steps to be expressed in
a more natural (more human readable) manner, compared to binary machine codes. Thankfully, there exist
‘machine code programs’ that read-in such ‘more natural’ programs and convert them into ‘machine code
programs’ or immediately carry out those ‘naturally expressed’ steps.
Python is such a ‘more natural way’ of expressing programming steps.

Contents 7
8 Contents
1 | Basic Computer Organization

1.1 The von Neumann Architecture

1.1.1 John von Neumann

From: Oxford Reference4

“Hungarian-born US mathematician, creator of the theory of games and pioneer in the development of the
modern computer. Born in Budapest, the son of a wealthy banker, von Neumann was educated at the uni-
versities of Berlin, Zürich, and Budapest, where he obtained his PhD in 1926. After teaching briefly at the
universities of Berlin and Hamburg, von Neumann moved to the USA in 1930 to a chair in mathematical
physics at Princeton. In 1933, he joined the newly formed Institute of Advanced Studies at Princeton as one
of its youngest professors. By this time he had already established a formidable reputation as one of the most
powerful and creative mathematicians of his day. In 1925 he had offered alternative foundations for set the-
ory, while in his Mathematischen Grundlagen der Quantenmechanik (1931) he removed many of the basic
doubts that had been raised against the coherence and consistency of quantum theory. In 1944, in collab-
oration with Oskar Morgenstern (1902–77), von Neumann published The Theory of Games and Economic
Behaviour. A work of great originality, it is reputed to have had its origins at the poker tables of Princeton and
Harvard. The basic problem was to show whether it was possible to speak of rational behaviour in situations
of conflict and uncertainty as in, for example, a game of poker or wage negotiations. In 1927 von Neumann
proved the important theorem that even in games that are not fully determined, safe and rational strategies
exist. With entry of the USA into World War II in 1941 von Neumann, who had become an American citizen
in 1937, joined the Manhattan project (for the manufacture of the atom bomb) as a consultant. In 1943 he
became involved at Los Alamos on the crucial problem of how to detonate an atom bomb. Because of the
enormous quantity of computations involved, von Neumann was forced to seek mechanical aid. Although
the computers he had in mind could not be made in 1945, von Neumann and his colleagues began to design
Maniac I (Mathematical analyser, numerical integrator, and computer). Von Neumann was one of the first
to see the value of a flexible stored program: a program that could be changed quite easily without altering
the computer’s basic circuits. He went on to consider deeper problems in the theory of logical automata and
finally managed to show that self-reproducing machines were theoretically possible. Such a machine would
need 200 000 cells and 29 distinct states. Having once been caught up in affairs of state von Neumann found
4
https://www.oxfordreference.com/view/10.1093/oi/authority.20110803120234729

9
it difficult to return to a purely academic life. Thereafter much of his time was therefore spent, to the regret of
his colleagues, advising a large number of governmental and private institutions. In 1954 he was appointed
to the Atomic Energy Commission. Shortly after this, cancer was diagnosed and he was forced to struggle
to complete his last work, the posthumously published The Computer and the Brain (1958).”

Fig. 1.1.1: John von Neumann (1903 – 1957)

1.1.2 Components of the von Neumann Architecture

The von Neumann architecture (Fig. 1.1.2) defines the basic structure, or outline, used in most computers
today. Proposed in 1945 by von Neumann, it consists of two distinct units: An addressable memory and a
Central Processing Unit (CPU). All the encoded actions and data are stored together in the memory unit.
The CPU, querying these actions, the so-called instructions, executes them one by one, sequentially (though,
certain instructions may alter the course of execution order).

10 Chapter 1. Basic Computer Organization

Fig. 1.1.2: A block structure view of the von Neumann Architecture.

The CPU communicates with the memory via two sets of wires, namely the address bus and the data bus, plus
a single R/W wire (Fig. 1.1.2). These busses consist of several wires and carry binary information to/from
the memory. Each wire in a bus carries one bit of the information (either a zero (0) or a one (1)). Today’s
von Neumann architectures are working on electricity, and therefore, these zeros and ones correspond to
voltages. A one indicates usually the presence of a 5V and a zero denotes the absence of it.

1.2 The Memory

The memory can be imagined as pigeon holes organized as rows (Fig. 1.2.1). Each row has eight pigeon
holes, each being able to hold a zero (0) or one (1) – in electronic terms, each pigeon hole is capable of
storing a voltage (can you guess what type of an electronical component a pigeon hole is?). Each such row
is named to be of the size byte; i.e., a byte means 8 bits.

1.2. The Memory 11

Fig. 1.2.1: The memory is organized as a stack of rows such that each row has an associated address.

Each byte of the memory has a unique address. When the address input (also called address bus – Fig. 1.1.2)
of the memory is provided a binary number, the memory byte that has this number as the address becomes
accessible through the data output (also called output data bus). Based on W/R wire being set to Write (1)
or Read (0), the action that is carried out on the memory byte differs:
• W/R wire is set to WRITE (1) :
The binary content on the input data bus is copied into the 8-bit location whose address is provided
on the address bus, the former content is overwritten.
• W/R wire is set to READ (0) :
The data bus is set to a copy of the content of 8-bit location whose address is provided on the address
bus. The content of the accessed byte is left intact.
The information stored in this way at several addresses live in the memory happily, until the power is turned
off.
The memory is also referred as Random Access Memory (RAM). Some important aspects of this type of
memory have to be noted:
• Accessing any content in RAM, whether for reading or writing purposes, is only possible when the
content’s address is provided to the RAM through the address bus.
• Accessing any content takes exactly the same amount of time, irrespective of the address of the content.
In todays RAMs, this access time is around 50 nanoseconds.
• When a content is overwritten, it is gone forever and it is not possible to undo this action.

12 Chapter 1. Basic Computer Organization

An important question is who sets the address bus and communicates through the data bus (sends and receives
bytes of data). As depicted in Fig. 1.1.2, the CPU does. How this is done on the CPU side will become clear
in the next section.

1.3 The CPU

The Central Processing Unit, which can be considered as the ‘brain’ of a computer, consists of the following
units:
• Control Unit (CU), which is responsible for fetching instructions from the memory, interpreting (‘de-
coding’) them and executing them. After executing an instruction finishes, the control unit continues
with the next instruction in the memory. This “fetch-decode-execute” cycle is constantly executed by
the control unit.
• Arithmetic Logic Unit (ALU), which is responsible for performing arithmetic (addition, subtraction,
multiplication, division) and logic (less-than, greater-than, equal-to etc.) operations. CU provides the
necessary data to ALU and the type of operation that needs to be performed, and ALU executes the
operation.
• Registers, which are mainly storage units on the CPU for storing the instruction being executed, the
affected data, the outputs and temporary values.
The size and the quantity of the registers differ from CPU model to model. They generally have size in the
range of [2-64] bytes and most registers on today’s most popular CPUs have size 64 bits (i.e. 8 bytes). Their
quantity is not high and in the range of [10-20]. The registers can be broadly categorized into two: Special
Purpose Registers and General Purpose Registers.
Two special purpose registers are worth mentioning to understand how a CPU’s Fetch-Decode-Execute cycle
runs. The first is the so-called Program Counter (PC) and the second is the Instruction Register (IR).
• Input/Output connections, which connect the CPU to the other components in the computer.

1.4 The Fetch-Decode-Execute Cycle

The CPU is in fact a state machine, a machine that has a representation of its current state. The machine,
being in a state, reads the next instruction and executes the instruction according to its current state. The state
consists of what is stored in the registers. Until it is powered off, the CPU follows the Fetch-Decode-Execute
cycle (Fig. 1.4.1) where each step of the cycle is based on its state. The control unit is responsible for the
functioning of the cycle.

1.3. The CPU 13

Fig. 1.4.1: The CPU constantly follows the fetch-decode-execute cycle while the computer is running a
program.

1- The Fetch Phase

The cycle starts with the Fetch Phase. At the beginning of this phase, the CPU has the address (the position
in the memory) of the next instruction in the PC (Program Counter) Register. During this phase, the address
bus is set to this address in the PC register and the R/W wire is set to Read (0). The memory responds to this
by providing the memory content at the given address on the data bus.
How many bytes are sent through the data bus is architecture dependent. Usually it is 4-8 bytes. These bytes
are received into the IR (Instruction Register).
2- The Decode Phase
At the beginning of this phase, the IR is assumed to be holding the current instruction. The content of the
first part of the IR electronically triggers some action. Every CPU has an electronically built-in hard-wired
instruction table in which every possible atomic operation that the CPU can carry out has an associated binary
code, called operation code (opcode in short). This table differs from CPU brand to brand.
There are three types of instructions:
• Data manipulation: Arithmetic/Logic operations on/among registers,
• Data transfer: Memory-to-Register, Register-to-Memory, Register-to-Register transfers,
• Control flow of execution: Instructions that stop execution, jump to a different part of the memory for
next instruction, instead of the next one in the memory.
Let us assume that our instruction looks like this:

Opcode Effected data or address

0001 0110

This is an 8-byte instruction that has the first 4 bits as representing the opcode. The designer could have
designed the CPU such that the opcode 0001 denotes an instruction for reading data from the memory,
writing data to the memory or adding the contents of the two registers etc. The remaining four bits then

14 Chapter 1. Basic Computer Organization

contain the parameters of the instruction, which are the data to be operated on, the address in the memory or
the codes of the registers etc.
Let us assume that this 8-bit example instruction (i.e. the opcode 0001) denotes an addition on two registers
and that the remaining 4 bits encode the registers in question, with 01 denoting one register and 10 the other
register. Prior to the instruction, we can assume the two registers to contain integers, and after the instruction
is executed, one of the registers will be incremented by the amount of the other (by means of integer addition).
Although this was a simple and hypothetical example, it illustrates how modern CPUs can decode an in-
struction and decipher its elements. Though, the length of an instruction and the variety of instructions are
clearly different.
3- The Execute Phase
As the name implies, the electronically activated and initialized circuitry carries out the instruction in this
phase. Depending on the instruction, the registers, the memory or other components are effected. When the
instruction completes, the PC is updated by one unless it was a control flow changing instruction in which
case the PC is updated to the to-be-jumped address in the memory (in some designs, the PC can be updated
in the fetch phase, after fetching the instruction). Not all instructions take the same amount of time to be
carried out. Floating point division, for example, takes much more time compared to others.
A CPU goes through the Fetch-Decode-Execute cycle until it is powered off. What happens at the very
beginning? The electronics of the CPU is manufactured such that, when powered up, the PC register has a
very fixed content. Therefore, the first instruction is always fetched from a certain position.
An intelligent question would be “when does the CPU jump from one state to another?”. One possible
answer is: whenever the previous state is completed electronically, a transition to the next state is performed.
Interestingly, this is not true. The reality is that there is an external input to the CPU from which electronic
pulses are fed. This input is called the system clock and each period of it is named as a clock cycle. The best
performance would be that each phase of the fetch-decode-execute cycle is completed in one-and-only-one
clock cycle. On modern CPUs, this is true for addition instruction, for example. But there are instructions
(like floating point division) which take about 40 clock cycles.
What is length of a clock cycle? CPUs are marked with their clock frequency. For example, Intel’s latest
processor, i9, has a maximal clock frequency of 5GHz (that is 5×109 pulses per second). So, since (period) =
1/(frequency), for this processor a clock cycle is 200 pico seconds. This is such a short time that light would
travel only 6 cm.
A modern CPU has many more features and functional components: interrupts, ports, various levels of
caches are a few of them. To cover them is certainly out of the scope of this course material.

1.5 The Stored Program Concept

In order for the CPU to compute something, the corresponding instructions to do the computation have to
be placed into the memory (how this is achieved will become clear in the next chapter). These instructions
and data that perform a certain task are called a Computer Program. The idea of storing a computer program
into the memory to be executed is coined as the Stored Program Concept.
What does a stored program look like? Below you see a real extract from the memory, a program that
multiplies two integer numbers sitting in two different locations in the memory and stores the result in another
memory location (to save space consecutive 8 bytes in the memory are displayed in a row, the next row
displays the next 8 bytes):

1.5. The Stored Program Concept 15

01010101 01001000 10001001 11100101 10001011 00010101 10110010 00000011
00100000 00000000 10001011 00000101 10110000 00000011 00100000 00000000
00001111 10101111 11000010 10001001 00000101 10111011 00000011 00100000
00000000 10111000 00000000 00000000 00000000 00000000 11001001 11000011
...
11001000 00000001 00000000 00000000 00000000 00000000

Unless you have a magical talent, this should not be understandable to you. It is difficult because it is just a
sequence of bytes. Yes, the first byte is presumably an instruction, but what is it? Furthermore, since we do
not know what it is, we do not know whether it is followed by some data or not, so we cannot say where the
second instruction starts. However, the CPU for which these instructions were written for would know this,
hard-wired in its electronics.
When a programmer wants to write a program at this level, i.e. in terms of binary CPU instructions and binary
data, s/he has to understand and know each instruction the CPU can perform, should be able to convert data
to some internal format, to make a detailed memory layout on paper and then to start writing down each bit
of the memory. This way of programming is an extremely painful job; though it is possible, it is impractical.
Alternatively, consider the text below:

main:
pushq %rbp
movq %rsp, %rbp
movl alice(%rip), %edx
movl bob(%rip), %eax
imull %edx, %eax
movl %eax, carol(%rip)
movl $0, %eax
leave
ret
alice:
.long 123
bob:
.long 456

Though pushq and moveq are not immediately understandable, the rest of the text provides some hints.
alice and bob must be some programmer’s name invention, e.g. denoting variables with values 123 and
456 respectively; imull must have something to do with ‘multiplication’, since only registers can be subject
to arithmetic operations; %edx and %eax must be some denotation used for registers; having uncovered this,
movls start to make some sense: they are some commands to move around data… and so on. Even without
knowing the instruction set, with a small brainstorming we can uncover the action sequence.
This text is an example assembly program. A human invented denotation for instructions and data. An
important piece of knowledge is that each line of the assembler text corresponds to a single instruction. This
assembly text is so clear that even manual conversion to the cryptic binary code above is feasible. Form now
on, we will call the binary code program as a Machine Code Program (or simply the machine code).
How do we automatically obtain machine codes from assembly text? We have machine code programs that
convert the assembly text into machine code. They are called Assemblers.
Despite making programming easier for programmers, compared to machine codes, even assemblers are
insufficient for efficient and fast programming. They lack some high-level constructs and tools that are
necessary for solving problems easier and more practical. Therefore higher level languages that are much
easier to read and write compared to assembly are invented.

16 Chapter 1. Basic Computer Organization

We will cover the spectrum of programming languages in more detail in the next chapter.

1.6 Pros and Cons of the von Neuman Architecture

The von Neumann architecture has certain advantages and disadvantages:

Advantages
• CPU retrieves data and instruction in the same manner from a single memory device. This simplifies
the design of the CPU.
• Data from input/output (I/O) devices and from memory are retrieved in the same manner. This is
achieved by mapping the device communication electronically to some address in the memory.
• The programmer has a considerable control of the memory organization. So, s/he can optimize the
memory usage to its full extent.
Disadvantages
• Sequential instruction processing nature makes parallel implementations difficult. Any parallelization
is actually a quick sequential change in tasks.
• The famous “Von Neumann bottleneck” : Instructions can only be carried out one at a time and se-
quentially.
• Risk of an instruction being unintentionally overwritten due to an error in the program.
An alternative to the von Neumann Architecture is the Harvard Architecture5 which could not resist the test
of time due to crucial disadvantages compared to the von Neumann architecture.

1.7 Peripherals of a computer

Though it is somewhat contrary to your expectation, any device outside of the von Neumann structure,
namely the CPU and the Memory, is a peripheral. In this aspect, even the keyboard, the mouse and the
display are peripherals. So are the USB and ethernet connections and the internal hard disk. Explaining the
technical details of how those devices are connected to the von Neumann architecture is out of the scope of
this book. Though, we can summarize it in a few words.
All devices are electronically listening to the busses (the address and data bus) and to a wire running out
of the CPU (which is not pictured above) which is 1 or 0. This wire is called the port_io line and tells the
memory devices as well as to any other device that listens to the busses whether the CPU is talking to the
(real) memory or not. If it is talking to the memory all the other listeners keep quiet. But if the port_io
line is 1, meaning the CPU doesn’t talk to the memory but to the device which is electronically sensitive to
that specific address that was put on the address bus (by the CPU), then that device jumps up and responds
(through the data bus). The CPU can send as well as receive data from that particular device. A computer
has some precautions to prevent address clashes, i.e. two devices responding to the same address information
in port_io.
Another mechanism aids communication requests initiated from the peripherals. Of course it would be
possible for the CPU from time to time stop and do a port_io on all possible devices, asking them for any
data they want to send in. This technique is called polling and is extremely inefficient for devices that send
asynchronous data (data that is send in irregular intervals): You cannot know when there will be a keyboard
5
https://en.wikipedia.org/wiki/Harvard_architecture

1.6. Pros and Cons of the von Neuman Architecture 17

entry so, in polling, you have to ask very frequently the keyboard device for the existence of any data. Instead
of dealing with the inefficiency of polling, another mechanism is built into the CPU. The interrupt mechanism
is an electronic circuitry of the CPU which has inlets (wires) connected to the peripheral devices. When a
device wants to communicate with (send or receive some data to/from) the CPU they send a signal (1) from
that specific wire. This gets the attention of the CPU, the CPU stops what it is doing at a convenient point
in time, and asks the device for a port_io. So the device gets a chance to send/receive data to/from the CPU.

1.8 The running of a computer

When you power on a computer, it first goes through a start-up process (also called booting), which, after
performing some routine checks, loads a program from your disk called Operating System.

1.8.1 Start up Process

At the core of a computer is the von Neumann architecture. But how a machine code finds its way into the
memory, gets settled there, so that the CPU starts executing it, is still unclear.
When you buy a brand new computer and turn it on for the first time, it does some actions which are traceable
on its display. Therefore, there must be a machine code in a memory which, even when the power is off, does
not lose its content, very much like a flash drive. It is electronically located exactly at the address where the
CPU looks for its first instruction. This memory, with its content, is called Basic Input Output System, or
in short BIOS. In the former days, the BIOS was manufactured as write-only-once. To change the program,
a chip had to be replaced with a new one. The size of the BIOS of the first PCs was 16KB, nowadays it is
about 1000 times larger, 16MB.
When you power up a PC the BIOS program will do the following in sequence:
• Power-On Self Test, abbreviated as POST, which determines whether the CPU and the memory are
intact, identifies and if necessary, initializes devices like the video display card, keyboard, hard disk
drive, optical disc drive and other basic hardware.
• Looking for an operating system (OS): The BIOS program goes through storage devices (e.g. hard disk,
floppy disk, USB disk, CD-DVD drive, etc.) connected to the computer in a predefined order (this
order is generally changeable by the user) and looks for and settles for the first operating system that
it can find. Each storage device has a small table at the beginning part of the device, called the Master
Boot Record (MBR), which contains a short machine code program to load the operating system if
there is one.
• When BIOS finds such a storage device with an operating system, it loads the content of the MBR into
the memory and starts executing it. This program loads the actual operating system and then runs it.
BIOS is nowadays replaced by Unified Extensible Firmware Interface (UEFI) which introduces more capa-
bilities such as faster hardware check, better user interface and more security. With UEFI, MBR is replaced
by GPT (Globally-unique-identifier Partition Table) to allow larger disks, larger partitions (drives) and better
recovery options.

18 Chapter 1. Basic Computer Organization

1.8.2 The Operating System

The operating system is a program that, after being loaded into the memory, manages resources and services
like the use of memory, the CPU and the devices. It essentially hides the internal details of the hardware and
makes the ugly binary machine understandable and manageable to us.
An OS has the following responsibilities:
• Memory Management: Refers to the management of the memory connected to the CPU. In modern
computers, there is more than one machine code program loaded into the memory. Some programs are
initiated by the user (like a browser, document editor, Word, music player, etc.) and some are initiated
by the operating system. The CPU switches very fast from one program (this is called a process) in the
memory to another. The user (usually) does not feel the switching. The memory manager keeps track
of the space allocated by processes in the memory. When a new program (process) is being started, it
has to be placed into the memory. The memory manager decides where it is going to be placed. Of
course, when a process ends, the place in the memory occupied by the process has to be reclaimed;
that is the memory manager’s job. It is also possible that, while running, a process demands additional
space in the memory (e.g. a photoshop-like program needs more space for a newly opened JPG image
file) then the process makes this demand to the memory manager, which grants it or denies it.
• Process (Time) Management: As said above, a modern memory generally contains more than one
machine code program. An electronic mechanism forces the CPU to switch to the Time Manager
component of the OS. At least 20 times a second, the time manager is invoked to make a decision on
behalf of the CPU: Which of the processes that sit in the memory will be run during the next period?
When a process gets the turn, the current state of the CPU (the contents of all registers) is saved to
some secure position in the memory, in association to the last executing process. From that secure
position, the last saved state information which going to take the turn is found and the CPU is set to
that state. Then the CPU, for a period of time executes that process. At the end of that period, the
CPU switches over to the time manager and the time manager makes a decision for the next period.
Which process will get the turn? And so on. This decision making is a complex task. Still there are
Ph.D. level research going on on this subject. The time manager collects some statistics about each
individual process and its system resource utilization. Also there is the possibility that a process has a
high priority associated due to several reasons. The time manager has to solve a kind of optimization
problem under some constraints. As mentioned, this is a complex task and a hidden quality factor of
an OS.
• Device Management: All external devices of a computer have a software plug-in to the operating sys-
tem. An operating system has some standardized demands from devices and these software plug-ins
implement these standardized functionality. This software is usually provided by the device manufac-
turer and is loaded into the operating system as a part of the device installing process. These plug-ins
are named as device drivers.
An Operating System performs device communication by means of these drivers. It does the following
activities for device management:
– Keeps tracks of all devices’ status.
– Decides which process gets access to the device when and for how long.
– Implements some intelligent caching, if possible, for the data communication with the device.
– De-allocates devices.
• File Management: A computer is basically a data processing machine. Various data are produced or
used for very diverse purposes. Textual, numerical, audio-visual data are handled. Handling data also
includes storing and retrieving it on some external recording device. Examples for such recording

1.8. The running of a computer 19

devices are hard disks, flash drives, CDs and DVDs. Data is stored on these devices as files. A file is
a persistent piece of information that has a name, some meta data (e.g. information about the owner,
the creation time, size, content type, etc.) and the data.
The organizational mechanism for how files are stored on devices is called the file system. There are
various alternatives to do this. FAT16, FAT32, NTFS, EXT2, EXT3, ExFAT, HFS+ are a few of about
a hundred (actually the most common ones). Each has its own pros and cons as far as max allowed
file size, security, robustness (repairability), extensibility, metadata, lay out policies and some other
aspects are concerned. Files are most often managed in a hierarchy. This is achieved by a concept of
directories or folders. On the surface (as far as the user sees them), a file system usually consist of
files separated into directories where directories can contain files or other directories.
The file manager is responsible for creation & initialization of a file system, inclusion and removal
of devices from this system and management of all sorts of changes in the file system: Creation,
removal, copying of files and directories, dynamically assigning access rights for files and directories
to processes etc.
• Security: This is basically for the maintenance of the computer’s integrity, availability, and confiden-
tiality. The security of a computer exists at various layers such as:
– maintaining the physical security of the computer system,
– the security of the information the system is in hold of, and
– the security of the network to which the computer is connected.
In all of these, the operating system plays a vital role in keeping the security. Especially the second
item is where the operating system is involved at most. Information, as you know by now, is placed
in the computer in two locations. The internal memory and the external storage devices. The internal
memory is in hold of the processes and the processes should not interfere with each other (unless
specifically intended). Actually in a modern day computer, there can be more than one user working
at the same time on the computer. Their processes running in the memory as well as their files being
on the file system must remain extremely private and separate. Even their existence has to be hidden
from every other user.
Computers are connected and more and more integrated in a global network. This integration is done
on a spectrum of interactions. In the extreme case, a computer can be solely controlled over the
network. Of course this is done by supplying some credentials but, as you would guess, such a system
is prone to malicious attacks. An operating system has to take measures to protect the computer system
from such harms. Sometimes it is the case that bugs of the OS are discovered and exploited in order
to breach security.
• User Interface: As the computer’s user, when you want to do anything you do this by ordering the
operating system to do it. Starting/terminating a program, exploring or modifying the file system,
installing/uninstalling a new device are all done by “talking” to the operating system. For this purpose,
an OS provides an interface, coined as the user interface. In the older days, this was done by typing
some cryptic commands into a typewriter at a console device. Over the years, the first computer with a
Graphical User Interface (GUI) emerged. A GUI is a visual interface to the user where the screen can
host several windows each dedicated to a different task. Elements of the operating system (processes,
files, directories, devices, network communications) and their status are symbolized by icons and the
interactions are mostly via moving and clicking a pointing device which is another icon (usually a
movable arrow) on the screen. The Xerox Alto6 introduced on March 1973 was the first computer that
had a GUI. The first personal computer with a GUI was Apple Lisa7 , introduced in 1983 with a price
6
https://github.com/sinankalkan/CENG240/blob/master/figures/XeroxAlto.jpg?raw=true
7
https://github.com/sinankalkan/CENG240/blob/master/figures/AppleLisa.jpg?raw=true

20 Chapter 1. Basic Computer Organization

of $10,000. Almost three years later, by the end of 1985, Microsoft released its first OS with a GUI:
Windows 1.0. The archaic console typing still exists, in the form a type-able window, called terminal,
which is still very much favoured among programming professionals because it provides more control
for the OS.

1.9 Important Concepts

We would like our readers to have grasped the following crucial concepts and keywords from this chapter:
• The von Neumann Architecture.
• The interaction between the CPU and the memory via address, R/W and data bus lines.
• The crucial components on the CPU: The control unit, the arithmetic logic unit and the registers.
• The fetch-decode-execute cycle.
• The stored program concept.
• Operating system and its responsibilities.

1.10 Further Reading

• Computer Architectures:
– Von Neumann Architecture: http://en.wikipedia.org/wiki/Von_Neumann_architecture
– Harvard Architecture: http://en.wikipedia.org/wiki/Harvard_architecture
– Harvard vs. Von Neumann Architecture: http://www.pic24micro.com/harvard_vs_von_
neumann.html
– Quantum Computer: http://en.wikipedia.org/wiki/Quantum_computer
– Chemical Computer: http://en.wikipedia.org/wiki/Chemical_computer
– Non-Uniform Memory Access Computer: http://en.wikipedia.org/wiki/Non-Uniform_
Memory_Access
• Running a computer:
– Booting a computer: https://en.wikipedia.org/wiki/Booting
– History of operating systems: https://en.wikipedia.org/wiki/History_of_operating_systems
– Operating systems: https://en.wikipedia.org/wiki/Operating_system

1.9. Important Concepts 21

1.11 Exercises

• To gain more insight, play around with the von Neumann machine simulator at http://vnsimulator.
altervista.org
• Using Google and the manufacturer’s web site, find the following information for your desktop/laptop:
– Memory (RAM) size
– CPU type and Clock frequency
– Data bus size
– Address bus size
– Size of the general purpose registers of the CPU
– Harddisk or SSD size and random access time

Opcode Affected address

0001 0100

• Assume that we have a CPU that can execute instructions with the format and size given above.
– What is the number of different instructions that this CPU can decode?
– What is the maximum number of rows in the memory that can be addressed by this CPU?

22 Chapter 1. Basic Computer Organization

2 | A Broad Look at Programming and
Programming Languages

2.1 How do we solve problems with programs?

The von Neumann machine, on which computers’ design is based, makes a clear distinction between instruc-
tion and data (do not get confused by the machine code holding both data and instructions: The data field in
such instructions are generally addresses of the data to be manipulated and therefore, data and instructions
exist as different entities in memory). Due to this clear distinction between data and instruction, the solutions
to world problems were approached and handled with this distinction in mind (Fig. 2.1.1):
“For solving world problems, the first task of the programmer is to identify the information
to be processed to solve the problem. This information is called data. Then, the programmer
has to find an action schema that will act upon this data, carry out those actions according to
the plan, and produce a solution to the problem. This well-defined action schema is called an
algorithm.”
[From: G. Üçoluk, S. Kalkan, Introduction to Programming Concepts with Case Studies in
Python, Springer, 20128 ]
8
https://link.springer.com/book/10.1007/978-3-7091-1343-1

23
Fig. 2.1.1: Solving a world problem with a computer requires first designing how the data is going to be
represented and specifying the steps which yield the solution when executed on the data. This design of
the solution is then written (implemented) in a programming language to be executed as a program such
that, when executed, the program outputs the solution for the world problem. [From: G. Üçoluk, S. Kalkan,
Introduction to Programming Concepts with Case Studies in Python, Springer, 20129 ]

2.2 Algorithm

An algorithm is a step-by-step procedure that, when executed, leads to an output for the input we provided.
If the procedure was correct, we expect the output to be the desired output, i.e. the solution we wanted for
the algorithm to compute.
Algorithms can be thought of as recipes for cooking. This analogy makes sense since we would define a
recipe as a step-by-step procedure for cooking something: Each step performs a little action (cutting, slicing,
stirring etc.) that brings us closer to the outcome, the meal.
This is exactly the case in algorithms as well: At each step, we make a small progress towards the solution
by performing a small computation (e.g. adding numbers, finding the minimum of a set of real numbers etc.).
The only difference with cooking is that each step needs to be understandable by the computer; otherwise,
it is not an algorithm.
The origins of the world ‘algorithm’ The word ‘algorithm’ comes from the Latin word algorithmi, which is
the Latinized name of Al-Khwarizmi. Al-Khwarizmi was a Persian Scientist who has written a book on al-
9
https://link.springer.com/book/10.1007/978-3-7091-1343-1

24 Chapter 2. A Broad Look at Programming and Programming Languages

gebra titled “The Compendious Book on Calculation by Completion and Balancing” during 813–833 which
presented the first systematic solution of linear and quadratic equations. His contributions established alge-
bra, which stems from his method of “al-jabr” (meaning “completion” or “rejoining”). The reason the world
algorithm is attributed to Al-Khwarizmi is because he proposed systematic methods for solving equations
using sequences of well-defined instructions (e.g. “take all variables to the right. divide the coefficients by
the coefficient of x…”) – i.e. using what we call today as algorithms.

Fig. 2.2.1: Muhammad ibn Musa al-Khwarizmi (c. 780 – c. 850)

Are algorithms the same thing as programs?

It is very natural to confuse algorithms with programs as they are both step-by-step procedures. However,
algorithms can be studied and they were invented long before there were computers or programming lan-
guages. We can design and study algorithms without using computers with just a pen and paper. A program,
on the other hand, is just an implementation of an algorithm in a programming language. In other words,
algorithms are designs and programs are the written forms of these designs in programming languages.

2.2.1 How to write algorithms

As we have discussed above, before programming our solution, we first need to design it. While designing
an algorithm, we generally use two mechanisms:
1. Pseudo-codes. Pseudo-codes are natural language descriptions of the steps that need to be followed
in the algorithm. It is not as specific or restricted as a programming language but it is not as free as
the language we use for communicating with other humans: A pseudo-code should be composed of
precise and feasible steps and avoid ambiguous descriptions.
Here is an example pseudo-code:
Algorithm 1. Calculate the average of numbers provided by the user.

Input: N -- the count of numbers

Output: The average of N numbers to be provided
Step 1: Get how many numbers will be provided and store that in N
Step 2: Create a variable named Result with initial value 0
Step 3: Execute the following step N times:
(continues on next page)

2.2. Algorithm 25
(continued from previous page)
Step 4: Get the next number and add it to Result
Step 5: Divide Result by N to obtain the average

2. Flowcharts. As an alternative to pseudocodes, we can use flowcharts while designing algorithms.

Flowcharts are diagrams composed of small computational elements that describe the steps of the
algorithm. An example in Fig. 2.2.2 illustrates what kind of elements are used and how they are
brought together to describe an algorithm.
Flowcharts can be more intuitive to work with. However, for complex algorithms, flowcharts can get very
large and prohibitive to work with.

26 Chapter 2. A Broad Look at Programming and Programming Languages

Fig. 2.2.2: Flowcharts describe relationships by using basic geometric symbols and arrows. The program
start or end is depicted with an oval. A rectangular box denotes a simple action or status. Decision making is
represented by a diamond and a parallelogram of the Input/Output process. A silhouette of a TV tube means
displaying a message. The Internet is portrayed as a cloud.

2.2. Algorithm 27
2.2.2 How to compare algorithms

If two algorithms find the same solution, are they of the same quality? For a second, recall a game we used
to play when we were in primary school: “Guess My Number”.
The rule is as follows: There is a setter and a guesser. The setter sets a number from 1 to 1000 which s/he
does not tell. The guesser has to find this number. At each turn of the game, the guesser can propose any
number from 1 to 1000. The setter answers by one of following:
• HIT: The guesser found the number.
• LESSER: The hidden number is less than the proposed one.
• GREATER: The hidden number is greater than the proposed one.
In how many turns the number is found is recorded. The guesser and the setter switch. This goes on for
some agreed count of rounds. Whoever has a lower total count of turns wins.
Many of you have played this game and certainly have observed that there are three categories of children:
1. Random guessers: Worst category. Usually they cannot keep track of the answers and just based on
the last answer, they randomly utter a number that comes to their mind. Quite possibly they repeat
themselves.
2. Sweepers: They start at either 1 or 1000, and then systematically increase or decrease their proposal,
e.g.:
• -is it 1000? Answer: LESSER
• -is it 999? Answer: LESSER
• -is it 998? Answer: LESSER
… and so on. Certainly at some point such players do get a HIT. There is a group which decreases the
number by two or three as well. With a first GREATER reply, they start to increment by one.
3. Middle seekers: Keeping a possible lower and a possible upper value based on the reply they got, at
every stage they propose the number just in the middle of lower and upper values, e.g.:
• -is it 500? Answer: LESSER
• -is it 250? Answer: LESSER
• -is it 125? Answer: GREATER
• -is it 187? (which was (125+250)/2) Answer: GREATER
• … and so on.
All three categories actually adopt different algorithms, which will find the answer in the end. However, as
you may have realised even as you were a child, the first group performs the worst, then comes the second
group. The third group, if they do not make mistakes, is unbeatable.
In other words, algorithms that aim to solve the same problem may not be of the same “quality”: Some
perform better. This is the case for all algorithms and one of the challenges in Computer Science is to find
“better” algorithms. But, what is “better”? Is there a quantitative measure for “better”ness? The answer is
yes.
Let us look at this in the child game described above. First consider the last group’s algorithm (the middle
seekers). At every turn, this kind of seeker narrows down the search space by a factor of 1/2. Starting with
1000 numbers, the search space is reduced as follows: 1000, 1000/2, 1000/22 , … So, in the worst case,

28 Chapter 2. A Broad Look at Programming and Programming Languages

it will take m turns until 1000/2m gets down to 1 (the one remaining number, which has to be the hidden
number). In other words, in the worst case, 1000/2m = 1 and from this we can derive m = log2 (1000).
For 1000, this means approximately m = 10 turns. If we double the range, m would change only by 1 (yes,
think about it, only 11 turns).
We call such an algorithm of “order log(n)” or more technically, O(log(\)). In our case 1000 determines the
‘size’ of the problem. This is symbolized with n. O(log(\)) is the quantitative information about the algo-
rithm which signifies that the solution time is proportional to log(n). This information about an algorithm
is named as complexity.
What about the sweepers algorithm for the problem above? In the worst case, the sweeper would ask a
question 1000 times (the correct number is at the other end of the sequence). If the size (1000 in our case)
is symbolized with n, then it will take a time proportional to n to reach the solution. In other words this
algorithm’s complexity is O(\).
Certainly the algorithm that has O(log(\)) is better than the one with O(\), which is illustrated in Fig. 2.2.3.
In other words, an O(log(\)) algorithm requires less number of steps and is likely to run faster than the one
with O(\) complexity.

Fig. 2.2.3: A plot of various complexities.

2.3 Data Representation

The other crucial component of our solutions to world problems is the data representation, which deals with
encoding the information regarding the problem in a form that is most suitable for our algorithm.
If our problem is the calculation of the average of grades in a class, then before implementing our solution,
we need to determine how we are going to represent (encode) the grades of students. This is what we are
going to determine in the ‘data representation’ part of our solution and to discuss in Chapter 3.

2.3. Data Representation 29

2.4 The World of Programming Languages

Since the advent of computers, many programming languages have been developed with different designs
and levels of complexity. In fact, there are about 700 programming languages - see, e.g. the list of program-
ming languages10 - that offer different abstraction levels (hiding the low-level details from the programmer)
and computational benefits (e.g. providing built-in rule-search engine).
In this section, we will give a flavour of programming languages in terms of abstraction levels (low-level
vs. high-level – see Fig. 2.4.1) as well as the computational benefits they provide.

Fig. 2.4.1: The spectrum of programming languages, ranging from low-level languages to high-level lan-
guages and natural languages.

2.4.1 Low-level Languages

In the previous chapter, we introduced the concept of machine code program. A machine code program is
an aggregate of instructions and data, all being represented in terms of zeros (0) and ones (1). A machine
code is practically unreadable and very burdensome to create, as we have seen before and illustrated below:

01010101 01001000 10001001 11100101 10001011 00010101 10110010 00000011

00100000 00000000 10001011 00000101 10110000 00000011 00100000 00000000
00001111 10101111 11000010 10001001 00000101 10111011 00000011 00100000
00000000 10111000 00000000 00000000 00000000 00000000 11001001 11000011
...
11001000 00000001 00000000 00000000 00000000 00000000

To overcome this, assembly language and assemblers were invented. An assembler is a machine code pro-
gram that serves as a translator from some relatively more readable text, the assembly program, into machine
code. The key feature of an assembler is that each line of an assembly program corresponds exactly to a sin-
gle machine code instruction. As an example, the binary machine code above can be written in an assembly
language as follows:

main:
pushq %rbp
movq %rsp, %rbp
movl alice(%rip), %edx
movl bob(%rip), %eax
(continues on next page)
10
https://en.wikipedia.org/wiki/List_of_programming_languages_by_type

30 Chapter 2. A Broad Look at Programming and Programming Languages

(continued from previous page)
imull %edx, %eax
movl %eax, carol(%rip)
movl $0, %eax
leave
ret
alice:
.long 123
bob:
.long 456

Pros of assembly:
• Instructions and registers have human recognizable mnemonic words associated. Like integer addition
instruction being ADDI, for example.
• Numerical constants can be written down in human readable, base-10 format, the assembler does the
conversion to internal format.
• Implements naming of memory positions that hold data. In other words, assembly has a primitive
implementation of the variable concept.
Cons of assembly:
• No arithmetic or logical expressions.
• No concept of functions.
• No concept of statement grouping.
• No concept of data containers.

2.4.2 High-level Languages

To overcome the limitations of binary machine codes and the assembly language, more capable Programming
Languages were developed. We call these languages High-level languages. These languages hide the low-
level details of the computer (and the CPU) and allow a programmer to write code in a more human-readable
form.
A high-level programming language (or an assembly language) is defined, similar to a natural language,
by syntax (a set of grammar rules governing how to bring together words) and semantics (the meaning –
i.e. what is meant by the sequences of words in the syntax) associated for the syntax. The syntax is based
on keywords from a human language (due to historical reasons, English). Using human-readable keywords
ease comprehension.
The following example is a program expressed in Python that asks for a Fahrenheit value and prints its
conversion into Celsius:

Fahrenheit = input("Please Enter Fahrenheit value:")

print("Celsius equivalent is:", (Fahrenheit − 32) * 5/9)

Here input and print are keywords of the language. Their semantics is self explanatory. Fahrenheit is a
naming we have chosen for a variable that will hold the input value.
High-level languages (HL-languages from now on) implement many concepts which are not present at the
machine code programming level. Most outstanding features are:

2.4. The World of Programming Languages 31

• human readable form of numbers and strings (like decimal, octal, hexadecimal representations for
numbers),
• containers (automatic allocation for places in the memory to hold, access and name data),
• expressions (calculation formulas based on operators which have precedences the way we are used to
from mathematics),
• constructs for repetitive execution (conditional re-execution of code parts),
• functions,
• facilities for data organization (ability to define new data types based on the primitive ones, organizing
them in the memory in certain layouts).

2.4.3 Implementing with a High-level Language: Interpreter vs. Compiler

We can implement our solution in a high-level programming language in two manners:

1. Compilative Approach. In this approach, a translator, called compiler, takes a high-level program-
ming language program as input and converts all actions in the program into a machine code program
(Fig. 2.4.2). The outcome is a machine code program that can be run any time (by asking the OS to
do so) and does the job described in the high-level language program.
Conceptually this is correct, but actually, this schema has another step in-the-loop. The compiler produces
an almost complete machine code with some holes in it. These holes are about the parts of the code which
is not actually coded by the programmer, but filled in from a pre-created machine code library (it is actually
named as library). A program, named linker fills those holes. The linker knows about the library and patches
in the parts of the code that are referenced by the programmer.

Fig. 2.4.2: A program code in a high-level language is first translated into machine understandable binary
code (machine code) which is then loaded and executed on the machine to obtain the result. [From: G.
Üçoluk, S. Kalkan, Introduction to Programming Concepts with Case Studies in Python, Springer, 201211 ]

2. Interpretive Approach. In this approach, a machine code program, named as interpreter, when run,
inputs and processes the high-level program line by line (Fig. 2.4.3). After taking a line as input, the
11
https://link.springer.com/book/10.1007/978-3-7091-1343-1

32 Chapter 2. A Broad Look at Programming and Programming Languages

actions described in the line are immediately executed; if the action is printing some value, the output
is printed right away; if it is an evaluation of a mathematical expression, all values are substituted
and at that very point-in-time, the expression is evaluated to calculate the result. In other words, any
action is carried out immediately when the interpreter comes to its line in the program. In practice, it is
always possible to write down the program lines into a file, and make the interpreter read the program
lines one by one from that file as well.

Fig. 2.4.3: Interpreted languages (e.g. Python) come with interpreters that process and evaluate each action
(statement) from the user on the run and returns an answer. [From: G. Üçoluk, S. Kalkan, Introduction to
Programming Concepts with Case Studies in Python, Springer, 201212 ]

Which approach is better?

Both approaches have their benefits. When a specific task is considered, compilers generate fast executing
machine codes compared to the same task being carried out by an interpreter. On the other hand compilers
are unpleasant when trial-and-errors are possible while developing the solution. Interpreters, on the other
hand, allow making small changes and the programmer receives immediate responses, which makes it easier
to observe intermediate results and adjust the algorithm accordingly. However, interpreters are slower since
they involve an interpretation component while running the code. Sometimes this slowness is by a factor
of 20. Therefore, the interpretive approach is good for quick implementations whereas using a compiler is
good for computation-intense big projects or time-tight tasks.
12
https://link.springer.com/book/10.1007/978-3-7091-1343-1

2.4. The World of Programming Languages 33

2.4.4 Programming-language Paradigms

As we mentioned before, there are more than 700 programming languages. Certainly some are for academic
purposes and some did not gain any popularity. But there are about 20 programming languages which are
commonly used for writing programs. How do we choose one when implementing our solution?
Picking a particular programming language is not just a matter of taste. During the course of the evolution
of the programming languages, different strategies or world views about programming have also developed.
These world views are reflected in the programming languages.
For example, one world view regards the programming task as transforming some initial data (the initial
information that defines the problem) into a final form (the data that is the answer to that problem) by applying
a sequence of functions. From this perspective, writing a program consists of defining some functions which
are then used in a functional composition; a composition which, when applied to some initial data, yields the
answer to the problem.
This concept of world views are coined as programming paradigms. The Oxford dictionary defines the word
paradigm as follows:
paradigm |’parǝ,dïm|
noun
A world view underlying the theories and methodology of a particular scientific subject.
Below is a list of some major paradigms:
• Imperative: Is a paradigm where programming statements and their composition directly map to the
machine code segments, so that the whole machine code is covered.
• Functional: In this paradigm, solving a programming task is to construct a group of functions so that
their ‘functional composition’ acting on the initial data produces the solution.
• Object oriented: In this paradigm the compulsory separation (due to the von Neumann architecture)
of algorithm from data is lifted, and algorithm and data are reunited under an artificial computational
entity: the object. An object has algorithmic properties as well as data properties.
• Logical-declarative: This is the most contrasting view compared to the imperative paradigm. The idea
is to represent logical and mathematical relations among entities (as rules) and then ask an inference
engine for a solution that satisfies all rules. The inference engine is a kind of ‘prover’, i.e. a program,
that is constructed by the inventor of the logical-declarative programming language.
• Concurrent: A paradigm where independent computational entities work towards the solution of a
problem. For problems that can be solved by a divide-and-conquer strategy, this paradigm is very
suitable.
• Event driven: This paradigm introduces the concept of events into programming. Events are assumed
to be asynchronous and they have ‘handlers’, i.e. programs that carry out the actions associated with
a particular event. Programming graphical user interfaces (GUIs) is usually performed using event-
driven languages: An event in a GUI is generated e.g. when the user clicks the “Close” button, which
triggers the execution of a handler function that performs the associated closing action.
In contrary to the layman programmers’ assumption, these paradigms are not mutually exclusive. Many
paradigms can very well co-exist in a programming language together. At a meta level, we can call them
‘orthogonal’ to each other. This is why we have so many programming languages around. A language can
provide imperative as well as functional and object-oriented constructs. Then it is up to the programmer to
blend them in his or her particular program. As it is with many ‘world views’ among humans, in the field

34 Chapter 2. A Broad Look at Programming and Programming Languages

of programming, fanaticism exists too. You can meet individuals that do only functional programming or
object-oriented programming. We better consider them outliers.
Python, the subject language of this book, supports strongly the imperative, functional and object-orient
paradigms. It also provides some functionality in other paradigms by some modules.

2.5 Introducing Python

After having provided background on the world of programming, let us introduce Python: Although it is
widely known to be a recent programming language, Python’s design, by Guido van Rossum13 , dates back
to 1980s, as a successor of the ABC programming language. The first version was released in 1991 and
with the second version released in 2000, it started gaining a wider interest from the community. After it
was chosen by some big IT companies as the main programming language, Python became one of the most
popular programming languages.
An important reason for Python’s wide acceptance and use is its design principles. By design, Python is a
programming language that is easier to understand and to write but at the same time powerful, functional,
practical and fun. This has deep roots in its design philosophy (a.k.a. The Zen of Python14 ):
“Beautiful is better than ugly. Explicit is better than implicit. Simple is better than complex.
Complex is better than complicated. Readability counts. [… there are 14 more]”
Python is multi-paradigm programming language, supporting imperative, functional and object-oriented
paradigms, although the last one is not one of its strong suits, as we will see in Chapter 7. Thanks to its
wide acceptance especially in the open-source communities, Python comes with or can be extended with an
ocean of libraries for practically solving any kind of task.
The word ‘python’ was chosen as the name for the programming language not because of the snake species
python but because of the comedy group Monty Python15 . While van Possum was developing Python, he
read the scripts of Monty Python’s Flying Circus and thought ‘python’ was “short, unique and mysterious”16
for the new language. To make Python more fun to learn, earlier releases heavily used phrases from Monty
Python in example programming codes.
With version 3.9 being released in October 2020 as the latest version, Python is one of the most popular
programming languages in a wide spectrum of disciplines and domains. With an active support from the
open-source community and big IT companies, this is not likely to change in the near future. Therefore, it
is in your best interest to get familiar with Python if not excel in it.
This is how the Python interpreter looks like at a Unix terminal:

$ python3
Python 3.8.5 (default, Jul 21 2020, 10:48:26)
[Clang 11.0.3 (clang-1103.0.32.62)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>>

The three symbols >>> indicate that the interpreter is ready to collect our computational demands, e.g.:

13
https://en.wikipedia.org/wiki/Guido_van_Rossum
14
https://en.wikipedia.org/wiki/Zen_of_Python
15
https://en.wikipedia.org/wiki/Monty_Python
16
https://docs.python.org/2/faq/general.html#why-is-it-called-python

2.5. Introducing Python 35

>>> 21+21
42

where we asked what was 21+21 and Python responded with 42, which is one small step for a man but one
giant leap for mankind.

2.6 Important Concepts

We would like our readers to have grasped the following crucial concepts and keywords from this chapter:
• How we solve problems using computers.
• Algorithms: What they are, how we write them and how we compare them.
• The spectrum of programming languages.
• Pros and cons of low-level and high-level languages.
• Interpretive vs. compilative approach to programming.
• Programming paradigms.

2.7 Further Reading

• The World of Programming chapter available at: https://link.springer.com/chapter/10.1007/

978-3-7091-1343-1_1
• Programming Languages:
– For a list of programming languages: http://en.wikipedia.org/wiki/Comparison_of_
programming_languages
– For a comparison of programming languages: http://en.wikipedia.org/wiki/Comparison_of_
programming_languages
– For more details: Daniel P. Friedman, Mitchell Wand, Christopher Thomas Haynes: Essentials
of Programming Languages, The MIT Press 2001.
• Programming Paradigms:
– Introduction: http://en.wikipedia.org/wiki/Programming_paradigm
– For a detailed discussion and taxonomy of the paradigms: P. Van Roy, Programming Paradigm
for Dummies: What Every Programmer Should Know, New Computational Paradigms for Com-
puter Music, G. Assayag and A. Gerzso (eds.), IRCAM/Delatour France, 2009 http://www.info.
ucl.ac.be/~pvr/VanRoyChapter.pdf
– Comparison between Paradigms: http://en.wikipedia.org/wiki/Comparison_of_programming_
paradigms

36 Chapter 2. A Broad Look at Programming and Programming Languages

2.8 Exercise

• Draw the flowchart for the following algorithm:

Step 1: Get a list of N numbers in a variable named Numbers

Step 2: Create a variable named Sum with initial value 0
Step 3: For each number i in Numbers, execute the following line:
Step 4: if i > 0: Add i to Sum
Step 5: if i < 0: Add the square of i to Sum
Step 5: Divide Sum by N

• What is the complexity of Algorithm 1 (in Section 2.2)?

• What is the complexity of the following algorithm?

Step 1: Get a list of N numbers in a variable named Numbers

Step 2: Create a variable named Mean with initial value 0
Step 3: For each number i in Numbers, execute the following line:
Step 4: Add i to Mean
Step 5: Divide Mean by N
Step 6: Initialize a variable named Std with value 0
Step 7: For each number i in Numbers, execute the following line:
Step 8: Add the square of (i-Mean) to Std
Step 9: Divide Std by N and take its square root

• Assuming that a step of an algorithm takes 1 second, fill in the following table for different algorithms
for different input sizes (n):
Input Size
O(log \)
O(\)
O(\ log \)
O(\∈ )
O(\∋ )
O(∈\ )
O(\\ )
n = 102
n = 103
n = 104
n = 105
• Assume that we have a parser than can process and parse natural language descriptions (without any
syntactic restrictions) for programming a computer. Given such a parser, do you think we would use
natural language to program computers? If no, why not?
• For each situation below, try to identify which paradigm is more suitable compared to the others:
• Writing a program which should take an image as input, an RGB color value and find all pixels in the
image that match the given color.

2.8. Exercise 37
• Writing a theorem proving program.
• Writing the auto pilot program flying an airplane.
• Writing a document editing program as an alternative to Microsoft Word.

38 Chapter 2. A Broad Look at Programming and Programming Languages

3 | Representation of Data

Fig. 3.1: Solving a world problem with a computer requires first designing how the data is going to be
represented and specifying the steps which yield the solution when executed on the data. This design of
the solution is then written (implemented) in a programming language to be executed as a program such
that, when executed, the program outputs the solution for the world problem. [From: G. Üçoluk, S. Kalkan,
Introduction to Programming Concepts with Case Studies in Python, Springer, 201217 ]

39
At this stage, you may be wondering what ‘structured data’ is and how it differs from ‘data’. As we already
know, data is stored in the memory. Let us illustrate this with an example: Assume that we have a table full
of rows such that each row holds an angle value and its cosine value (both expressed as decimal numbers).
How would you organize the storing of these rows in the memory? Two straightforward options are:
a) Row-by-row:
Anglevalue1
Cosinevalue1
Anglevalue2
Cosinevalue2
..
.
Anglevaluen
Cosinevaluen
b) Column-by-Column:
Anglevalue1
Anglevalue2
..
.
Anglevaluen
Cosinevalue1
Cosinevalue2
..
.
Cosinevaluen
Apart from this ordering, there is the issue of whether we should sort the values. If yes by which one: the
angle or the cosine values? In a descending order or ascending order? All these questions are what we try
to determine in a structured data.
There are other ways to organize data even for the example data as simple as the above example table.
Unfortunately, these are beyond the scope of this book.
Now, we will look into the atomic structure of data representation: i.e. how integers, floating points and
characters are represented.
The electronic architecture used for the von Neumann machine, namely the CPU and the memory, is based on
the presence and absence of a fixed voltage. We conceptualize this absence and presence by ‘0’ and ‘1’. This
means that any data that is going to be processed on this architecture has to be converted to a representation
of ‘0’s and ’1’s. We, though, keep in mind, that any ’1’ (or ‘0’) in our representation is actually the presence
(or absence) of a voltage in a specific point of the electronic circuitry.

3.1 Representing integers

Mathematics already has a solution for representing integers in binary via base-2 calculation or representa-
tion. The idea behind base-2 representation is the same as representing base-10 (decimal) integers. Namely,
we have a sequence of digits, each of which can be either 0 or 1. Counting starts from the right-most digit.
A ‘1’ in a position k means “additively include a value of 2k ”:
P osition : n n − 1 ... 2 1 0
M eaning : 2n 2n−1 ... 22 21 20
The following is an example for expressing the integer 181 in base-2:
17
https://link.springer.com/book/10.1007/978-3-7091-1343-1

40 Chapter 3. Representation of Data

In other words, (181)10 = (10110101)2 .
Converting a decimal number into binary
A decimal number can be easily converted into binary by repeatedly dividing the number by 2, as shown in
Fig. 3.1.1. At each step, the quotient of the previous step is divided by two. This division is repeated until
the quotient is zero. At this point, the binary sequence of remainders is the representation of the decimal
number.
Having doubts? You can easily cross-check your calculation by multiplying each bit by its value (2i if the
bit is at position i) and sum up the values (like we did above).

Fig. 3.1.1: The division method for converting a decimal number into binary.

This looks easy. However, this is just a partial solution to the ‘binary representation of integers’ problem: It
does not answer how we can represent negative integers.

3.1. Representing integers 41

3.1.1 Sign-Magnitude Notation

Decimal arithmetics provide the minus sign (-) for representing negativity. However, electronics of the von
Neumann machinery requires that the minus is also represented by either a ‘1’ or ‘0’. We can reserve a bit,
for example the left-most digit, for this purpose. If we do this, when the left-most digit is a ‘1’, the rest
of the digits are encoding the magnitude of the ‘negative’ integer. One point to be mentioned is that this
requires a fixed size (a fixed number of digits) for the ‘integer representation’. In this way, the electronics
can recognize the sign bit and the magnitude part. This is called the sign-magnitude notation.
Sadly, this notation has certain disadvantages:
1. Addition and subtraction
Consider adding two integers. Based on their signs, we have the following possibilities:
• positive + positive
• negative + positive
• positive + negative
• negative + negative
Then, to be able to perform addition, the electronics must do the following:
• if both integers are positive then the result is positive, obtained by adding the magnitudes (and not
setting the sign bit).
• if both integers are negative, then the result is negative, obtained by adding the magnitudes and setting
the sign bit to negative.
• otherwise, if one is negative and the other is positive then:
1. find the bigger magnitude,
2. subtract the smaller magnitude from the bigger one, obtain the result,
3. if the bigger magnitude integer was negative the result is negative (set the sign bit) else don’t set
the sign bit in result.
This was only for the case of addition, a similar electronic circuitry is needed for subtracting two integers.
This technique has the following drawbacks: * Requires separate electronics for subtraction. * Requires
electronics for magnitude-based comparison and an algorithm implementation for setting the sign bit. *
Electronically, it has to differentiate among addition and subtraction.
2. Representing number zero
Another limitation of the sign-magnitude notation is that number zero (0)10 has two different representations,
e.g. in a 4-bit representation:
• (1 000)2 ==> (−0)10
• (0 000)2 ==> (+0)10
In modern computers, we use a method that does not have these drawbacks.

42 Chapter 3. Representation of Data

3.1.2 Two’s Complement Representation

Two’s complement representation can be considered as an extension of the sign-magnitude notation:

• The positive integers are represented by their base-2 representation.
• For negative integers, we need a one-to-one mapping for all negative integers bounded by value to a
binary representation (also bounded by value), so that :
1. The sign bit is set to 1, and
2. When the whole binary representation (including the sign bit) is treated as a single binary number
it operates correctly under addition. When the result, obtained purely by addition, produces a
sign bit, this means the result is the encoding of a negative integer.
There are two alternatives for this mapping: One’s-complement and two’s complement. In this section, we
will introduce the more popular one, namely the two’s complement representation:
If we are given n binary digits to be used (for a 32-bit computer, n = 32; for a 64-bit computer, it is 64), then
we are able to represent integer values in the range [−2n−1 , 2n−1 −1]. Then, two’s complement representation
of an integer can be obtained as follows:
• If the integer is positive simply, convert it to base-2.
• If it is negative: Let the magnitude (its absolute value) be p, then
1. convert p to base-2
2. negate this base-2 representation by flipping all 1s to 0s and all 0s to 1s: 1 ↔ 0.
3. add 1 to the result of the negation.

3.1. Representing integers 43

Here are some examples for 8-bit numbers (note that the valid decimal range is [−128, 127]):

Integer 2′ s complement Pos./Neg. Action:Binary Result

0 00000000 − Direct base-2 encoding:
1 00000001 P ositive Direct base-2 encoding:
−1 11111111 N egative magnitude:00000001
inverse:11111110
+1:11111111
−2 11111110 N egative magnitude:00000010
inverse:11111101
+1:11111110
18 00010010 P ositive Direct base-2 encoding:
−18 11101110 N egative magnitude:00010010
inverse:11101101
+1:11101110
(3.1.1)
−47 11010001 N egative magnitude:00101111
inverse:11010000
+1:11010001
−111 10010001 N egative magnitude:01101111
inverse:10010000
+1:10010001
111 01101111 P ositive Direct base-2 encoding:
112 01110000 P ositive Direct base-2 encoding:
−128 10000000 N egative magnitude:10000000
inverse:01111111
+1:10000000
−129 Not Calculated N egative (Out of valid range):
128 Not Calculated P ositive (Out of valid range):

3.1.3 Why does Two’s Complement work?

The Two’s Complement method may sound arbitrary at first. However, there are solid reasons for why it
works. To be able to explain why it works, let us first revisit our basic computer organization from Chapter
1:
The CPUs can only understand and work with fixed-length representations: Assume that our computer is an
8-bit computer such that registers in the CPU holding data and the arithmetic-logic unit can only work with
8 bits. The fixed-length design of the CPU has a severe implication. Consider the following 8-bit number
(which is 28 − 1) in a register in the CPU:
1 1 1 1 1 1 1 1
If you add 1 to this number, the result would be:
0 0 0 0 0 0 0 0
In other words, we lose the 9th bit since the CPU cannot fit that into the 8-bit representation. Mathematically,
this arithmetic corresponds to modular arithmetic: For our example, the modulo value is 28 and the addition
that we just performed can be written mathematically as:

(28 − 1) + 1 ≡ 0. (mod 28 ) (3.1.2)

44 Chapter 3. Representation of Data

What does this have to do with the Two’s Complement method? Consider taking the negative of a number
a in a 2n -modulo system:

−(a) ≡ 0 − a ≡ 2n − a. (mod 2n ) (3.1.3)

Now let us rewrite this in a form that will seem familiar to us:
2n − a ≡ 2n − 1 + 1 − a ≡ (2n − 1 − a) + 1
|{z} (mod 2n )
| {z } (3.1.4)
Invert the bits of a Add 1 to the inverted bits

In other words, Two’s Complement representation uses the negative value of a number relying on the modular
arithmetic, i.e. the fixed-length representation of the CPUs. This technique is not new and was used in
mechanical calculators long before there were computers.

3.1.4 Benefits of the Two’s Complement Representation

Let us revisit the limitations of the sign-magnitude representation:

• Addition and subtraction: With the two’s complement method, we do not need to check the signs
of the numbers and perform addition and subtraction using just an addition circuitry. E.g. (+2)10 +
(−3)10 is just equal to (−1)10 without doing anything extra other than plain addition (for a 4-bit
representation):
(0010)2 (+2)10
(1101)2 (−3)10
(3.1.5)
+
(1111)2 (−1)10
• Representation of +0 and -0: Another issue with the sign-magnitude representation was that +0 and
-0 had different representations. In the Two’s Complement representation, we see that this is resolved
– for example (for a 4-bit representation):
• (+0)10 = (0000)2
• (−0)10 = −(0000)2 = (1111)2 + (1)2 = (0000)2 , where we used Two’s Complement to convert
−(0000)2 into (1111)2 + (1)2 by flipping the bits and adding 1.
The fifth bit is lost because we have only four bits for representation.

3.1.5 PRACTICE TIME

Please follow the Colab link at the top of the page to have a practical session on Two’s Complement repre-
sentation.

3.2 Representing real numbers

Floating point is the data type used to represent non-integer real numbers. On today’s computers, all non-
integer real numbers are represented using the floating point data type. Since all integers are real numbers,
we could represent integers also using the floating point data type. Although you could do so, this is usually
not preferred, since floating point operations are more time consuming compared to integer operations. Also,
there is the danger of precision loss, which we will discuss later.

3.2. Representing real numbers 45

Almost all processors have adopted the IEEE 754 binary floating point standard for binary representation
of floating point numbers. The standard allocates 32 bits for the representation, although there is a recent
64-bit definition which is based on the same layout idea just with some more bits.
Let us see how we represent a floating point number with an example. Let us consider a decimal number
with fraction: 12263.921875. This number can be represented in binary as two binary numbers: the whole
part in binary and the fractional part in binary. Then, we can join them with a period.
To see this, let us first dissect decimal (base-10) 12263.921875:

104 103 102 101 100 . 10−1 10−2 10−3 10−4 10−5 10−6
(3.2.1)
1 2 2 6 3 . 9 2 1 8 7 5

Keeping the denotational similarity, but switching to (base-2), in other words to binary, we can express the
same number as:
213 212 211 210 29 28 27 26 25 24 23 22 21 20 . 2−1 2−2 2−3 2−4 2−5 2−6
1 0 1 1 1 1 1 1 1 0 0 1 1 1 . 1 1 1 0 1 1
(3.2.2)

How did we obtain the fractional part? By multiplying the fractional by two until we obtain zero for the
fraction part, as illustrated in Fig. 3.2.1. This is in certain ways the reverse of what we did for converting the
whole part into binary, in the previous section.

Fig. 3.2.1: The multiplication method for converting a fractional number into binary.

At this stage, it is worth mentioning that it is not always possible to convert the fractional part into binary
with finite number of bits. In other words, it is quite possible to have a finite number of fractional digits in
one base, and infinitely many in another base for the same value. We will come back to this point later.
Depending on the size of the whole part, the period could be anywhere. Therefore, the next step towards
obtaining the IEEE 754 representation is to reposition the period that separates the whole part from the
fraction, to be placed just after the leftmost ‘1’. To be able to do this without changing the value of the
number, a multiplicative factor of 2n has to be introduced, where n is how many bits the period is moved.
If it is moved to the left, it has a positive value, otherwise it is negative. This factor is important, because it
will contribute to the representation.
For our example, the period has to be moved left by 13 digits. Therefore, the multiplicative factor (to keep

46 Chapter 3. Representation of Data

the value the same) is ×2+13 . At this stage our representation has become:

20 . 2−1 2−2 2−3 2−4 2−5 2−6 2−7 2−8 2−9 2−10 2−11 2−12 2−13 2−14 2−15 2−1
2+13 × 1 . 0 1 1 1 1 1 1 1 0 0 1 1 1 1 1 1 0
(3.2.3)

Since this is doable for all values (except 0.0, which will be dealt with with an exception), there is no need
to keep a record of the whole part, i.e. the only remaining ’1’ value to the left of the period. Also, the period
is always there, so we can simply drop it. The mantissa part of the representation is obtained by keeping
exactly the first 23 bits of what is left. This leads to the following 23 bits for our example (note the extra
zeros at the end, added to fill complete the representation to 23 bits):
0 1 1 1 1 1 1 1 0 0 1 1 1 1 1 1 0 1 1 0 0 0 0

3.2.1 The IEEE754 Representation

Fig. 3.2.2: The 32-bit IEEE754 floating point representation.

The exponent of the multiplicative factor, namely n (which can be negative) in 2n , becomes a part of the
representation – see Fig. 3.2.2. To get rid of the minus sign problem, a constant value, 127, is added to n.
This value becomes the exponent part of the representation.
Adding a constant value to a number to be able to represent positive and negative numbers in binary is
called the excess representation, or k-excess representation, with k being the constant number that is added,
e.g. k = 127. Why we use k-excess representation is going to be clear when we compare it with two’s
complement representation in Table 3.2.1. You should see from the table that if the binary representation of
a decimal number is larger, the decimal number is also larger with the k-excess representation; however, this
is not the case for the two’s complement representation. This is important for comparing two floating point
numbers: Without decoding the whole floating point representation, which is expensive, we can just look at
the k-excess representation of the exponents and the fractional parts to compare numbers.

3.2. Representing real numbers 47

Table 3.2.1: A comparison between the k-excess representation
and the two’s complement representation.
Decimal number k-excess (k = 8) Two’s complement
7 1111 0111
6 1110 0110
5 1101 0101
4 1100 0100
3 1011 0011
2 1010 0010
1 1001 0001
0 1000 0000
−1 0111 1111
−2 0110 1110
−3 0101 1101
−4 0100 1100
−5 0011 1011
−6 0010 1010
−7 0001 1001
−8 0000 1000

In our example the factor was 213 . Adding k = 127 yields 13 + 127 = 140, which has an 8-bit represen-
tation of 10001100. This will become the exponent portion (the 8 green bits in Figure 3.4) of the IEEE754
representation.
It is possible that the value that is going to be represented is a negative value. This information is stored as
a ‘1’ in the first bit of the representation. For a positive value, a ‘0’ bit is used.
Finally, our example gets a IEEE754 representation as:
0 1 0 0 0 1 1 0 0 0 1 1 1 1 1 1 1 0 0 1 1 1 1 1 1 0 1 1 0 0 0 0
How is real number ‘0.0’ represented?
Floating point number zero is represented as all zero bits in the positional range [1-31]. The zeroth bit,
namely the sign bit, can be either ‘0’ or ‘1’.

3.2.2 Information loss in floating-point representations

The condition for a floating point number to be represented “exactly” by the IEEE754 standard is for the
number to be equal to:

X
m
f= digitk × 2k , (3.2.4)
k=n

where m and n are two integers so that |m−n| ≤ 24 (it is fine if the equation is not very clear – we will show
some examples below). In addition to this, there is the constraint that |n| ≤ 127 (think about why). Nothing
can be done about the second constraint, but as far as the first constraint is concerned, practically we can
approximate the number leaving off (truncating) the less significant bits (those to the right, or mathematically
those with a smaller k value in the sum above), so that |m − n| ≤ 24 is satisfied.
Actually, many rational numbers, even, are not expressible in the form of the sum above: As an example, try
deriving the representation of 4.1 and see whether you can represent it in binary with finite number of bits.

48 Chapter 3. Representation of Data

√ √
In addition, we have infinitely many irrational numbers which do not have finite fractions: e.g. 2, 3, π, e.
To be able to use all these numbers in computers, we approximate them. Do not worry, it is something quite
common in applied sciences.
Approximation comes with some dangers: When you subtract two truncated numbers which are close to
each other, then the result you obtain has a high imprecision. This also is the case with multiplication and
division of relatively big numbers.
Though we have not started working with Python yet, we will display some self explanatory examples and
comment on them:

• >>> 0.9375 - 0.9

0.03749999999999998

The first line is some input that we typed in, to be carried out, and the following line is what the Python
interpreter returned as the result. Surprise! (Actually a bad one!) In contrary to our expectation, the
result is not 0.0375.
PN The reason is that 0.9375 is one of the rare numbers that could be represented as
a finite sum k=1 (digitk · 2−k ), digitk ∈
0, 1 such that N stays in the IEEE representation limit. Actually for 0.9375, digit = [1, 1, 1, 1] and
N = 4.
On the other hand, quite unexpectedly, 0.9 is not so. It cannot be expressed as a similar sum where N
stays in the IEEE representation limit. Actually, N extents to +∞ for 0.9. Hence the digitn sequence
has to be truncated before it can stored in the IEEE representation. The bits that do not fit into the
representation are simply ignored: In other words, the number loses its precision.
This is combined with a similar representation problem of 0.0375, which leads to another loss and we
arrive at 0.03749999999999998. Bottom line: many numbers cannot exactly be represented in the
IEEE754 format, and this causes precision loss.
• Things can get even worse. Consider:

>>> 2000.0041 - 2000.0871

-0.0829999999998563

>>> 2.0041 - 2.0871

-0.08299999999999974

Actually both results should have been -0.0830. Despite having an imprecision, the imprecision is
not consistent. This is because the loss in the first example (the one where the whole part is 2000) is
bigger than the loss in the second one since 2000 needs more bits to be represented compared to 2.
• Pi (π) is a transcendental number. The fractional part never stops, in any base. Let us give it a shot on
the Python interpreter:

>>> PI = 3.141592653589793238462643383279502884197169399375105820974944592307816406286
>>> print(PI)
3.141592653589793

Ok, from this we understand that the IEEE representation could only accommodate so many bits for
the 15 places in the fractional part. That looks quite precise, but let us take a look at the sin and cos
values:

3.2. Representing real numbers 49

>>> sin(PI)
1.2246467991473532e-16
>>> cos(PI)
-1.0

Interestingly, we received a slight error for the sine value, which is different from 0.0. But when it
comes to the cosine value, we were lucky that imprecision somewhat cancelled out and gave us the
correct result.
• We are thought since primary school that addition is associative. Hence A + (B + C) is the same as
(A + B) + C. In floating point arithmetic, this may not be so:

>>> A = 1234.567
>>> B = 45.67834
>>> C = 0.0004
>>> AB = A + B
>>> BC = B + C
>>> print (AB+C)
1280.2457399999998

>>> print (A+BC)

1280.2457400000001

This is again a combination of the precision loss phenomena introduced above. Most of the interme-
diate steps of a calculation have precision losses of their own.
As a final word about using floating point numbers, it is worth stressing a common mistake commonly made
but has nothing to do with the precision loss mentioned. Let us assume you provide your program the Pi
number to be 3.1415. You do your calculations and obtain floating point numbers with 14-15 digit fractional
parts. Knowing about the precision loss, you assume that maybe a couple of the last digits are wrong but at
least 10 digits after the decimal point are correct. However, this is a mistake: You made an approximation
in the 5th digit after the decimal point in number Pi which will propagate through your calculation. It can
get worse (for example if you subtract two very close numbers and use it in the denominator), but can never
get better. Your best chance is to get a correct result in the 4th digit after the decimal point. The following
digits, as far as precision is concerned, are bogus.
So, what can we do? Here are some rules of thumb about using floating point numbers:
• It is in your best interest to refrain from using floating points. If it is possible transform the problem
to the integer domain.
• Use the most precise type of floating point provided by your high-level language, some languages
provide you with 64 bit or even 128 bit floats, use them.
• Use less precision floating points only when you are in short of memory.
• Subtracting two floating points close in value has a potential danger.
• If you add or subtract two numbers which are magnitude-wise not comparable (one very big the other
very small), it is likely that you will lose the proper contribution of the smaller one. Especially when
you iterate the operation (repeat it many times), the error will accumulate.
• You are strongly advised to use well-known, commonly used scientific computing libraries instead of
coding floating point algorithms by yourself.

50 Chapter 3. Representation of Data

3.2.3 PRACTICE TIME

Please follow the Colab link at the top of the page to have a practical session on the IEEE754 floating point
representation.

3.3 Numbers in Python

Python provides the following representations for numbers:

• Integers: You can use integers as you are used to from your math classes. Interestingly Python adopts
a seamless internal representation so that integers can effectively have any number of digits. The
internal mechanism of Python switches from the CPU-imposed fixed-size integers to some elaborated
big-integer representation silently when needed. You do not have to worry about it. Furthermore, bear
in mind that “73.” is not an integer in Python. It is a floating point number (73.0). An integer cannot
have a decimal point as part of it.
• Floating point numbers (float in short): In Python, numbers which have decimal point are taken
and represented as floating point numbers. For example, 1.45, 0.26, and -99.0 are float but 102 and -8
are not. We can also use the scientific notation (a × 10b ) to write floating point numbers. For example,
float 0.0000000436 can be written in scientific notation as 4.36 × 10−8 and in Python as 4.36E-8 or
4.36e-8.
• Complex numbers: In Python, complex numbers can be created by using j after a floating point
√ imaginary part: e.g. 1.5-2.6j for the complex number (1.5 + 2.6i).
number (or integer) to denote the
The j symbol (or i) represents −1. There are other ways to create complex numbers, but this is the
most natural way, considering your previous knowledge from high school.

3.4 Representing text

As we said in the first lines of this chapter, programming is mostly about a world problem which generally
includes human related or interpretable data to be processed. These data do not consist of numbers only but
can include more sophisticated data such as text, sound signals and pictures. We leave the processing of
sound and pictures out of this book’s scope. Text, though, is something we have to study.

3.4.1 Characters

Written natural languages consist of basic units called graphemes. Alphabetic letters, Chinese-Japanese-
Korean characters, punctuation marks, numeric digits are all graphemes. There are also some basic actions
that go commonly hand in hand with textual data entry. “Make newline”, “Make a beep sound”, “Tab”,
“Enter” are some examples. These are called “unprintables”.
How can we represent graphemes and unprintables in binary? Graphemes are heavily culture dependent.
The shapes do not have a numerical foundation. As far as computer science is concerned, the only way to
represent such information in numbers is to make a table and build this table into electronic input/output
devices. Such a table will have two columns: The graphemes and unprintables in one column and the
assigned binary code in the other, e.g.:

Grapheme or Unprintable Binary Code

.. ..
. .

3.3. Numbers in Python 51

Throughout the history of computers, there has been several such tables, mainly constructed by computer
manufacturers. In time, most of them vanished and only one survived: The ASCII (American Standard
Code for Information Interchange) table which was developed by the American National Standards Institute
(ANSI). This American code, developed by Americans, is naturally quite ‘American’. It incorporates all
characters of the American-English alphabet, including, for example, the dollar sign, but stops there. The
table does not contain a single character from another culture (for example, even the pound sign ‘£’ is not in
the table).
The ASCII table has 128 lines. It maps 128 American graphemes and unprintables to 7-bit long codes. Since
the 7-bit long code can be interpreted also as a number, for convenience, this number is also displayed in the
ASCII table – see Fig. 3.4.1.

Fig. 3.4.1: The ASCII table.

Do not worry, you do not have to memorize it, even professional computer programmers do not. However,
some properties of this table has to be understood and kept in mind:
• The general layout of the ASCII table:

52 Chapter 3. Representation of Data

Dec. Range Property
0-31 Unprintables
32 Space char.
33-47 Punctuations
48-57 Digits 0-9
58-64 Punctuations
65-90 Upper case letters
91-96 Punctuations
97-122 Lower case letters
123-127 Punctuations

• There is no logic in the distribution of the punctuations.

• It is based on the English alphabet, characters of other languages are simply not there. Moreover, there
is no mechanism for diacritics.
• Letters are ordered and uppercase letters come first in the table (have a lower decimal value)
• Digits are also ordered but are not represented by their numerical values. To obtain the numerical
value for a digit, you have to subtract 48 form its ASCII value.
• The table is only and only about 128 characters, neither more nor less. There is nothing like Turkish-
ASCII, French-ASCII. The extensions, where the 8th bit is set has nothing to do with the ASCII table.
• Python makes use of ASCII character representation.
The frustrating discrepancies and shortcomings of the ASCII table have led the programming society to
seek a solution. A non-profit group, the Unicode Consortium, was founded in the late 80s with the goal of
providing a substitute for the current character tables, which is also compliant (backward compatible) with
them. The Unicode Transformation Format (UTF) is their suggested representation scheme.
This UTF representation scheme has variable length and may include components of 1-to-4 8-bit wide (in
the case of UTF-8) or 16-bit wide components of 1-to-2 (in the case of UTF-16). UTF is now becoming
part of many recent high-level language implementations, including Python, Java, Perl, TCL, Ada95 and
C#, gaining wide popularity.

3.4.2 Strings

Text is as vital a data as numbers are. Text is expressed as character sequences. These sequences are named
as strings. But we have here a problem. As introduced, numbers (integers and floating points) have a niche
in the CPU. There are instructions designed for them: we can store and retrieve them to/from the memory;
we can perform arithmetical operations among them. A character data can be represented and processed as
well because they are mapped to one byte integers. But when it comes to strings, the CPU does not have any
facility for them.
The only reasonable way is to store the codes of each character that make up a string into the memory
in consecutive bytes. Does this solve the problem of ‘representation’? Unfortunately no. The trouble is
determining how to know where the string ends. Two methods come to mind:
1. Prior to the string characters, store the length (the number of characters in the string) as an integer of
fixed number of bytes.
2. Store a special byte value, which is not used to represent any other character, at the end of the string
characters.

3.4. Representing text 53

3.5 Containers

The representation of a string is maybe the most simple representation of a data that is not directly supported
by the CPU. However, it gives an idea for what can be done in similar cases. The key concept is to find a
layout that is a mapping from the space of the data to an organization in the memory.
Also notice that whatever is placed in the memory has an ‘address’ and the value of the address can also
become a part the organization. As far as the string is concerned, the usage of the ‘address concept’ was
simple: A single address marks the start of the string. When we want to process the string, we go to this
address and then start to process the character codes sequentially.
It is possible to have data organizations which include addresses of other data organizations. In other words,
it is possible to jump from one group of data to some other in the memory. This type of organizations are
name Data Structures in Computer Science. So, string is a data structure. As far as Python is concerned,
there are several other data structures. They are coined in Python as containers. In addition to strings Python
provide lists, tuples, sets and dictionaries as containers.

3.6 Representing truth values (Booleans)

Boolean is another data type that has its roots in the very structure of the CPU. Answers to all questions asked
to the CPU are either true or false. The logic of a CPU is strictly based on binary evaluation system. This
logic system is coined as Boolean logic. It was introduced by George Boole in his book “The Mathematical
Analysis of Logic” (1847).
It is tightly connected to the binary 0 and 1 concepts. In all CPUs, falsity is represented with a 0 whereas
truth is represented with a 1 and on some with any value which is not 0.

3.7 Important Concepts

We would like our readers to have grasped the following crucial concepts and keywords from this chapter:
• Sign-magnitude notation and two’s complement representation for representing integers.
• The IEEE754 standard for representing real numbers.
• Precision loss in representing floating point numbers.
• Representing characters with the ASCII table.
• Representing truth values.

3.8 Further Reading

• Two’s Complement: http://en.wikipedia.org/wiki/Two%27s_complement

• The Method of Complements: https://en.wikipedia.org/wiki/Method_of_complements
• Excess-k Representation: https://en.wikipedia.org/wiki/Offset_binary
• IEEE 754 Floating Point Standard: http://en.wikipedia.org/wiki/IEEE_754-2008
• ASCII: http://en.wikipedia.org/wiki/ASCII

54 Chapter 3. Representation of Data

• UTF—UCS Transformation Format: http://en.wikipedia.org/wiki/UTF-8

3.9 Exercises

• By hand, find the 5-bit Two’s Complement representation of the following numbers: 4, −5, 1,−0, 11.
• Represent 6 and (-7) in a 5-bit sign-magnitude notation and add them up in binary, without looking at
their signs.
• Find the IEEE 754 32-bit representation of the following floating point numbers: 3.3, 3.37, 3.375.

3.9. Exercises 55
56 Chapter 3. Representation of Data
4 | Dive into Python

(C) Copyright Notice: This chapter is part of the book available athttps://pp4e-book.github.io/and copying,
distributing, modifying it requires explicit permission from the authors. See the book page for details:https:
//pp4e-book.github.io/
After having covered some background on how a computer works, how we solve problems with computers
and how we can represent data in binary, we are ready to interact with Python and to learn representing data
and performing actions in Python.
The Python interpreter that we introduced in Chapter 2 waits for our computational demands if not executing
something already. Those demands that describe an algorithm have to be expressed in Python as a sequence
of ‘actions’ where each action can serve two broad purposes:
• Creating or modifying data: These actions take in data, perform sequential, conditional or repetitive
execution on the data, and produce other data. Computations that form the bases for our solutions are
going to be these kinds of actions.
• Interacting with the environment: Our solutions will usually involve interacting with the user or the
peripherals of the computer to take in (input) data or take out (output) data.
Irrespective of its purpose, an action can be of two types:
• Expression: An expression (e.g. 3 + 4 * 5) specifies a calculation, which, when evaluated (per-
formed), yields some data as a result. An expression can consist of:
– basic data (integer, floating point, boolean etc.) or container data (e.g. string, list, set etc.).
– expressions involving operations among data and other expressions.
– functions acting on expressions.
• Statement: Unlike an expression, a statement does not return data as a result and can be either basic
or compound:
– Basic statement: A basic statement can be e.g. for storing the result of an expression in a memory
location (an assignment statement for further use (in following actions), deleting an item from
a collection of data etc. Each statement has its special syntax that generally involves a special
keyword.
– Compound statement: Compound statements are composed of other statements and executing
the compound statement means executing the statements in the compound.
Naming and Printing Data
For better illustrations and examples, we will use two concepts in the first part of the chapter before they are
introduced. Let us briefly describe them and leave the coverage of the details to their individual sections:
• Variables: In programming languages, we can give a name to data and use that name to access it, e.g.:

57
>>> a = 3
>>> 10 + a
13

We call such names variables and the action a = 3 is called an assignment. We defer a more detailed
coverage until Section 4.4.
• Printing data: Python provides the print() function to display data items on screen:

print(item1, item2, ..., itemN)

For example:

>>> print('Python', 'is', 'so', 'fun')

Python is so fun

4.1 Basic Data

Let us remember what basic data types we had:

• Numbers
– Integers
– Floating points
– Complex numbers
• Booleans
Python is a language in which all arithmetic operations among the same type of numbers are pro-
vided very much as expected from our math knowledge. Furthermore, mixed type operations (e.g. subtract-
ing a floating point number form an integer) are also defined.
In programming, being able to ask questions about data is of vital importance. The atomic structures for ask-
ing questions are the comparison operations (e.g. “is a value equal to another value”, or “is a value greater
than another value”, etc.). Operators that serve these purposes do exist in Python and provide resulting val-
ues that are True or False (Booleans). It is also possible to combine such questions under logical opera-
tions. The and, or and not operators stand for conjunction, disjunction and negation, correspondingly. Need-
less to say, these operators also return Boolean values.

4.1.1 Numbers in Python

Python provides the following representations for numbers (the following is an essential reminder from the
previous chapter):
• Integers: You can use integers as you are used to from your math classes. Interestingly Python adopts
a seamless internal representation so that integers can effectively have any number of digits. The
internal mechanism of Python switches from the CPU-imposed fixed-size integers to some elaborated
big-integer representation silently when needed. You do not have to worry about it. Furthermore, bear
in mind that “73.” is not an integer in Python. It is a floating point number (73.0). An integer cannot
have a decimal point as part of it.

58 Chapter 4. Dive into Python

• Floating point numbers (float in short): In Python, numbers which have decimal point are taken
and represented as floating point numbers. For example, 1.45, 0.26, and -99.0 are float but 102 and -8
are not. We can also use the scientific notation (a × 10b ) to write floating point numbers. For example,
float 0.0000000436 can be written in scientific notation as 4.36 × 10−8 and in Python as 4.36E-8 or
4.36e-8.
• Complex numbers: In Python, complex numbers can be created by using j after a floating point
√ imaginary part: e.g. 1.5-2.6j for the complex number (1.5 − 2.6i).
number (or integer) to denote the
The j symbol (or i) represents −1. There are other ways to create complex numbers, but this is the
most natural way, considering your previous knowledge from highschool.
More on Integers and Floating Point Numbers
Python provides int data type for integers and float data type for floating point numbers. You can easily
play around with int and float numbers and check their type as follows:

>>> 3+4
7
>>> type(3+4)
<class 'int'>
>>> type(4.1+3.4)
<class 'float'>

where <class 'int'> indicates type int.

In Python version 3, integers do not have fixed-size representation and their size is only limited by your
available memory. In Python version 2, there were two integer types: int, which used the fixed length
representation supported by the CPU, and long type, which was unbounded. Since this book is based on
Python version 3, we will assume that int refers to an unbounded representation.
As for the float type, Python uses the 64-bit IEEE754 standard, which allows representing numbers in the
range [2.2250738585072014E-308, 1.7976931348623157E+308].
Useful Operations:
The following operations can be useful while working with numbers:
• abs(<Number>): Takes the absolute value of the number.
• pow(<Number1>, <Number2>): Takes the power of <Number1>, i.e. <Number1><Number2>
• round(<FloatNumber>): Rounds the floating point number to the closest integer.
• Functions from the math library: sqrt(), sin(), cos(), log(), etc (see the Python documentation18
for a full list). This requires importing from the built-in math library first as follows:

>>> from math import *

>>> sqrt(10)
3.1622776601683795
>>> log10(3.1622776601683795)
0.5

18
https://docs.python.org/3/library/math.html

4.1. Basic Data 59

4.1.2 Boolean Values

Python provides the bool data type which allows only two values: True and False. For example:

>>> type(True)
<class 'bool'>
>>> 3 < 4
True
>>> type(3 < 4)
<class 'bool'>

Also due to decades of programming experience, Python converts several instances of other data types to
some certain boolean values, if used in place of a boolean value. For example,
• 0 (the integer zero)
• 0.0 (the floating point zero)
• "" (the empty string)
• [] (the empty list)
• {} (the empty dictionary or set)
are interpreted as False. All other values of similar kinds are interpreted as True.
Useful Operations:
With boolean values, we can use not (negation or inverse), and and or operations:

>>> True and False

False
>>> 3 > 4 or 4 < 3
False
>>> not(3 > 4)
True

and returns a True value only if both of its operands are True, otherwise it returns False. or returns True
if any or both of its operands are True. The following table gives result of the boolean operations for the
given operand pair:

a b a and b a or b
True True True True
True False False True
False True False True
False False False False

60 Chapter 4. Dive into Python

4.2 Container data (str, tuple, list, dict, set)

Up to this point we have seen the basic data types. They are certainly needed for computation but many
world problems for which we seek computerized solutions need more elaborate data. Just to mention a few:
• Vectors
• Matrices
• Ordered and unordered sets
• Graphs
• Trees
Vectors and matrices are used in almost all simulations/problems of the physical world; sets are used to keep
any property information as well as orders of items, equivalences; graphs are necessary for many spatial
problems; trees are vital to representing hierarchical relational structures, action logics, organizing data for
a quick search.
Python provides five container types for these:
1. String (``str``): A string can hold a sequence of characters or only a single character. A string cannot
be modified after creation.
2. List (``list``): A list can hold ordered sets of all data types in Python (including another list). A list’s
elements can be modified after creation.
3. Tuple (``tuple``): The tuple type is very similar to the list type but the elements cannot be modified
after creation (similar to strings).
4. Dictionary (``dict``): A very very efficient method to form a mapping from a set of numbers,
booleans, strings and tuples to any set of data in Python. Dictionaries are easily modifiable and ex-
tendable. Querying the mapping of an element to the ‘target’ data is performed in almost constant time
(regardless of how many elements the dictionary has). In Computer Science terms, it is a hash table.
5. Set (``set``): The set type is equivalent to sets in mathematics. The element order is undefined. (We
deemphasize the use of set).
The first three, namely String, List and Tuple are called sequential containers. They consist of consecutive
elements indexed by integer values starting at 0. Dictionary is not sequential, element indexes are arbitrary.
For simplicity we will abbreviate sequential containers as s-containers.
All these containers have external representations which are used in inputting and outputting them (with all
their content). Below we will walk through some examples to explain them.
Mutability vs. Immutability
Some container types are created ‘frozen’. After creating them you can wholly destroy them but you cannot
change or delete their individual elements. This is called immutability. Strings and tuples are immutable
whereas lists, dictionaries and sets are mutable. With a mutable container, adding new elements and changing
or deleting existing ones is possible.

4.2. Container data (str, tuple, list, dict, set) 61

4.2.1 Accessing elements in sequential containers

All containers except set reveal their individual elements by an indexing mechanism with brackets:

Fig. 4.2.1: How elements of a container are accessed

For the s-containers, the index is an ordinal number where counting starts with zero. For dictionaries, the
index is a Python data item from the source (domain) set. A negative index (usable only on s-containers) has
a meaning that the (counting) value is relative to the end. A negative index can be converted to a positive
index by adding to the negative value the length of the container. It is nothing but index obtained by adding
the length of the container.

Below we have an s-container that has a length of (n + 1)

(careful: indexing started at 0!):

···
□0 □1 □n−1 □n
[0] [1 ] ··· [n − 1] [n] (4.2.1)

[−(n + 1)][−n] ··· [−2] [−1]

As you surely have observed, when you add (n + 1) to the negative index you obtain the positive one.
Slicing
s-containers provide a rich mechanism, called slicing, that allows accessing multiple elements at once.
Thanks to this mechanism, you can define a start and end index and obtain the portion that lies in between
(Fig. 4.2.2):
• The element at the start index is the first to be accessed.
• The end index is where accessing stops (the element at the end index is not accessed – i.e. the end
index is not inclusive).
• It is also possible to optionally define an increment (jump amount) between indexes. After the element
at [start] is accessed first, [start + increment] is accessed next. This goes on until the accessed
position is equal or greater than the end index. For negative indexing, a negative increment has to
work from the bigger index towards the lesser, so (start index > end index) is expected.

Fig. 4.2.2: Accessing multiple elements of an s-container is possible via the slicing mechanism, which spec-
ifies a starting index, an ending index and an index increment between elements.

62 Chapter 4. Dive into Python

Below, with the introduction of strings, we will have extensive examples on slicing.
If the s-container is immutable (e.g. string and tuple containers) then slicing creates a copy of the sliced data.
Otherwise, i.e. if the s-container is mutable (i.e. the ‘list’ container) then slicing provides direct access to the
original elements and therefore, they can be updated, which updates the original s-container.

4.2.2 Useful operations common to containers

The following operations are common to all or a subset of containers:

1- Number of elements:
For all containers len() is a built-in function that returns the count of elements in the container that is given
as argument to it, e.g.:

>>> len("Five")
4

2- Concatenation:
String, tuple and list data types can be combined using ‘+’ operation:

where the containers need to be of the same type. For example:

>>> "Hell" + "o"

'Hello'

3- Repetition: String, tuple and list data types can be repeated using “*” operation:

where the container is copied ‘’ many times. For example:

>>> "Yes No " * 3

'Yes No Yes No Yes No '

4- Membership: All containers can be checked for whether they contain a certain item as an element using
in and not in operations:

<item> not in <Container>

Of course, the result is either True or False. For dictionaries in tests if the domain set contains the element,
for others it simply tests if element is a member.

4.2. Container data (str, tuple, list, dict, set) 63

4.2.3 String

As was explained in the previous chapter, a string is used to hold a sequence of characters. Actually, it is a
container where each element is a character. However, Python does not have a special representation for a
single character. Characters are represented externally as strings containing a single character only.
Writing strings in Python
In Python a string is denoted by enclosing the character sequence between a pair of quotes (') or double
quotes ("). A string surrounded with triple double quotes (""" . . . """) allows you to have any combination
of quotes and line breaks within a sequence, and Python will still view it as a single entity.
Here are some examples:
• "Hello World!"
• 'Hello World!'
• 'He said: "Hello World!" and walked towards the house.'
• "A"
• """ Andrew said: "Come here, doggy". The dog barked in reply: 'woof' """
The backslash (\) is a special character in Python strings, also known as the escape character. It is used
in representing certain, the so called, unprintable characters: (\t) is a tab, (\n) is a newline, and (\r) is a
carriage return. Table 4.2.1 provides the full list.

Table 4.2.1: The list of escape characters in Python.

Escape Sequence Meaning
\\ Backslash (\)
\' Single quote (')
\" Double quote (")
\a ASCII Bell (BEL)
\b ASCII Backspace (BS)
\f ASCII Formfeed (FF)
\n ASCII Linefeed (LF)
\r ASCII Carriage Return (CR)
\t ASCII Horizontal Tab (TAB)
\v ASCII Vertical Tab (VT)
\ooo ASCII character with octal value ooo
\xhh ASCII character with hex value hh
\uhhhh UNICODE character with hex value hhhh

Conversely, prefixing a special character with (\) turns it into an ordinary character. This is called escaping.
For example, (\') is the single quote character. 'It\'s raining' therefore is a valid string and equivalent
to "It's raining". Likewise, (") can be escaped: "\"hello\"" is a string that begins and ends with the
literal double quote character. Finally, (\) can be used to escape itself: (\\) is the literal backslash character.
For '\onnn', nnn is a number in base 8, for '\xnn' nn is a number in base 16 (including letters from A to F
as digits for values 10 to 15).
In Python v3, all strings use the Unicode representation where all international symbols are possible, e.g.:

64 Chapter 4. Dive into Python

>>> a = "Fıstıkçı şahap"
>>> a
'Fıstıkçı şahap'

Examples with strings

Let us look at some examples to see what strings are in Python and what we can do with them:

>>> "This is a string"

"This is a string"
>>> "This is a string"[0]
'T'
>>> s = "This is a string"
>>> print(s[0])
T
>>> print(s[0],s[1],s[8],s[14],s[15])
T h a n g

Since strings are immutable an attempt to change a character in a string will badly fail:

>>> s = "This is a string"

>>> print(s)
This is a string
>>> s[2] = "u"
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: 'str' object does not support item assignment

Bottom line: You cannot change a character in a created string.

Let us go through a sequence of examples. We encourage you to run the following examples in the Colab
version of this chapter. Feel free to change the text and rerun the examples.

my_beautiful_string = "The quick brown fox jumps over the lazy dog"
print("THE STRING:", my_beautiful_string, len(my_beautiful_string), "CHARACTERS")

THE STRING: The quick brown fox jumps over the lazy dog 43 CHARACTERS

my_beautiful_string[0]

'T'

my_beautiful_string[4]

'q'

my_beautiful_string[0:4]

'The '

4.2. Container data (str, tuple, list, dict, set) 65

my_beautiful_string[:4]

'The '

my_beautiful_string[4:]

'quick brown fox jumps over the lazy dog'

my_beautiful_string[10:15]

'brown'

my_beautiful_string[:-5]

'The quick brown fox jumps over the laz'

my_beautiful_string[-8:-5]

'laz'

my_beautiful_string[:]

'The quick brown fox jumps over the lazy dog'

my_beautiful_string[::-1]

'god yzal eht revo spmuj xof nworb kciuq ehT'

my_beautiful_string[-6:-9:-1]

'zal'

my_beautiful_string[0:15:2]

'Teqikbon'

Strings are used to represent non-mathematical textual information. Common places where strings are used
are:
• Textual communation in natural language with the user of the program.
• Understandable labeling of parts of data: City names, names of individuals, addresses, tags, labels.
• Denotation needs of human-to-human interactions.

66 Chapter 4. Dive into Python

Useful operations with strings:
• String creation: In addition to using quotes for string creation, the str() function can be used to create
a string from its argument, e.g.:

>>> str(4)
'4'
>>> str(4.578)
'4.578'

• Concatenation, repetition and membership:

>>> 'Programming' + ' ' + 'with ' + 'Python is' + ' fun!'
'Programming with Python is fun!'
>>> 'really fun ' * 10
'really fun really fun really fun really fun really fun really fun really fun really␣
,→fun really fun really fun '
>>> 'fun' in 'Python'
False
>>> 'on' in 'Python'
True

• Evaluate a string: If you have a string that is an expression describing a computation, you can use the
eval() function to evaluate the computation and get the result, e.g.:

>>> s = '3 + 4'

>>> eval(s)
7

Deletion and Insertion from/to strings

Since strings are immutable, this is not possible. The only way is to create a new string, making use of slicing
and concatenation operation (using +), then replacing the new created string in same place of the former one.
For example:

>>> a = 'Python'
>>> a[0] = 'S'
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: 'str' object does not support item assignment
>>> b = 'S' + a[1:]
>>> b
'Sython'

4.2.4 List and tuple

Both list and tuple data types have a very similar structure on the surface: They are sequential containers that
can contain any other data type (including other tuples or lists) as elements. The only difference concerning
the programmer is that tuples are immutable whereas lists are mutable. As previously said, being immutable
means that, after being created, it is not possible to change, delete or insert any element in a tuple.
Lists are created by enclosing elements into a pair of brackets and separating them with commas,
e.g. ["this", "is", "a", "list"]. Tuples are created by enclosing elements into a pair of parenthe-

4.2. Container data (str, tuple, list, dict, set) 67

ses, e.g. ("this", "is", "a", "tuple"). There is no restriction on the elements: They can be any data
(basic or container).
Let us look at some examples for lists:
• [9,3,1,-1,6]
• []
• [2020]
• [3.1415, 2.718281828]
• [["pi,3.1415"], ["e",2.718281828], 1.41421356]
• ['the' 'quick','brown','fox','jumped','over','the','lazy','dog']
• [10, [5, [3, [[30, [[30, [], []]], []]], []], [8, [], []]], [30, [], []]]
• [[1,-1,0],[2,-3.5,1.1]]
and some examples for tuples:
• ('north','east','south','west')
• ()
• ('only',)
• ('A',65,"1000001","0x41")
• ("abx",[1.32,-5.12],-0.11)
Of course, tuples can become list members as well, or vice versa:
• [("ahmet","akhunlar",("deceased", 1991)), ("huri","huriyegil","not born")]
• (["ahmet","akhunlar",("deceased", 1991)], ["huri","huriyegil","not born"])
Programmers generally prefer using lists over tuples since they allow changing elements, which is often a
useful facility in many problems.
Lists (and tuples) are used whenever there is a need for an ordered set. Here are a few usecases for lists and
tuples:
• Vectors.
• Matrices.
• Graphs.
• Board game states.
• Student records, address book, any inventory.
Useful operations with lists and tuples
1- Deletion from lists
As far as deletion is concerned, you can use two methods:
• Assigning an empty list to the slice that is going to be removed:

68 Chapter 4. Dive into Python

>>> L = [111,222,333,444,555,666]
>>> L[1:5] = []
>>> print(L)
[111, 666]

• Using the del statement on the slice that is going to be removed:

>>> L = [111,222,333,444,555,666]
>>> del L[1:5]
>>> print(L)
[111, 666]

2- Insertion into lists

For insertion, you can use three methods:
• Using assignment with a degenerate use of slicing:

>>> L = [111,222,333,444,555,666]
>>> L[2:2] = [888,999]
>>> print(L)
[111, 222, 888, 999, 333, 444, 555, 666]

• The second method can insert only one element at a time, and requires object-oriented features, which
will be covered in Chapter 7.

>>> L = [111,222,333,444,555,666]
>>> L.insert(2, 999)
>>> print(L)
[111, 222, 999, 333, 444, 555, 666]

where the insert function takes two parameters: The first parameter is the index where the item will be
inserted and the second parameter is the item to be inserted.
• The third methods uses the append() method to insert an element only to the end or extend() to
append more than one element to the end:

>>> L = [111,222,333,444,555]
>>> L.append(666)
>>> print(L)
[111, 222, 333, 444, 555, 666]
>>> L.extend([777, 888])
[111, 222, 333, 444, 555, 666, 777, 888]

3- Data creation with ``tuple()`` and ``list()`` functions: Similar to other data types, tuple and list data types
provide two functions for creating data from other data types.
4- Concatenation and repetition with lists and tuples: Similar to strings, + and * can be used respectively to
concatenate two tuples/lists and to repeat a tuple/list many times.
5- Membership: Similar to strings, in and not in operations can be used to check whether a tuple/list
contains an element.
Here is an example that illustrates the last three items:

4.2. Container data (str, tuple, list, dict, set) 69

>>> a = ([3] + [5])*4
>>> a.append([3, 5])
>>> print(a)
[3, 5, 3, 5, 3, 5, 3, 5, [3, 5]]
>>> a.extend([3, 5])
>>> print(a)
[3, 5, 3, 5, 3, 5, 3, 5, [3, 5], 3, 5]
>>> b = tuple(a)
>>> print(b)
(3, 5, 3, 5, 3, 5, 3, 5, [3, 5], 3, 5)
>>> [3, 5] not in b
False
>>> [5, 3] not in b
True # test for single element, not subsequence

As you have recognized, the examples with some containers included two types of constructs which were
not covered yet: One is the use of ‘functions’ on containers, e.g. the use of len(). Due to your high school
background, this is certainly easy to understand. The second construct is something new. It appears that we
can use some functions suffixed by a dot to a container (e.g. the append and insert usages in the examples
above). This construct is an internal function call of data structure called object. Containers are actually
objects and in addition to their data containment property, they also have some defined action associations.
These actions are named as member functions, and called (applied) on that object (in our case -the container-)
by means of this dot notation:

•.f (□) has the conceptual meaning of f (•, □) (4.2.2)

(Don’t try this explicitly, it will not work. The equivalence is just ‘conceptual’: meaning the function receives
the object internally, as a ‘hidden’ argument). All these will be covered in details in Chapter 7. Till then, for
sake of completeness, from time to time we will referring to this notation. Till then simply interpret it based
on the equivalence depicted above.
Example: Matrices as Nested Lists
In many real-world problems, we often end up with a set of values that share certain semantics or function-
ality, and benefit from representing them in a regular grid structure that we call matrix:
 
a11 a12 ... a1n
 a21 a22 ... a2n 
 
A= . .. .. , (4.2.3)
 .. . . 
am1 am2 ... amn

which has n columns and m rows, and we generally shorten this as m × n and say that matrix A has size
m × n. The following set of equations describe relations among a set of variables and called system of
equations:

3x + 4y + z = 4,
−3x + 3y + 5 = 3, (4.2.4)
x+y+z = 0,

which can be represented with matrices as:

    
3 4 1 x 4
−3 3 5 y  = 3 , (4.2.5)
1 1 1 z 0

70 Chapter 4. Dive into Python

which defines the same problems in terms of matrices and matrix multiplication. It is okay if you are not
familiar with matrix multiplication – we will briefly explain it and use it as an example in the next chapter.
Writing a problem in a matrix form like we did above allows us to group the related aspects of the problem
together and focus on these groups for solving a problem. In our example, we grouped coefficients in a
matrix and this allows us to analyze and manipulate the coefficients matrix to check whether for this system
of equations there is a solution, whether the solution is unique or what the solution is. All these are questions
that are studied in Linear Algebra and beyond the scope of our book.
Now let us see how we can represent matrices in Python. A very straightforward approach that would also
allow changing elements of the matrix later on is to use lists in a nested form, as follows:

>>> A = [[3, 4, 1],

... [-3, 3, 5],
... [1, 1, 1]]
>>> A
[[3, 4, 1], [-3, 3, 5], [1, 1, 1]]

where each row is represented as a list and a member of the outer list. This would allow us to access entries
like this:

>>> A[1]
[-3, 3, 5]
>>> A[1][0]
-3

Although we can represent matrices like this, there are very advanced libraries that make representing and
working with matrices more practical, as we will see in Chapter 10. For an extended coverage on matrices
and matrix operations, we refer to Matrix algebra for beginners19 (by Jeremy Gunawardena) or Chapter 2.2
of “Mathematics for Machine Learning”20 (by Marc Peter Deisenroth, A. Aldo Faisal and Cheng Soon Ong).

4.2.5 Dictionary

Dictionary is a container data type where accessing items can be performed with indexes that are not numer-
ical; in fact, in dictionaries, indexes are called keys. A list, tuple, or string data type stores a certain element
at each numerical index (key). Similarly, a dictionary stores an element (value) for each key. In other words,
a dictionary is just a mapping from keys to values (Fig. 4.2.3).
The keys can be only immutable data types, i.e., numbers, strings, or tuples (with only immutable elements)
– some other immutable types that we do not cover in the book are also possible. Since lists and dictionaries
are mutable, they cannot be used for keys for indexing. As for values, there is no limitation on the data type.
A dictionary is a discrete mapping from a set of Python elements to another set of Python elements. Itself,
as well as its individual elements, are mutable (you can replace them). Moreover, it is possible to add and
remove items from the mapping.
19
http://vcp.med.harvard.edu/papers/matrices-1.pdf
20
https://mml-book.github.io/book/mml-book.pdf

4.2. Container data (str, tuple, list, dict, set) 71

Fig. 4.2.3: A dictionary provides a mapping from keys to values.

A dictionary is represented as key-value pairs, each separated by a column sign (:) and all enclosed in a pair
of curly braces. The dictionary in Fig. 4.2.3 would be denoted as:

{34:"istanbul",
58:"sivas",
"istanbul":[[7,"tepeli","sehir"],(41.0383,28.9703),34,15.5e6],
"murtaza abi":"sivas",
("ahmet",(8,14)):"almanya","james bond":7,
("ahmet",(1,6)):["anne","anneanne"],
("ahmet",7):"sivas",
"karsiyaka":35.5}

Similar to other containers, we usually store containers under a variable. Let us assume the dictionary above
was assigned to a variable with the name conno and look at some examples:

conno = {34:"istanbul", 58:"sivas", "istanbul":[[7,"tepeli","sehir"],(41.0383,28.9703),34,15.

,→5e6], "murtaza abi":"sivas", ("ahmet",(8,14)):"almanya","james bond":7, ("ahmet",(1,6)):[
,→"anne","anneanne"], ("ahmet",7):"sivas", "karsiyaka":35.5}
print(conno["murtaza abi"])
print(conno["istanbul"])

sivas
[[7, 'tepeli', 'sehir'], (41.0383, 28.9703), 34, 15500000.0]

Let us ask for something that does not exist in the dictionary:

>>> print(conno["ankara"])
KeyError: 'ankara'

Ups, that was bad. Though we have a decent method to test for the existence of a key in a dictionary, namely
the use of in:

72 Chapter 4. Dive into Python

print("ankara" in conno)

False

It is also possible to remove/insert mappings from/to a dictionary:

print("conno has this many keys:", len(conno))

conno["ankara"] = ["baskent", "anitkabir", 6]
print("conno has this many keys now:", len(conno))
print(conno["ankara"])
print(conno["murtaza abi"])
del conno["murtaza abi"]
print("murtaza abi" in conno)
print("After murtaza abi is deleted we have this many keys:", len(conno))

conno has this many keys: 9

conno has this many keys now: 10
['baskent', 'anitkabir', 6]
sivas
False
After murtaza abi is deleted we have this many keys: 9

The benefit of using a dictionary is ‘timewise’. The functionality of a dictionary could be attained by using
(key, value) tuples inserted into a list. Then, when you need the value of a certain key, you can search one-by-
one each (key, value)-tuple element of the list until you find your key in the first position of a tuple element.
However, this will consume a time proportional to the length of the list (worst case). On the contrary, in a
dictionary, this time is almost constant.
Moreover, a dictionary is more practical to use since it already provides accessing elements in a key-based
fashion.
Useful Operations with Dictionaries
Dictionaries support len() and membership (in and not in) operations that we have seen above. You can
also use <dictionary>.values() and <dictionary>.keys() to obtain lists of values and keys respec-
tively.

4.2.6 Set

Sets are created by enclosing elements into a pair of curly-braces and separating them with commas. Any
immutable data type, namely a number, a string or a tuple, can be an element of a set. Mutable data types
(lists, dictionaries) cannot be elements of a set. Being mutable, sets cannot be elements of a sets.
Here is a small example with sets:

a = {1,2,3,4}
b = {4,3,4,1,2,1,1,1}
print (a == b)
a.add(9)
a.remove(1)
print(a)

4.2. Container data (str, tuple, list, dict, set) 73

True
{2, 3, 4, 9}

The most functionalities of sets can be undertaken by lists. Furthermore, lists do not possess the restrictions
sets do. On the other hand, especially membership tests are much faster with sets, since member repetition
is avoided.
Frozenset
Python provides an immutable version of the set type, called frozenset. A frozenset can be constructed
using the frozenset() function as follows:

>>> s = frozenset({1, 2, 3})

>>> print(s)
frozenset({1, 2, 3})

Being immutable, frozensets can be a member of a set or a frozenset.

Useful Operations with Sets
Apart from the common container operations (len(), in and not in), sets and frozensets support the fol-
lowing operators:
• S1 <= S2: True if S1 is a subset of S2.
• S1 >= S2: True if S1 is a superset of S2.
• S1 | S2: Union of the sets (equivalent to S1.union(S2)).
• S1 & S2: Intersection of the sets (equivalent to S1.intersection(S2)).
• S1 - S2: Set difference (equivalent to S1.difference(S2)).
The following are only applicable with sets (and not with forezensets) as they require a mutable container:
• S.add(element): Add a new element to the set.
• S.remove(element): Remove element from the set.
• S.pop(): Remove an arbitrary element from the set.

4.3 Expressions

Expressions such as 3 + 4 describe calculation of an operation among data. When an expression is evaluated,
the operations in the expression are applied on the data specified in the expression and a resulting value is
provided.
Operations can be graphically illustrated as follows:

□1 ⊙ □2 (4.3.1)

where ⊙ is called the operator, and □1 and □2 are called operands. This was a binary operator; i.e. it acted
on two operands.
We can also have unary operators:

⊙□ (4.3.2)

74 Chapter 4. Dive into Python

or operators that have more than two operands (see chained comparison operators below).
Before we can cover how such operations are evaluated, let us look at commonly used operations (operators)
in Python.

4.3.1 Arithmetic, Logic, Container and Comparison Operations

Python provides the operators in Table 4.3.1 for arithmetic (addition, subtraction, multiplication, division, ex-
ponentiation), logic (and, or, not), container (indexing, membership) and comparison (less, less-than, equal-
ity, not-equality, greater, greater-than) operations.
Note that operators such as + and * have different meanings on different data types. For numerical data, they
mean addition and multiplication whereas for container data, they mean concatenation and repetition.

Table 4.3.1: Arithmetic, Logic, Container and Comparison opera-

tors in Python.
Operator Operation Result Type
[] Indexing Any data type
** Exponentiation Numeric
* Multiplication or Repetition Numeric or container
/ Division Numeric (floating point)
// Integer Division Numeric (integer)
+ Addition or concatenation Numeric or container
- Subtraction Numeric
< Less than Boolean
<= Less than or equal to Boolean
> Greater than Boolean
>= Greater than or equal to Boolean
== is equal to Boolean
!= is not equal to Boolean
in is a member Boolean
not in is not a member Boolean
not logical negation Boolean
and logical and Boolean
or logical or Boolean

Below are some illustrations:

S1 = "Four"
S2 = "Five"
B1 = len(S1) < len(S2)
print("B1 is: ", B1)
B2 = S1 != S2
print("B2 is: ", B2)
B3 = B1 or B2
print("B3 is: ", B3)

B1 is: False
B2 is: True
B3 is: True

4.3. Expressions 75
4.3.2 Exercise

Give one example for each row in Table 4.3.1. Please complete this exercise in the Colab version of the
chapter.

4.3.3 Evaluating Expressions

In the previous section, we have seen simple use of operators in an expression. In many cases, we combine
several operators for brevity and readability, e.g. 2.3 + 3.4 * 4.5. This expression can be evaluated in
two different ways:
• (2.3 + 3.4) * 4.5, which would yield 25.65.
• 2.3 + (3.4 * 4.5), which would yield 17.599999999999998 in Python.
Since the results are very different, it is very important for a programmer to know in which order operators
are evaluated when they are combined. There are two rules that govern this:
1. Precedence: Each operator has an associated precedence (priority) based on which we can determine
which operator is going to be evaluated first. E.g. multiplication has higher precedence than addition,
and therefore, 2.3 + 3.4 * 4.5 would be evaluated as 2.3 + (3.4 * 4.5) in Python.
2. Associativity: If two operators have the same precedence, evaluation order is determined based on
associativity. Associativity can be from left to right or from right to left.
For the operators, the complete associativity and precedence information are listed in Table 4.3.2.

Table 4.3.2: Precedence and associativity for the operators in Table

4.3.1.
Operator Precedence Associativity
[] Left-to-right
1.

** Right-to-left
2.

*, /, //, % Left-to-right
3.

+, - Left-to-right
4.

<, <=, >, >=, ==, !=, in, not in Special

not Unary
6.

and Left-to-right (with short-cut)

or Left-to-right (with short-cut)

76 Chapter 4. Dive into Python

Therefore, according to Table 4.3.2, a sophisticated expression such as 2**3**4*5-2//2-
1 is equivalent to 2**81*5-1-1 which is equivalent to 12089258196146291747061760-2 which
is 12089258196146291747061758.
Below are some notes and explanations regarding expression evaluation:
(i) The Special keyword in Table 4.3.2 means some treatment which is common to mathemat-
ics but not programming. In that sense Python is unique among commonly used programming lan-
guages. If ⊙i is any boolean comparison operator and □j is any numerical expression, the sequence of
□1 ⊙1 □2 ⊙2 □3 ⊙3 □4 · · · □n−1 ⊙n−1 □n (4.3.3)
is interpreted as:

□1 ⊙1 □2 and □2 ⊙2 □3 and □3 ⊙3 □4 ··· □n−1 ⊙n−1 □n (4.3.4)

(ii) It is always possible to override the precedence by making use of parenthesis. We are familiar
with this since our primary school days.
(iii) If two numeric operands are of the same type, then the result is of that type unless the operator is /
(for which the result is always a floating point number). Also comparison operators return bool typed
values.
(iv) If two numeric operands that enter an operation are of different types, then a computation occurs ac-
cording to the following rules:
• if one operand is integer and the other is floating point: The integer is converted to floating point.
• if one operand is complex: The complex arithmetic is carried out according to the rules of math-
ematics among the real/imaginary part coefficients (which are either integers or floating points).
Each of the two resulting coefficients are separately checked for having zero (.0) fractional part.
If so, that one is converted to integer.
(v) Except for the two logical operators and and or, all operators have an evalua-
tion scheme which is coined as eager evaluation. Eager evaluation is the strat-
egy where all operands are evaluated first and then the semantics of the opera-
tors kicks in, providing the result. Here is an example: Consider mathematical expression of:
0*(2**150-3**95) As a human being, our immediate reaction would be: Anything mul-
tiplied with 0 (zero) is 0, therefore we do not have to compute the two huge exponentia-
tions (the second operand). This is taking a short-cut in evaluation and is certainly far from ea-
ger evaluation. Eager evaluation would evaluate the internals of the parenthesis, obtain -
693647454339354238433323618063349607247325483 and then multiply this with 0 to obtain 0.
Yes, Python would go this ‘less intelligent’ way and do eager evaluation. The logical op-
erators and and or, though, do not adopt eager evaluation. On the contrary, they use short-
cuts (this is known as based-on-need evaluation in computer science).
In a conjunctive expression like:

□1 and □2 and □3 and · · · and □n (4.3.5)

The evaluation proceeds as follows:

4.3. Expressions 77
Fig. 4.3.1: Logical AND evaluation scheme

Similarly, a disjunctive expression:

□1 or □2 or □3 or · · · or □n (4.3.6)

has the following evaluation scheme:

Fig. 4.3.2: Logical OR evaluation scheme

Although high-level languages provide mechanisms for evaluating expressions involving multiple operators,
it is not a good programming practice to leave multiple operators without parentheses. A programmer should
do his/her best to write code that is readable and understandable by other programmers and that does not
include any ambiguity whatsoever. This includes expressions.

4.3.4 Implicit and Explicit Type Conversion

In Python, when you apply a binary operator on items of two different data types, it tries to convert one data
to another one if possible. This is called implicit type conversion. For example,

>>> 3+4.5
7.5
>>> 3 + True
4
>>> 3 + False
3

78 Chapter 4. Dive into Python

which illustrates that an integer is converted to a float, True is converted to integer 1, and False is converted
to integer zero.
Although Python can do such conversions, it is a good programming practice to make these conversions
explicit and make the intention clear to the reader. Explicit type conversion, also called as type casting, can
be performed using the keyword for the target type as a function. For example:

>>>> 1.1*(7.1+int(2.5*5))
21.01

Not all conversions are possible, of course. Implicit conversions are allowed only for the basic data types as
illustrated below:

Fig. 4.3.3: Type casting among numeric and Boolean

Type casting can accommodate conversion between a wider spectrum of data types, including containers:

>>> str(34)
'34'
>>> list('34')
['3', '4']

4.4 Basic Statements

Now, let us continue with actions that do not provide (return) us data as a result.

4.4.1 Assignment Statement and Variables

When we perform computation in Python, we either print the result and/or keep the result for further com-
putations that we are going to perform. Variables help us here as named memory positions in which we can
store data. You can imagine them as pigeonholes that are able to hold a single data item.
There are two methods to create and store some data into a variable. Here we will mention the overwhelm-
ingly used one, the second is more implicit and will be introduced when we deal with functions.
A variable receives data for storage by the use of the assignment statement. It has the following form:
V ariable = Expression
Although the equal sign resembles an operator that is placed between two operands, it is not an operator:
Operators return a value, and in Python, the assignment is not an operator and it does not return a value (this
can be different in other programming languages).
The action semantics of the assignment statement is simple:

4.4. Basic Statements 79

1. The Expression is evaluated.
2. If the V ariable does not exist, it is created.
3. The evaluation result is stored into the V ariable (by doing so, any former value, if existing, is purged).
After assignment, the value can be repetitively used. To use the value in a computation, we can use the name
of the variable. Here is an example:

>>> a = 3
>>> b = a + 1
>>> (a-b+b**2)/2
7.5

It is quite common to use a variable on both sides of the assignment statement. For example:

>>> a = 3
>>> b = a + 1
>>> a = a + 1
>>> (a-b+b**2)/2
8.0

The expression on the second line uses the 3 value for a. In the next line, namely the third line, again 3 is
used for a in the expression (a+1). Then, the result of the evaluation (3+1) which is a 4 is stored in variable
a. The variable a had a previous value of 3; that value is purged and replaced by 4. The former value is not
kept anywhere and cannot be recovered.
Multiple assignments
It is possible to have multiple variables assigned to the same value, e.g.:

>>> a = b = 4

assigns an integer value of 4 to both a and b. After the multi-assignment above, if you change b to some
other value, the value of a will still remain to be 4.
Multiple assignment with different values Python provides a powerful mechanism for providing different
values to different variables in assignment:

>>> a, b = 3, 4
>>> a
3
>>> b
4

This is internally handled by Python with tuples and equivalent to:

>>> (a,b) = (3, 4)

>>> a
3
>>> b
4

This is called tuple matching and would also work with lists (i.e. [a, b] = [3, 4]).
Swapping values of variables Tuple matching has a very practical benefit: Let us say you want to swap the
values in variables. Normally, this requires the use of a temporary variable:

80 Chapter 4. Dive into Python

>>> print(a,b)
3 4
>>> temp = a
>>> a = b
>>> b = temp
>>> print(a,b)
4 3

With tuple matching, we can do this in one line:

>>> print(a,b)
3 4
>>> a,b = b,a
>>> print(a,b)
4 3

Frequently-asked questions about assignments

• QUESTION: Considering the example below, one may have doubts about the value in b: Is it updated?
On the fourth line, there is an expression using b: Which value is it refering to? 4 or 5? When we use
b in a following expression, will the ‘definition’ be retrived and recalculated?

>>> a = 3
>>> b = a + 1
>>> a = a + 1
>>> (a-b+b**2)/2
8.0

• ANSWER: No. Statements are executed only once: the moment they are entered (it is possible to
repetitively execution a statement but here it is not used). Each assignment in the example above is
executed only once. No reevaluation is performed, the use of a variable in an expression, the ``a`` and
those ``b``’s, refer solely to the last calculated and stored values: In the evaluation of ``(a-b+b**2)/2``,
``a`` is 4 and ``b`` is 4.
• QUESTION: We had variables in math, especially in middle school and high school. So, this is very
similar to that right? But I am confused having seen a line like a = a + 1. What’s happening? The
as cancel out and we are left with 0 = 1?
• ANSWER: The use of the equality sign (=) is confusing to some extent. It does not stand for a balance
among the left-hand-side and the right-hand-side. Do not interpret it as an ‘equality of mathematics’.
It has a absolutely different semantics. As said, it only means:
1. First calculate the right-hand side,
2. then store the result into an (electronic) pigeon hole which has the name label given to the left-
hand-side of the equal sign.
• QUESTION: I typed in the following lines:

>>> x = 5
>>> y = 3
>>> x = y
>>> y = x
>>> print x,y
3 3

4.4. Basic Statements 81

However, it should have been 3 5, right? Or am I doing something wrong?
• ANSWER: You are somehow missing the time flow. Statements are executed in order:
1. First: ``x`` is set to 5.
2. Second: ``y`` is set to 3.
3. Third: ``x`` is set to the value stored in ``y`` which is 3. ``x`` now holds 3.
4. Fourth: ``y`` is set to the value stored in ``x`` which is 3 (just look to the 3. item which right
above this one). ``y`` now holds 3.
5. Fifth: print both the values ``x`` and ``y``. Both are ``3`` and they got printed.

4.4.2 Variables & Aliasing

There is something peculiar about lists that we need to be careful while assigning them to variables. First
consider the following code:
“The first example”

print("address of 5: ", id(5))

a = 5
print("address of a: ", id(a))
b = a
print("address of b: ", id(b))
a = 3
print("address of a: ", id(a))
print("b is: ", b)

address of 5: 4311538352
address of a: 4311538352
address of b: 4311538352
address of a: 4311538288
b is: 5

Here, we used the id() function to display the address of variables in the memory to help us understand
what happens in those assignments: * The second line creates 5 in memory and links that with a. * The
fourth line links the content of a with variable b. So, they are two different names for the same content. *
The sixth line creates a new content, 3, and assigns it to a. Now, a points to a different memory location than
b. * b still points to 5, which is printed.
Now keep the task the same but change the data from integer to a list:
“The second example”

a = [5,1,7]
b = a
print("addresses of a and b: ", id(a), id(b))
print("b is: ", b)
a = [3,-1]
print("addresses of a and b: ", id(a), id(b))
print("b is: ", b)

82 Chapter 4. Dive into Python

addresses of a and b: 4344021184 4344021184
b is: [5, 1, 7]
addresses of a and b: 4345519360 4344021184
b is: [5, 1, 7]

In other words, this works similar to the first example, as expected. But now, consider the following slightly
different example:
“The third example”

a = [5,1,7]
b = a
print("addresses of a and b: ", id(a), id(b))
print("b is: ", b)
a[0] = 3
print("addresses of a and b: ", id(a), id(b))
print("b is: ", b)

addresses of a and b: 4345520000 4345520000

b is: [5, 1, 7]
addresses of a and b: 4345520000 4345520000
b is: [3, 1, 7]

To our surprise, the first element’s replacement of a got also reflected in b. This should not be surprising
since both a and b point to the same memory location and since a list is mutable, when we change an element
in that memory location, it affects both a and b that are just names for that memory location.
Although the examples above used lists for illustration purposes, aliasing pertains to all mutable data types.
Aliasing is a powerful concept that can be hazardous and beneficial depending on the content:
• If you carelessly assign mutable variable to another variable, changes on one variable is going to reflect
the other one. If this was not intended and the location in the code where the aliasing was initiated
could not be identified, you may lose hours or days trying to identify what the problem is with your
code.
• Aliasing can be beneficial especially when we want our changes on one variable to be reflected on
another. This will be useful for passing data to functions and getting the results from functions.

4.4.3 Naming variables

Programmers usually select names for variables so that it indicates what the content will be. Variable names
may be arbitrarily long. They may contain letters (from the English alphabet) as well as numbers and under-
scores, but they must start with a letter or an underscore. While using upper case letters is allowed, bear in
mind that programmers reserve starting with an upper case to differentiate a property (scope) of the variable
which you will learn later (when we introduce functions).
Here are a few examples for variable names (all are different):

y y1 Y Y_1 _1
te mperature temperat ure_today Tempera tureToday C umulative coo rdinate_x
a1b145c a 1_b1_45_c s_s_ _ ___

4.4. Basic Statements 83

As you might have recognized, though they are perfectly valid, the five examples in the last line do not make
much sense. Here is a short list that you should prefer to follow in variable naming:
• Name variables in the context of the value they are going to store. For example

>>> a = b * c

is syntactically correct, but its purpose is not evident. Contrast this with:

>>> salary = hours_worked * hourly_pay_rate

• Use different variables for different data, a.k.a. the ‘Single Responsibility Principle’. For example,
even if you could use the same variable in a statement for counting and in another statement, for
holding the largest grade, this is not recommended: Do not economise with variables. Devise two
different variables where each of which will reflect the semantics and the context, uniquely.
• Variable names should be pronounceable, which makes them easier to remember.
• Do not use variable names which could misguide you or whoever look into your code.
• Use i,j,k,m,n for counting only: Programmers implicitly recognize them as integer holding (the
historical reason for this dates to 60 years ago).
• Use x,y,z for coordinate values or multi-variate function arguments.
• Do not use single character variable l as it can be easily confused with 1 (one).
• If you are going to use multiple words for a variable, choose one of these:
– Put all words into lowercase, affix them using _ as separator (e.g. highest_midterm_grade,
shortest_path_distance_up_to_now).
– Put all words but the first into first-capitalized-case, put the first one into lowercase, affix them
without using any separator (e.g.highestMidtermGrade, shortestPathDistanceUpToNow).
Reserved names
The following keywords are being used by Python already and therefore, you cannot use them to name your
variables.

and def exec if not return

assert del finally import or try
break elif for in pass while
class else from is print yield
continue except global lambda raise

4.4.4 Other Basic Statements

Python has other basic statements listed below:

pass, del, return, yield, raise, break, continue, import, future, global, nonlocal.
Among these, we have seen del and we will see some others in the rest of the book.
print used to be a statement in version 2, which however changed in version 3: print is a function in
version 3 and therefore, it has a value (which we will see later).

84 Chapter 4. Dive into Python

4.5 Compound Statements

Like other high-level languages, Python provides statements combining many statements as their parts. Two
examples are:
• Conditional statements where different statements are executed based on the truth value of a condition,
e.g.:

if <boolean-expression>:
statement-true-1
statement-true-2
...
else:
statement-false-1
statement-false-2
...

• Repetitive statements where some statements are executed more than once depending on a condition
or for a fixed number of times, e.g.:

while <boolean-condition>:
statement-true-1
statement-true-2
...

which executes the statements while the condition is true.

We will see these and other forms of compound statements in the rest of the book.

4.6 Basic actions for interacting with the environment

In our programs, we frequently require obtaining some input from the user or displaying some data to the
user. We can use the following for these purposes.

4.6.1 Actions for input

In Python, you can use the input() function for obtaining input from the user, e.g.:

>>> s = input("Now enter your text: ")

Now enter your text: This is the text I entered
>>> print(s)
This is the text I entered

If you expect the user to enter an expression, you can evaluate it using the eval() function as we explained
in Section 4.2.3.

4.5. Compound Statements 85

4.6.2 Actions for output

For displaying data to the screen, Python provides the print() function:

print(item1, item2, ..., itemN)

For example:

>>> print('Python', 'is', 'so', 'fun')

Python is so fun

In many cases, we end up with strings that have placeholders for data items which we can fill in using a
formatting function as follows:

>>> print("I am {0} tall, {1} years old and have {2} eyes".format(1.86, 20, "brown"))
I am 1.86 tall, 20 years old and have brown eyes

Alternatively, instead of using integers for the placeholders, we can give names to the placeholders:

>>> print("I am {height} tall, {age} years old and have {eyes} eyes. Did I tell you that I␣
,→was {age}?".format(age=20, eyes="brown", height=1.86))
I am 1.86 tall, 20 years old and have brown eyes. Did I tell you that I was 20?

The format() function provides much more functionalities than the ones we have illustrated. However, this
extent should be sufficient for general uses and this book. The reader interested in a complete coverage is
referred to the Python’s documentation on string formatting21 .
We can also ask Python to directly take in the values of the variables for which we provide the names:

>>> age = 20
>>> height = 1.70
>>> eye_color = "brown"
>>> print(f"I am {height} tall, {age} years old and have {eye_color} eyes")
I am 1.7 tall, 20 years old and have brown eyes

4.7 Actions that are ignored

In Python, we have two actions that are ignored by the interpreter:

4.7.1 Comments

Like other high-level languages, Python provides programmers mechanisms for writing comments on their
programs:
• Comments with #: When Python encounters # in your code, it ignores the rest of the line, assumes
the current line is finished, evaluates & runs the current line and continues interpretation with the next
line. For example:
21
https://docs.python.org/3/library/string.html

86 Chapter 4. Dive into Python

>>> 3 + 4 # We are adding two numbers here
7

• Multi-line comments with triple-quotes: If you wish to provide comments longer than one line, you
can use triple quotes:

"""
This is a multi-line comment.
We are flexible with the number of lines &
characters,
spacing. Python
will ignore them.
"""

Triple-quote comments are generally used by programmers to write documentation-level explanations and
descriptions for their codes. There are document-generation tools that process triple-quotes for automatically
generating documents for codes.
As we have seen before, the triple-quote comments are actually strings in Python. Therefore, if you use
triple-quotes for providing a comment, it should not overlap with an expression or statement line in your
code.

4.7.2 Pass statements

Python provides the pass statement that is ignored by the interpreter. The pass statement is generally used in
incomplete function implementations or compound statements to place a dummy holder (some instruction)
such that the interpreter does not complain about a statement being missing. For example:

if <condition>:
pass # @TODO fill this part
else:
statement-1
statement-2
...

4.8 Actions and data packaged in libraries

Many high-level programming languages like Python provide a wide spectrum of actions and data predefined
and organized in ‘packages’ that we call libraries. For example, there is a library for mathematical functions
and constant definitions which you can access using from math import * as follows:

>>> pi
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
NameError: name 'pi' is not defined
>>> sin(pi)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
NameError: name 'sin' is not defined
>>> from math import *
(continues on next page)

4.8. Actions and data packaged in libraries 87

(continued from previous page)
>>> pi
3.141592653589793
>>> sin(pi)
1.2246467991473532e-16

Below is a short list of libraries that might be useful:

Library Description
math Mathematical functions and definitions
cmath Mathematical functions and definitions for complex numbers
fractions Rational numbers and arithmetic
random Random number generation
statistics Statistical functions
os Operating system functionalities
time Time access and conversion functionalities

Of course, the list is too wide and it is not practical to list and explain all libraries here. The interested reader
is direct to to check the comprehensive list at Python docs22 .
The import statement that we used above loads the library and makes its contents directly accessible to us
by directly using their names (e.g. sin(pi)). Alternatively, we can do the following:

>>> import math

>>> math.sin(math.pi)
1.2246467991473532e-16

which requires us to specify the name math everytime we need to access something from it. We could also
change the name:

>>> import math as m

>>> m.sin(m.pi)
1.2246467991473532e-16

However, we discourage this way of using libraries until Chapter 7 where we introduce the concept of objects
and object-oriented programming.
However, to be able to learn what is available in a library, you can use this form (with dir() function):

>>> import math

>>> dir(math)
['__doc__', '__file__', '__loader__', '__name__', '__package__', '__spec__', 'acos', 'acosh
,→', 'asin', 'asinh', 'atan', 'atan2', 'atanh', 'ceil', 'comb', 'copysign', 'cos', 'cosh',
,→'degrees', 'dist', 'e', 'erf', 'erfc', 'exp', 'expm1', 'fabs', 'factorial', 'floor', 'fmod
,→', 'frexp', 'fsum', 'gamma', 'gcd', 'hypot', 'inf', 'isclose', 'isfinite', 'isinf', 'isnan
,→', 'isqrt', 'ldexp', 'lgamma', 'log', 'log10', 'log1p', 'log2', 'modf', 'nan', 'perm',
,→'pi', 'pow', 'prod', 'radians', 'remainder', 'sin', 'sinh', 'sqrt', 'tan', 'tanh', 'tau',

,→'trunc']

22
https://docs.python.org/3/library/

88 Chapter 4. Dive into Python

4.9 Providing your actions to the interpreter

Python provides different mechanisms for you to provide your actions and get them executed.

4.9.1 Directly interacting with the interpreter

As we have seen up to now, we can interact with the interpreter directly by typing our actions to the interpreter.
When you are done with the interpreter, you can quit using quit() or exit() functions or by pressing CTRL-
D. E.g.:

$ python3
Python 3.8.5 (default, Jul 21 2020, 10:48:26)
[Clang 11.0.3 (clang-1103.0.32.62)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> print("Python is fun")
Python is fun
>>> print("Now I am done")
Now I am done
>>> quit()
$

where we see that after quitting the interpreter, we are back at the terminal shell.
Although this way of coding is simple and very interactive, when your code gets longer and more compli-
cated, it becomes difficult to manage your code. Moreover, when you exit the interpreter, all your actions
are lost and to be able to run them again, you need to type them again, from scratch. This can be tedious,
redundant, impractical and inefficient.

4.9.2 Writing actions in a file (script)

Alternatively, we can place our actions in a file with an .py extension in the filename as follows:

print("This is a Python program that reads two numbers from the user, adds the numbers and␣
,→prints the result\n\n")
[a, b] = input("Enter two numbers: ")
print("You have provided: ", a, b)
result = a + b
print("The sum is: ", result)

Assuming that you save these lines into a file named test.py, you can execute those statements from a
terminal window as follows:

$ cat test.py
print("This is a Python program that reads two numbers from the user, adds the numbers and␣
,→prints the result\n\n")
[a, b] = input("Enter two numbers: ")
print("You have provided: ", a, b)
result = a + b
print("The sum is: ", result)
$ python3 test.py
This is a Python program that reads two numbers from the user, adds the numbers and prints␣
,→the result

(continues on next page)

4.9. Providing your actions to the interpreter 89

(continued from previous page)

Enter a list of two numbers [a, b]: [3, 4]

You have provided: 3 4
The sum is: 7

It is possible to provide command-line arguments to your script and use the provided values in your script
(named test.py):

from sys import argv

print("The arguments of this script are:\n", argv)

exec(argv[1]) # Get a
exec(argv[2]) # Get b

print("The sum of a and b is: ", a+b)

which can be run as follows:

$ python3 test.py a=10 b=20

The arguments of this script are:
['test.py', 'a=10', 'b=20']
The sum of a and b is: 30

Note that this example used the function exec(), which executes the statement provided to the function as
a string argument. Compare this with the eval() function that we have introduced before: eval() takes an
expression whereas exec() takes a statement.

4.9.3 Writing your actions as libraries (modules)

Another mechanism for executing your actions is to place them into libraries (modules) and provide them to
Python using import statement, as we have illustrated in Section 4.7. For example, if you have a test.py
file with the following content:

a = 10
b = 8
sum = a + b
print("a + b with a =", a, " and b =", b, " is: ", sum)

In another Python script or in the interpreter, you can directly type:

>>> from test import *

a + b with a = 10 and b = 8 is: 18
>>> a
10
>>> b
8

In other words, what you have defined in test.py becomes accessible after being imported.

90 Chapter 4. Dive into Python

4.10 Important Concepts

We would like our readers to have grasped the following crucial concepts and keywords from this chapter
(all related to Python):
• Basic data types.
• Basic operations and expression evaluation.
• Precedence and associativity.
• Variables and how to name them.
• Aliasing problem.
• Container types.
• Accessing elements of a container type (indexing, negative indexing, slicing).
• Basic I/O.
• Commenting your codes.
• Using libraries.

4.11 Further Reading

• A detailed history of Python and its versions: https://en.wikipedia.org/wiki/History_of_Python

• The concept of aliasing: http://en.wikipedia.org/wiki/Aliasing_%28computing%29

4.12 Exercises

• Without using Python, determine the results of the following expressions and validate your answers
with Python:
• 2 - 3 ** 4 / 8 + 2 * 4 ** 5 * 1 ** 8
• 4 + 2 - 10 / 2 * 4 ** 2
• 3 / 3 ** 3 * 3
• Assuming that a is True, b is True and c is False, what would be the values of the following expres-
sions?
(a) not a == b + d < not a
(b) a == b <= c == True
(c) True <= False == b + c
(d) c / a / b
p
• The Euclidean distance between two points (a, b) and (c, d) is defined as: (a−c)2 + (b−d)2 . Write
a Python code that reads a, b, c, d from the user, calculates the Euclidean distance and prints the result.

4.10. Important Concepts 91

92 Chapter 4. Dive into Python
5 | Conditional and Repetitive Execution

5.1 Conditional execution

Asking (mostly) mathematical questions that have Boolean type answers and taking some actions (or not)
based on the Boolean answers are crucial tools in designing algorithms. This is so vital that there is no single
programming language that does not provide conditional execution.

5.1.1 if statement

The syntax of the conditional (if) statement is

if Boolean expression : Statement

The semantics is obvious. If the Boolean expression evaluates to True, then the Statement is carried out;
otherwise Statement is not executed.

>>> if 5 > 2: print("Hurray")

Hurray
>>> if 2 > 5: print("No I will not believe that !")

We observe that the Boolean expression in the last example evaluates to False and hence the statement with
the print function is not carried out.
How to group statements?
It is quite often that we want to do more then one action instead of executing a single statement. As a general
syntax rule of Python:
• switch to a new line,
• indent by one tab,
• write your first statement to be executed,

93
• switch to the next line,
• indent the same amount (usually your Python-aware editor will do it for you)
• write your second statement to be executed,
• and so on…
• To finish this writing task, and presumably enter a statement which is not part of the group action, just
do not intend.
It is very very important that the statements that are part of the if part have the same amount and type of
whitespace. E.g. if the first statement has 4 spaces, all following statements in the if statement need to
start with 4 spaces. If it is a tab character, all following statements in the if statement need to start with a
tab character. Since these whitespaces are not visible, it is very easy to make mistakes and get annoying
errors from the interpreter. To avoid this, we strongly recommend you to choose a style of indentation (e.g. 4
whitespaces or a single tab) and stick to it as a programming style.
Here is an example:

>>> if 5 > 2:
... print(7*6)
... print("Hurray")
... x = 6
42
Hurray
>>> print(x)
6

The >>> as well as ... prompts appear when you sit in front of the interpreter. When you type your programs
into a file and then run them, of course they will not show up. If you type in these small programs (we call
them snippets) just ignore the >>> and ... from the examples, as follows:

if 5 > 2:
print(7*6)
print("Hurray")
x = 6
print(x)

42
Hurray
6

As you continue practicing programming, you will soon come across a need to be able to define actions that
should be carried out in case a Boolean expression turns out to be False. If it were not provided, we would
have to ask the same question again, but this time negated:

>>> if x == 6: print("still it is six")

>>> if not (x == 6): print("it is no more six")

One of them will certainly fire (be true). Although this is a feasible solution, there is a more convenient way:
We can combine the two parts into one if-else statement with the following syntax:

if Boolean expression : Statement

94 Chapter 5. Conditional and Repetitive Execution

else : Statement

Making use of the else part, the last example would be written as:

>>> if x == 6: print("still it is six")

... else: print("it is no more six")

Of course, using multiple statements in the if part or the else part is possible by intending them to the same
level.

5.1.2 Exercise

In the following code cell, you are expected to insert a couple of lines so that the output is always the absolute
value of N (try the program for various values of N by altering the right-hand side of the assignment):

N = -100 # Feel free to change this to any numerical value #

# @TODO type in your code that changes the content of N to its absolute value, below this␣
,→line

# @TODO print the result

5.1.3 Nested if statements

It is quite possible to nest if-else statements: i.e. combine multiple if-else statements within each other.
There is no practical limit to the depth of nesting. This enables us to code a structure, called a decision tree.
A decision tree is a set of binary questions & answers where at every case of an answer either a new question
is asked or an action is carried out. Fig. 5.1.1 displays such a decision tree for forecasting rain based on the
temperature, wind and air pressure values.

Fig. 5.1.1: An example decision tree.

5.1. Conditional execution 95

The tree in Fig. 5.1.1 can be coded as nested if-else statements:

if temperature > 21:

if wind > 3.2:
if pressure > 66: rain = True
else: rain = False
else: rain = False
else:
if wind > 7.2: rain = True
else:
if pressure > 79: rain = True
else: rain = False

Sometimes the else case contains another if which has an else of its own. And this ladder goes on. For
such a case, elif serves as a combined keyword for the else if action, as follows:

if Boolean expression1 : Statement1

elif Boolean expression2 : Statement2

..
.
elif Boolean expressionn : Statementn

else : Statementotherwise

The flowchart in Fig. 5.1.2 explains the semantics of the if-elif-else combination.

96 Chapter 5. Conditional and Repetitive Execution

Fig. 5.1.2: The semantics of the if-elif-else combination.

There is no restriction on how many elif statements are used. Furthermore, the presence of the last statement
(the else) is optional.

5.1.4 Practice

Let us assume that you have a value assigned to a variable x. Based on this value, a variable s is going to be
set as:


 (x + 1)2 , x < 1

x√− 0.5, 1 ≤ x < 10
s= (5.1.1)

 x + 0.5, 10 ≤ x < 100

0, otherwise

It is convenient to make use of an if-elif-else structure:

5.1. Conditional execution 97

#@TODO Assign a value to x
x = 10

if x < 1: s = (x+1)**2
elif x < 10: s = x-0.5
elif x < 100: s = (x+0.5)**0.5
else: s = 0

print("s is: ", s)

s is: 3.24037034920393

5.1.5 Conditional expression

The if statement does not return a value. As said, that is so for all statements. They do not have a value of
their own. A non-statement alternative that has a return value is the conditional expression.
It also uses the if and else keywords. This time if is not at the start of a statement, but following a value.
Here is the syntax:

expressionY ES if Boolean expression else expressionN O

This whole structure is an expression. It yields a value. So, it can be used in any place an expression is
allowed to be used. It is evaluated as follows:
• First the Boolean expression is evaluated.
• If the outcome is True then expressionY ES is evaluated and used as value (expressionN O is left
untouched).
• If the outcome is False then expressionN O is evaluated and used as value (expressionY ES is left
untouched).
Though it is not compulsory, it is wise to use conditional expression enclosed in parenthesis to prevent wrong
sub-expression groupings.
Let us look at an example:

>>> x = -34.1905
>>> y = (x if x > 0 else -x)**0.5
>>> print(y)
5.84726431761

p
As you have observed, y is assigned the value |x|. For code readability, it is strongly advisable not to nest
conditional expressions. Instead, restructure your coding into nested if-else statements (by making use of
intermediate variables).

98 Chapter 5. Conditional and Repetitive Execution

5.2 Repetitive execution

Repeating a sequence of instructions over-and-over is a programming ingredient which is used frequently.

This action is called iteration or loop.
We use iteration
to systematically deal with all possible cases or all elements of a collection, or
to jump from one case to another as a function of the previous case.
The repetition of a sequence of instructions is carried out either for a known number of times or until a
criterion is met.
Iteration examples:
• Computing letter grades of a class (Type I, for all students).
• Finding the shortest path from city A to city B in a map (Type II).
• Root finding by Newton-Raphson method (Type II).
• Finding darkest and brightest point(s) in a image (Type I, for all pixels).
• Computing a function value using Taylor’s expansion (Type I).
• Computing the next move in a chess game (Type II).
• Obtaining the letter frequency of a text (Type I, for all letters).
Python provides two statements for iteration. Namely, while and for. for is used mostly for (Type I)
iterations whereas while is used for both types.

5.2.1 while statement

The syntax of while resembles the synax of the if statement:

while Boolean expression : Statement

It is certainly possible to have a statement group subject to the while, as it was with the if statement. The
semantics can be expressed as the flowchart in Fig. 5.2.1:

Fig. 5.2.1: The flowchart illustrating how a while statement is executed.

5.2. Repetitive execution 99

Later in this chapter, when we have introduced the for statement as well, we will consider some special
statements (break and continue) that are allowed only under the scope of a while or for and are capable
of altering the execution flow (to some limited extent).
Since a while statement is a statement, it can be part of another while statement (or any other compound
statement):

while <condition-1>:
statement-1
statement-2
...
while <condition-2>:
statement-inner-1
statement-inner-2
...
statement-inner-M
... # statements after the second while
statement-N

Of course, there is no practical limit on the nesting level.

Let us look at several examples below to illustrate these concepts.

5.2.2 Examples with while statement

Example 1: Finding Average of Numbers in a List

Let’s say we have a list of numbers and we are asked to find their average. To make things easier to follow,
let us start with the mathematical definition:

1 X
N
avg(L) = Li , (5.2.1)
N
i=1

where Li is the ith number in the list which contains N numbers.

Let us see the algorithm before implementing a solution in Python (note that, in the equation, indexing started
with 1 – compare this with the following):

Input: A list L that includes N numbers

Output: avg, which holds the average of numbers in L

Step 1: Initialize a sum variable with value 0

Step 2: Initialize an index variable, i, with value 0
Step 3: While i is less than N, Execute Steps 4-5
Step 4: sum = sum + L[i]
Step 5: i = i + 1
Step 6: avg = sum/N

Before we proceed with the Python implementation, make sure that you have understood the algorithm. The
best way to make sure that is to take a pen & paper and go through each step while keeping a track of the
variables, as if you are the computer.
Here is the implementation in Python:

100 Chapter 5. Conditional and Repetitive Execution

# L: the list of numbers
# @TODO: Change the list and check if the code works
L = [10, -4, 4873, -18]
N = len(L)

sum = 0 # Step 1
i = 0 # Step 2

while i < N: # Step 3

sum = sum + L[i] # Step 4
i = i + 1 # Step 5

avg = sum / N # Step 6

print(avg)

1215.25

Have you noticed how close the Python code is to the algorithm? Python’s principles for keeping things
simple makes it so much easier to write that it can be very close to the pseudo-code.
Example 2: Standard Deviation
Now, let us look at a highly related problem, that of calculating the standard deviation. Standard deviation
can be formally defined as follows:
v
u
u1 X N
std(L) = t (Li − avg(L))2 . (5.2.2)
N
i=1

The extension of the previous example is rather straightforward therefore but we will write it down nonethe-
less since these are your first iterative examples.

L = [10, 20, 30, 40]

N = len(L)

# CALCULATE THE AVG FIRST (COPY-PASTE FROM THE FIRST EXAMPLE)

sum = 0
i = 0

while i < N:
sum = sum + L[i]
i = i + 1

avg = sum / N

# CALCULATE THE STD NOW

sum = 0
i = 0

while i < N:
sum = sum + (L[i] - avg)**2
i = i + 1

std = (sum / N)**0.5

print("Avg & Std of the list are: ", avg, std)

5.2. Repetitive execution 101

Avg & Std of the list are: 25.0 11.180339887498949

Exercise
Write down the algorithm that we used in this Python code as a pseudo-code.
Example 3: Factorial Factorial of a number, denoted by n!, is an important mathematical construct that we
frequently use while formulating our solutions. n! can be defined formally as:
(
n × (n − 1) × ...2 × 1, if n > 0
n! = (5.2.3)
1, if n = 0.
Let us first write down the algorithm:

Input: n
Output: n_factorial

Step 1: If n is zero or negative, n_factorial is 1. Return n_factorial and exit

Step 2: Initialize n_factorial with 1
Step 3: Initialize an index variable, i, with 1
Step 4: While i <= n, Execute Steps 5 and 6:
Step 5: n_factorial = n_factorial * i
Step 6: i = i + 1
Step 7: The result is n_factorial

which can be implemented in Python as follows:

# Let us choose an n value:

n = 5

if n <= 0: n_factorial = 1 # Step 1

elif n > 0: # Steps 2-6
n_factorial = 1 # Step 2
i = 1 # Step 3
while i <= n: # Step 4
n_factorial *= i # Step 5
i += 1 # Step 6

print("n! is ", n_factorial)

n! is 120

Exercise
Modify our factorial implementation such that variable i starts from n.
Example 4: Computing Sine with Taylor Expansion
Let us assume we want to compute sin(x), for a given value of x, up to a given precision. This is just for the
sake of having an understandable example. Actually today’s CPUs have embedded math coprocessors that
do this at microcode level. We will not use this convenience and do the computation ourselves.
The computation of many analytic functions are performed using the Taylor series expansion. The Taylor
series expansion of sin(x) around zero is:
x3 x5 x7
sin (x) ≈ x − + − ··· , (5.2.4)
3! 5! 7!

102 Chapter 5. Conditional and Repetitive Execution

which could also be written as:
∞
X (−1)k 2k+1
x . (5.2.5)
(2k + 1)!
k=0

Let us implement this. We will use the factorial example as a nested iteration. The outer iteration runs over
the terms in the summation, and continues until a term becomes too small (|term| < ϵ, where ϵ is a small
value, e.g. 10−12 ).

epsilon = 1.0E-12
x = 3.14159265359/4.0 # i.e. pi/4 (45 degrees in radians)
result = 0.0
k = 0
term = 2*epsilon #just a trick to bypass the first test
while abs(term) > epsilon:
# Calculate the denominator - i.e. (2k+1)!
factorial = i = 1
while i <= 2*k+1:
factorial *= i
i += 1

# Now calculate the term

term = (((-1)**k) / factorial) * (x**(2*k+1))

result += term
k += 1

This is a readable but at the same time quite an inefficient code:

* 2*k+1 is computed several times. Actually, do we need 2*k+1 at all? * It uses factorial computation
which does not make use of its preceeding computations. For example 9! is actually 9 × 8 × 7!. But it is
computed as 2 × 3 × 4 × 5 × 6 × 7 × 8 × 9. Therefore, there is significant inefficiency regarding the use of
factorial in the nested iteration. * Similar to the inefficiency with the factorial computation, the
computation of x(n+2) does not make use of the already computed xn value.
* Do we need to compute (−1)k to get an alternating sign? Can there be a simpler way to program this?

Let us investigate the proportionality of two consecutive terms (k = n) and (k = n + 1):

termn+1 (−1)(n+1) x2(n+1)+1 (2n + 1)! −x2

= × = . (5.2.6)
termn (2(n + 1) + 1)! (−1)n x2n+1 (2n + 2)(2n + 3)

This is interesting because it suggests us a relatively faster way to compute the next term in the series. We
do not have to compute xn for each new term that we are going to add. We also discover that there is nothing
magical about (2n + 1). It can simply be called as (d ≡ 2n + 1) which increments by 2 for each consecutive
term. Then we have
termn+1 −x2
= (5.2.7)
termn (d + 1)(d + 2)

Now, let us see the more efficient implementation in action:

5.2. Repetitive execution 103

epsilon = 1.0E-12
x = 3.14159265359/4.0 # i.e. pi/4 (45 degrees in radians)
x_square = x*x
term = x
result = term
d = 1 # 2*n+1
while abs(term) > epsilon:
term *= -x_square/((d+1)*(d+2))
result += term
d += 2

print("sin(x) [Ours] is: ", result)

# Compare this with the CPU-implemented version

from math import *
print("sin(x) [CPU ] is: ", sin(x))

sin(x) [Ours] is: 0.707106781186584

sin(x) [CPU ] is: 0.7071067811865841

5.2.3 for statement

The syntax of the for statement is:

for V ariable in Iterator : Statement

An iterator is an object that provides a countable number of values on demand. It has an internal mechanism
that responds to three requests:
1. Reset (initialize) it to the start.
2. Respond to the “Are we at the end?” question.
3. Provide the next value in line.
The for statement is a mechanism that allows traversing all values provided by an Iterator, one-by-one
assigning them to a V ariable and then executiong a given Statement.
The semantics of the for statement is displayed in Fig. 5.2.2:

104 Chapter 5. Conditional and Repetitive Execution

Fig. 5.2.2: The flowchart illustrating how a for statement is executed.

What iterators do we have?

1. All containers are also iterators (strings, lists, dictionaries, sets).
2. The built-in function range returns an iterator that generates a sequence of integers, starting from 0
(default), and increments by 1 (default), and stops before a given number:
range(start, stop, step)

Parameter Opt./Req. Default Description

start Optional 0 An integer specifying the start value.
stop Required An integer specifying the stop value (stop value will not be taken)
step Optional 1 An integer specifying the in crementation.

3. A user-defined iterator. Since this is an introductory book, we will not cover how this is done.

5.2.4 Examples with for statement

Example 1: Words Starting with Vowels and Consonants

Let us split a list of words into two lists: those that start with a vowel and those that start with a consonant.

mixed = ["lorem","ipsum","dolor","sit","amet,","consectetur","adipiscing","elit","sed","do",
,→"eiusmod","tempor","incididunt","ut","labore","et","dolore","magna","aliqua"]
vowels = []
consonants = []
for word in mixed:
if word[0] in ['a','e','i','o','u']: vowels += [word]
else: consonants += [word]

print("Starting with consonant:", consonants)

print("Starting with vowel:", vowels)

5.2. Repetitive execution 105

Starting with consonant: ['lorem', 'dolor', 'sit', 'consectetur', 'sed', 'do', 'tempor',
,→'labore', 'dolore', 'magna']

Starting with vowel: ['ipsum', 'amet,', 'adipiscing', 'elit', 'eiusmod', 'incididunt', 'ut',
,→'et', 'aliqua']

Exercise
Write this solution in pseudo-code.
Example 2: Run-length Encoding
Run-length encoding is a compression technique for data that repeats itself contiguously. Images bear such
data: For example, a clear sky portion is actually rows of ‘blue’. In the example below, we investigate a string
for consequent character occurrences: If found, repetitions are coded as a list of [character, repetition
count].

So, a text of "aaaaaaxxxxmyyyaaaassssssssstttuivvvv" gets encoded as:

[['a',6],['x',4],'m',['y',3],['a',4],['s',9],['t',3],'u','i',['v',4]].

text = "aaaaaaxxxxmyyyaaaassssssssstttuivvvv"
code_list = []
last_character = text[0]
count = 1

# Go over each character except for the first

for curr_character in text[1:]:
# If curr_character is equal to last_character, we found a duplicate
if last_character == curr_character:
count += 1
else:
# We have finished a sequence of same characters: Save the count and
# reinitialize last_character and count accordingly
code_list += [last_character if count==1 else [last_character, count]]
count = 1
last_character = curr_character

# handle the last_character here:

code_list += [last_character if count==1 else [last_character,count]]

print(code_list)

[['a', 6], ['x', 4], 'm', ['y', 3], ['a', 4], ['s', 9], ['t', 3], 'u', 'i', ['v', 4]]

Exercise
Modify this code such that it removes consecutive duplicate characters. In other words, "aaaabbbcdd"
should be reduced to "abcd".
Example 3: Permutations
Permutations are keys to shuffling a sequence of data. Permutations can be denoted in various forms. One
way is to give an order list: The ith element of the list stores the old position where the ith element of the
new sequence will come from. In our implementation counting starts from 0 (zero).

106 Chapter 5. Conditional and Repetitive Execution

word_list = ["he","came","home","late","yesterday"]
permutation = [4,3,2,0,1]
length = len(permutation)
new_list = [None]*length

for i in range(length):
new_list[i] = word_list[permutation[i]]

print(new_list)

['yesterday', 'late', 'home', 'he', 'came']

Example 4: List Split

In the example below the first element of the list is taken and the rest is split into two lists: Those that are
smaller than the first element, and those that are not.

list_to_split = [42, 59, 53, 84, 43, 8, 75, 34, 40, 89, 29, 15, 51, 6, 90, 32, 58, 77, 4, 24]
list_smaller = []
list_not_smaller = []

for x in list_to_split[1:]:
if x < list_to_split[0]: list_smaller += [x]
else: list_not_smaller += [x]

print("list head was. :",list_to_split[0])

print("smaller than head :", list_smaller)
print("not smaller than head :", list_not_smaller)

list head was. : 42

smaller than head : [8, 34, 40, 29, 15, 6, 32, 4, 24]
not smaller than head : [59, 53, 84, 43, 75, 89, 51, 90, 58, 77]

Example 5: Dot Product

A frequently used operation with vectors is the dot product between two vectors, e.g. u and v:

X
n
u·v= ui vi , (5.2.8)
i=1

where the vectors have n elements each. Let us implement this in Python:

# Define the vectors (as lists)

u = [1, 2, 4, 10]
v = [10, 4, 2, 1]
n = len(u)
dot_prod = 0

if n != len(v): print("Sizes don't match!")

else:
for i in range(n):
dot_prod += u[i] * v[i]
print("u . v is: ", dot_prod)

5.2. Repetitive execution 107

u . v is: 36

Exercise
Implement dot product between two vectors using a while statement.
Example 6: Angle between Two Vectors
The angle between two vectors u and v is closely related to their dot product:

u · v = ||u|| ||v|| cos(θ), (5.2.9)

where θ is the angle between the vectors, and || · || denotes the norm of a vector:
v
u n
√ uX
||u|| = u · u = t u2i . (5.2.10)
i=1

The angle can be derived from these as follows:

u·v
θ = arccos . (5.2.11)
||u|| ||v||

Let us implement this:

from math import sqrt, acos

# Define two vectors that have 90deg (pi/2) between them

u = [1, 0, 0]
v = [0, 1, 0]
n = len(u)

if n != len(v):
print("Sizes don't match!")

else:
# Calculate dot product & norms
dot_prod = u_norm_sum = v_norm_sum = 0
for i in range(n):
dot_prod += u[i] * v[i]
u_norm_sum += u[i] ** 2
v_norm_sum += v[i] ** 2

u_norm = sqrt(u_norm_sum)
v_norm = sqrt(v_norm_sum)

theta = acos(dot_prod / u_norm * v_norm)

print("angle between u and v is: ", theta)

angle between u and v is: 1.5707963267948966

Example 7: Matrix Multiplication

Our last example is matrix multiplication. As you know, a matrix A can be multipled with the matrix B if
the column-count (m) of A is equal to the row-count of B. The product matrix is of a size where row-count

108 Chapter 5. Conditional and Repetitive Execution

equals that of A and the column-count equals that of B.

X
m−1
(A · B)ij = Aik · Bkj . (5.2.12)
k=0

As we discussed in the previous chapter, since Python provides no two dimensional containers, but a very
fexible list container, matrices are mostly represented as lists of lists each of which represent a row. For
example, a 3 × 4 matrix:
 
−2 3 5 −1
 0 3 10 −7 (5.2.13)
11 0 0 −8

can be represented as:

A = [[-2,3,5,-1], [0,3,10,-7], [11,0,0,-8]]

This representation is also coherent with the indexing of matrices: Aij is accessible in Python as A[i][j].
The code below multiples two matrices, given in this representation, and prints the result.

A = [[-2,3,5,-1], [0,3,10,-7], [11,0,0,-8]]

B = [[2,1], [-1,1], [0,4], [8,0]]

# C = A*B

# Create a result matrix with entries filled with 0 (zeros)

C = []
for i in range(len(A)):
C += [[0] * len(B[0])] # This is done to overcome the aliasing problem

for i in range(len(C)):
for j in range(len(C[0])):
for k in range(len(B)):
C[i][j] += A[i][k]*B[k][j]

print(C)

[[-15, 21], [-59, 43], [-42, 11]]

5.2.5 continue and break statements

It is possible to alter the flow of execution in a while or for repetition. This is almost always used in
combination with statement grouping, where a group of statements is subject to the while or for.
Let’s assume you are executing a sequence of statements by means of statement grouping. Somewhere in the
process, without waiting for the terminating condition to become False in a while statement or exhausting
all items in the iterator in a for statement, you decide to terminate looping. The break statement does
exactly serve this purpose: The while (or for) is stopped immediately and the execution continues with the
statement after the while (or for).
Similarly, somewhere in the process you decide not to execute the rest of the statements in the grouping and
straight away continue with the test. This is achieved by a continue statement usage. See Fig. 5.2.3 and
Fig. 5.2.4 for how the continue statement changes the flow of execution.

5.2. Repetitive execution 109

Fig. 5.2.3: How continue and break statements change execution in a while statement.

Fig. 5.2.4: How continue and break statements change execution in a for statement.

5.2.6 Set and list comprehension

A well-known and extensively-used notation to describe a set in mathematics is the so-called ‘set-builder
notation’. This is also known as set comprehension. This notation has three parts (from left-to-right):
1. an expression in terms of a variable,
2. a colon or vertical bar separator, and
3. a logical predicate.
Something like this:

{x3 | x ∈ {0, 1, . . . , 7}} (5.2.14)

which defines the set:

{0, 1, 8, 27, 64, 125, 216, 343} (5.2.15)

110 Chapter 5. Conditional and Repetitive Execution

or as something more elaborate:

x3 | x ∈ {0, 1, . . . , 7} ∧ (x mod 2) = 1 (5.2.16)
defining
{1, 27, 125, 343} (5.2.17)
There exists a Python notation with the same semantics both for sets and lists:

>>> [x**3 for x in range(8)]

[0, 1, 8, 27, 64, 125, 216, 343]
>>> {x**3 for x in range(8)}
{0, 1, 64, 8, 343, 216, 27, 125}

The second example, which imposes a constraint on x to be an odd number, would be coded as:

>>> [x**3 for x in range(8) if x%2 == 1]

[1, 27, 125, 343]
>>> {x**3 for x in range(8) if x%2 == 1}
{1, 27, 125, 343}

Certainly the condition (following the if keyword) could have been constructed as a more complex boolean
expression.
The expression that can appear to the left of the for keyword can be any Python expression and container.
Here are a couple of more examples on matrices.

print( [[0 for i in range(3)] for j in range(4)] )

print( [[1 if i==j else 0 for i in range(3)] for j in range(3)] )
print( [[(i,j) for i in range(3)] for j in range(4)] )
print( [[(i+j)%2 for i in range(3)] for j in range(4)] )
print( [[i+3*j for i in range(3)] for j in range(4)] )

[[0, 0, 0], [0, 0, 0], [0, 0, 0], [0, 0, 0]]

[[1, 0, 0], [0, 1, 0], [0, 0, 1]]
[[(0, 0), (1, 0), (2, 0)], [(0, 1), (1, 1), (2, 1)], [(0, 2), (1, 2), (2, 2)], [(0, 3), (1,␣
,→3), (2, 3)]]

[[0, 1, 0], [1, 0, 1], [0, 1, 0], [1, 0, 1]]

[[0, 1, 2], [3, 4, 5], [6, 7, 8], [9, 10, 11]]

5.3 Important Concepts

We would like our readers to have grasped the following crucial concepts and keywords from this chapter:
• Conditional execution with if, if-else, if-elif-else statements.
• Nested if statements.
• Conditional expression as a form of conditional computation in an expression.
• Repetitive execution with while and for statements.
• Changing the flow of iterations with break and continue statements.
• Set and list comprehension as a form of iterative computation in an expression.

5.3. Important Concepts 111

5.4 Further Reading

• “Managing the size of a problem”, Chapter 4 of “Introduction to Programming Concepts with Case
Studies in Python”: https://link.springer.com/chapter/10.1007/978-3-7091-1343-1_4

5.5 Exercises

We have provided many exercises throughout the chapter, please study and answer them as well.
• Implement the while statement examples in Section 5.2.2 using for statement:
– Calculating the average of a list of numbers.
– Calculating the standard deviation of a list of numbers.
– Calculating the factorial of a number.
– Taylor series expansion of sin(x).
• Implement the for statement examples in Section 5.2.4 using while statement:
– Words with vowels and consonants.
– Run-length encoding.
– Permutation.
– List split.
– Matrix multiplication.
• What does the variable c hold after executing the following Python code?

c = list("address")
for i in range(0, 6):
if c[i] == c[i+1]:
for j in range(i, 6):
c[j] = c[j+1]

• Write a Python code that removes duplicate items in a list. E.g. [12, 3, 4, 12] should be changed
to [12, 3, 4]. The order of items should not change.
• Write a Python code that replaces a nested list of numbers with the average of the numbers. E.g. [[1,
3], [4, 5, 6], 7, [10, 20]] should yield [2, 5, 7, 15].
• Write a Python code that finds all integers dividing a given integer without a remainder. E.g. for
number 21, your code should compute [1, 3, 7].
• Write a Python code that converts an integer (less than 128) into binary string using a for/while state-
ment. E.g. 21 should be represented as 00010101.
• Write a Python code that convers a binary string like 00010101 into integer.

112 Chapter 5. Conditional and Repetitive Execution

6 | Functions

6.1 Why define functions?

There are four main reasons for using functions while developing solutions with a programming language:
1- Reusability
A function is defined at a single point in a program but it can be called from several different points. Assume
you need to calculate the run-length encoding of a string (the example code from the previous chapter) in
your program for n different strings in different positions of your code. In a programming language without
functions, you would have to copy-paste that code piece in n different places. Thanks to functions, we can
place the code piece in a function once (named encodeRunLen for example) and use it in n places via simple
function calls like enc = encodeRunLen(inpstr).
If functions would not be present, even if we would have the piece of code that does the actions of the
function, how would we return to the different calling points after the actions are completed? Something,
that is theoretically possible but extremely painful.
2- Maintenance
Using functions makes it easier to make updates to the algorithm. Let us explain this with the run-length
encoding example: Assume that we found a bug in our calculation or we have discovered a better algorithm
than the one we are using. Without functions, you would need to go through all your code, find the locations
where your run-length encoding algorithm is implemented, and update each and every one of them. This is
impractical and very tedious. With functions, it is sufficient if you just change the function that implements
the run-length encoding.
3- Structuredness
Let us have a look at a perfectly valid Python expression:

113
(b if b<c else c) if (a if a<b else b)<(b if b<c else c) else (a if a<b else b)

which certainly takes some time to parse and understand but it just performs the following:

max(min(a,b), min(b,c))

Assume instead of the long expression above, the one with min and max are used. Even if the definition of
max and min are not given yet, still one can grasp what the expression is up to. It is not only because the
expression is smaller, but because function naming can depict the meaning for the actions.
4- Benefits of the functional programming paradigm
If a programmer sticks to the functional paradigm, then he or she has to plot a functional decomposition
so that the solution of the problem is obtained by means of functional composition, something that is well-
known and understood through Mathematics. This type of a solution does not suffer from unexpected side
effects or variable alterations; therefore, testing and debugging are easier. Another benefit of the functional
paradigm is recursion: In other words, while defining a function, we can call the function that is being defined
– we will cover this seemingly confusing concept below. Moreover, we can define higher-order functions
that take functions as input and apply them to a sequence of data. All these benefits come free with using
functions, as we will illustrate below.
Even when we do not limit ourselves to ‘functional programming’, it is essential to use functions in profes-
sional programming. General algorithms that are coded as functions are becoming very valuable because
they can be reused again and again in new programs. Collections formed by such general purpose functions
are called libraries. For almost all application areas libraries have been created over decades.

6.2 Defining functions

For defining a function, Python adopts the following syntax:

def F unctionN ame ( P arameter1 , P arameter2 ,· · · ) : Statement

The Statement, which can be replaced by multiple statements like we did with other compound statements,
can make use of the P arameteri s to obtain the actual arguments sent to the function at call-time. If the
function is going to give a value in return to the call, this value is provided by a return statement.
F unctionN ame is the name of the function being defined. Python follows the same naming rules and
conventions used for naming variables – if you do not remember them, we recommend you to go back and
check the restrictions for naming variables in Chapter 4.
Here is a straightforward example. Let us assume that we want to implement the following function:
m1 m2
Fgravity (m1 , m2 , r) = G . (6.2.1)
r2
The following is the one-line Python code that defines it:

def F_gravity(m_1, m_2, r): return G * m_1 * m_2 / (r * r)

This function, F_gravity, returns a value. This is provided by the return statement. As you may have
observed, the values m1 , m2 and r are provided as the first, the second and the third parameters, named as
m_1, m_2 and r, respectively.

114 Chapter 6. Functions

But what about G? It is not provided in the arguments and used in the return expression for its value. If the
definition were a multi-statement definition (indented multi-statements following the def line), then Python
would seek for a defined value for G among them. But this is not even the case, so the last alternative is to
seek a value in the global environment, the environment in which F_gravity is called.
Therefore, before we call F_gravity, we must make sure that a value for a global variable G (which appar-
ently is the Gravitational Constant) is set. In the Scope of Variables section, we will be handling the rules
of this to a greater extent.

def F_gravity(m1, m2, r): return G * m1 * m2 / (r * r)

G = 6.67408E-11
print(F_gravity(1000, 20, 0.5), "Newton")

5.3392640000000006e-06 Newton

6.3 Passing parameters to functions

When a function is called, first its arguments are evaluated. The function will receive the results of these
evaluations. The parameter variables used in the definition are created, and set to the result of the argument
evaluations. Then the statement that follows the column (:) after the argument parenthesis is executed.
Certainly this statement can and will make use of the parameter variables (which are set to the result of the
argument evaluations).
Let us look at an example:

def F_gravity(m_1, m_2, r): return G * m_1 * m_2 / (r * r)

G = 6.67408E-11

S = 100
Q = 2000

print(F_gravity(Q+S, Q-S, ((504.3-66.1)2+(351.1-7.7)2)**0.5))

8.59177215925003e-10

This is a legitimate piece of code that performs the following in order:

• Defines the function F_gravity.
• Sets the value of the global variable G and S and Q
• Calls the (internally defined) print function which receives one argument. This argument is a function
call to F_gravity. So, in order to have a value to print, this function call has to be performed.
• A call to F_gravity is prepared: It has three arguments, each has to be evaluated and boiled down to
a value. So, in a left-to-right order the arguments are evaluated:
– Q+Sis evaluated, and the result is stored into m_1.
– Q-Sis evaluated, and the result is stored into m_2.
√
– Euclidean distance, (504.3−66.1)2 +(351.1−7.7)2 , is evaluated and the result is stored into r.

6.3. Passing parameters to functions 115

• Since all arguments are evaluated, the statement defining the function F_gravity can be executed.
• This statement is return G * m_1 * m_2 / (r * r), which requires the expression following the
return keyword to be evaluated and the result to be returned as the result of the function call.
• Now the print function has got its parameter evaluated. The defining action of the print function
can be carried out (This is an internally defined function, i.e. we cannot see its definition, but the
documents defining the language explain it).
The bottom line is: Before the function is called, there is a preparation phase where all arguments are eval-
uated and hence converted to a value. This has a name in computer science and is coined as Call-by-value.

6.3.1 Default Parameters

While defining a function, we can provide default values to parameters as follows:

def F unctionN ame ( P arameter1 = Exp1 , P arameter2 = Exp2 , · · · ) : Statement

where Expi is an expression. When calling the function, we may omit providing the parameters for which
the function has default values.
Let us have a look at a simple example:

def norm(x1, x2, norm_type="L1", verbose=True):

result = None
if norm_type == "L1": result = abs(x1) + abs(x2)
elif norm_type == "L2": result = (x1**2 + x2**2)**0.5
elif verbose: print("Norm not known:", norm_type)

if verbose: print(f"The {norm_type} norm of", [x1, x2], "is:", result)

return result

norm(3, 4) # CASE 1
norm(3, 4, "L2") # CASE 2
norm(3, 4, norm_type="L2") # CASE 3
norm(3, 4, verbose=False) # CASE 4: Does not print the value
norm(3, 4, verbose=True, norm_type="L1") # CASE 5
norm(x2=3, x1=4, verbose=True, norm_type="L1") # CASE 6

The L1 norm of [3, 4] is: 7

The L2 norm of [3, 4] is: 5.0
The L2 norm of [3, 4] is: 5.0
The L1 norm of [3, 4] is: 7
The L1 norm of [4, 3] is: 7

# Let us look at some cases that lead to errors:

>>> norm(3, 4, norm_type="L1", True) # CASE 7
File "<ipython-input-19-128dc3e43256>", line 2
norm(3, 4, norm_type="L1", True) # CASE 7
^
SyntaxError: positional argument follows keyword argument

116 Chapter 6. Functions

In this example, you should notice the following crucial points:
• We can call a function without providing any value for the parameters that have default values (CASE
1, .., CASE 4).
• We can swap the order of default parameters when we are calling the function (CASE 5 and CASE 6).
• When calling a function, we can give values to non-default parameters by using their names (CASE 6).
• The parameters after the default parameters need to be given names as well, both while defining the
function and calling the function (CASE 7).

6.4 Scope of variables

We extensively use variables while programming our solutions. Not all variables are accessible from every-
where. There is a governing set of rules for that determine which parts of a code can access a variable. The
code part from which a variable is accessible is called its scope.
In Python, there are four categories of scopes. As will be explained shortly, they are abbreviated with their
initials , which then combine to form the rule mnemonic LEGB standing for:
[Local < Enclosing < Global < Built-in]
1- Local Scope
Defining a variable in a function in Python makes it local.
Local variables are the most ‘private’ ones. Any code which is external to that function cannot see it. If you
make an assignment anywhere in the function to a variable, it is registered as a local variable. If you attempt
to refer to such a local variable prior to the assignment, Python will generate a run-time error.
The local variables are created each time the function is called (it is activated) and are annihilated (destructed)
the moment the function returns (finishes).
2- Enclosing Scope
As we mentioned before, it is possible to define functions inside functions. They are coined as local functions
and can refer (but not change) to the values of the local variables of its parent function (the function in which
the local function is defined). The scope of the outer function is the enclosing scope for the inner function.
Though the parent function’s variables are visible to the local function, the converse is not true. The parent
function cannot access its local function’s local variables. Therefore, the accessibility is from inward to
outward and for referring purposes only.
3- Global Scope
If a variable is not defined in a function then it is a global variable. It can be accessed outside as well as
inside of any function. If a global variable is accessed inside a function, it cannot be altered there. If it
is altered anywhere in the function, this is recognised at the declaration-time and that variable is declared
(created) as a local variable. A local variable that has the same name with a global variable conceals the
global variable.
If you want to change the value of a global variable inside a function, then you have to explicitly declare it
as a global variable by a statement:

global Variable

Then an assignment to Variable will not register it as a local variable but merely refer to the global Variable.
As said, since the system will recognize them automatically, there is no need to use the global statement for

6.4. Scope of variables 117

global variables which are referred to for their values only. However, it is a good programming habit to do so
in order to enhance readability and prevent accidental local variable creations (due to careless assignments
to global variables).
4- Built-in Scope
Built-in scope is a special Python scope that is created or loaded whenever you run a script or open an
interactive session. This is actually the scope in which the Python interpreter starts. It contains names such
as keywords, built-in functions, exceptions, and some other attributes. Entities in this scope are also available
from everywhere in your code.
An Interactive Example for Scope
Please follow the Colab link at the top of the page for an interactive demo on scope of variables.
A variable occurrence has different meanings for different positions in the program:
1. First time occurrence of the variable at the left-hand side of an assignment: This tells Python to declare
a variable in the current scope, the current function or the global scope. If there is a variable with the
same name in an outer scope, it is hidden as long as the new variable is active. This declaration will
be active for the scope it is created in including statements preceding it.
2. Second and later occurrences of the variable at the left-hand side of an assignment: They simply update
the previously declared variable.
3. Occurrences of the variable at the right-hand side of an assignment or any other place in an expression:
Python places value (content) of the variable in the expression position. Variable is searched in all
scopes from inner to outer (local, then enclosing, then global, then built-in) and the first found variable
is used. If the declaration of the variable is after its reference in current scope, Python will raise the
error 'Reference before assignment'.
Note that global declaration changes the behaviour above. It makes all occurrences to refer to the global
variable and does not declare a variable in the current scope with the globally defined names.

6.5 Higher-order functions

Python provides several flavours of the functional paradigm, one of the most powerful of which is the ability
to easily pass functions as parameters to other functions. No special syntax or rule is used for this: If you
use a parameter as if it is a function and when calling the higher-order function, pass a function name as a
parameter, Python will handle the rest.
Python provides two very practical higher-order functions for us:
• map function: The map function has the following syntax:
map(function, Iterator)
which yields an iterator where each element is function being applied on the corresponding element
in Iterator. Here is a simple example:

>>> abs_it = map(abs, [10, -4, 20, -100])

>>> for x in abs_it: print(x)
10
4
20
100

118 Chapter 6. Functions

• filter function: The filter function has a similar syntax to the map function:
filter(predicate, Iterator)
where predicate is a function with a boolean return value. The filter function applies predicate
to each element of the Iterator and only keeps the ones for which predicate returns True. For
example:

>>> def positive(x): return x > 0

>>> for x in filter(positive, [10, -4, 20, -100]): print(x)
10
20

Of course, the functional capabilities of Python is more diverse than we can cover in this introductory text-
book. The reader is referred to Python docs23 . The curious reader is especially recommended to look at
lambda expressions and the reduce function.

6.6 Functions in programming vs. functions in Mathematics

It is to be noted that functions have a differentiated meaning compared to functions in Mathematics.

• In Mathematics, a function is a mapping from a domain to a range (target); a function returns a value.
In Python: Functions may not return a value at all. If so, they are intended to code a parametrized
action. A few examples to this are:
– Drawing a red line from (x0 , y0 ) to (x1 , y1 ).
– Printing a separating line made of n ‘minus’ signs.
– Turning off the lights.
– Sending a digital initialization string to a device connected through the USB port.
• In Mathematics, a function definition is global. You do not have function definitions local to some
function.
In Python: Since a function definition is done by a Python Statement (or a group of statements –
grouped by indentation), it is quite possible to have another definition as part of the defining state-
ments. Such a function can only be usable in the Statement group that defines the outer function and
access its local variables and parameters. Outside of the function, this interior function (called a local
function) will not be visible nor usable.
• In Mathematics, an element from the domain is mapped to one and only one target element by the
function. The mapping does not change. Moreover, a mathematical function returns a value that only
depends on the arguments. Hence, for the same arguments, the result is the same.
In Python: Even if such a mapping exists (i.e. a function returns a value), it is possible that the function
acts differently due to different settings of the variables (outside of the function) that can be accessed in
the functions. Consider the F_gravity function above: A different setting for the G (the Gravitational
Constant) will yield a different F_gravity result even when the parameters are the same.
• In Mathematics, a function’s operation is reflected on the value it returns: It does not have side effects.
In other words, a mathematical function does not change the mathematical environment that is external
to it (variables or the ways other functions are evaluated).
23
https://docs.python.org/3/howto/functional.html

6.6. Functions in programming vs. functions in Mathematics 119

In Python: This may not be true. Consider the G value of the F_gravity function above. Hypotheti-
cally, it is possible that, at each call of F_gravity, the G value is increased slightly (1% for example):

>>> def F_gravity(m1, m2, r):

... global G # since we alter it
... G = 1.01 * G
... return G * m1 * m2 / (r * r)

>>> G = 6.67408E-11

>>> print(F_gravity(1000, 20, 0.5), "Newton")

5.39265664e-06 Newton

>>> print(F_gravity(1000, 20, 0.5), "Newton")

5.4465832064e-06 Newton

>>> print(F_gravity(1000, 20, 0.5), "Newton")

5.501049038464e-06 Newton

In these aspects, a function defined in Python has a greater freedom. As previously stated, there are paradigms
of programming. One of these paradigms, the functional paradigm, promotes the function’s usage to be
restricted to the mathematical sense. In other words, the functional paradigm followers deliberately renounce
the additional freedom programming languages like Python provides in function definitions and usages with
respect to the mathematical sense. As a rule-of-thumb we can say that deviating from the mathematical sense
makes your programming error prone.

6.7 Recursion

Recursion is a function definition technique in which the definition makes use of calls to the function being
defined. Although it is difficult to grasp, it allows compact and more readable code for recursive algorithms,
relations or mathematical functions.
It is customary to give the infamous example of the recursive definition of factorial. First, let us put it down
in mathematical terms:
N ! = (N − 1)! × N, for N ∈ N, N > 0
0! = 1,

Based on this, we can realize that:

• 6! is 6 × 5!
• 5! is 5 × 4!
• 4! is 4 × 3!
• 3! is 3 × 2!
• 2! is 2 × 1!
• 1! is 1 × 0!
• 0! is 1
Substituting any factorial value, based on this rule, one ends up with

6! = 6 × 5 × 4 × 3 × 2 × 1 × 1. (6.7.1)

120 Chapter 6. Functions

Making use of (N − 1)! while we are defining N ! is making a recursive definition. The Python equivalent
of the above given recursive mathematical definition of the factorial would be:

def factorial(N):
if N == 0: return 1
else: return N * factorial(N-1)

High-level languages are designed so that each call to the factorial function does not remember any previ-
ous call(s) even if they are still active (unfinished). For every new call, fresh positions for all local variables
are created in the memory without forgetting the former values of variables for the unfinished calls.
From the programmers point of view, the only concern is the algorithmic answer to the question:
“Having the recursive call returned the correct answer to the smaller problem, how can I cook
up the answer to the actual problem?”
It requires a kind of ‘stop thinking’ (leap of faith) action: ‘Stop thinking how the recursive call comes up
with the correct answer’. Just concentrate on the next step paraphrased in the quotation above.
Not to get into an infinite, non-stopping recursive loop, you should be careful about two aspects:
1. The recursive call should be made with a decrescent (getting smaller) argument. In the factorial ex-
ample, calculation of N ! recursively called (N − 1)!, i.e. a smaller value than N !.
2. The function definition must start with a test for the smallest possible argument case (in the factorial
example, this was 0 (zero)) where a value for this case is directly returned (in the factorial example
this was 1 (one)). This is called the termination or base condition.
Let us consider another example: We have a list of unordered numbers. Nothing is pre-known about the
magnitude of the numbers. The problem is to find the maximum of the numbers in that list.
Let us say you are given a list of, say, 900 numbers. For a second, assume that you have a very loving
grandpa. Your grandpa has a rule of his own: not to corrupt you, he will not solve the problem for you but,
if you reduce the problem slightly, his heart softens and provides the answer for the smaller problem.
Presumably, the simplest ‘reduction’ that you can do on a 900-element list is removing an element from it and
making it a list of 899 numbers. As far as Python lists are concerned, the simplest of the simple is to remove
the first element: If L is your list, L[0] is the first element and L[1:] is the list of remaining elements. So,
the trick is to remove the first element, take L[1:] to your grandpa (that is making the recursive call), and
ask for the solution. Your grandpa is very cooperative and gives you the solution (the biggest number) for
the list (L[1:]) you passed to him.
However, we are not finished: On one hand you have what your grandpa returned to you as the biggest
number of the slightly reduced list (L[1:]), and on the other hand, you have the removed (first) element
L[0]. Now, which one will you return as the solution of the 900 number list? Think about it!
Here is the solution in Python:

def find_max(L):
if len(L) == 1: return L[0] # Terminating condition
else:
grandpa_result = find_max(L[1:]) #recursion is on L with the 0th element␣
,→removed
if grandpa_result > L[0]: return grandpa_result
else: return L[0]

# Let's try with a simple list -- try changing M and check the result
(continues on next page)

6.7. Recursion 121

(continued from previous page)
M = [-100, 10, -10, 100]
print("The maximum element of M is:", find_max(M))

The maximum element of M is: 100

As you may have recognized, the definition started with testing for the smallest case. That is a single element
list for which the result (the biggest number in the list) is that single number itself.
Recursion helps writing elegant functions. However, it does not come for free. The price is in the time and
memory overhead spent on creating a new function call. Time may be something tolerable but memory is
limited. For functions that make thousands of recursive calls, you should better seek algorithmic solutions
that do not lean on recursion. Python has a default for recursion depth, which is set to 1000. This can be
altered. As said, it is better to seek a non-recursive solution unless you know very well of what you are
doing.
Any recursive function can be implemented using iterations. The reverse is also true: Any iterative function
can be implemented using recursion. The definition of the problem or the algorithm generally provides hints
for whether recursion or iteration is better for implementing your solution.

6.8 Function Examples

Example 1: Computing average and standard-deviation

In the previous chapter, we provided iterative definitions for calculating the average and the standard-
deviation of a list of numbers. Let us look at these definitions again with functions. For the sake of ease,
the mathematical definitions are provided again. Moreover, we will use this opportunity to talk more about
iterations as well.
Average of a set (S) of numbers is calculated by dividing the sum of the values in the set by their number.
So, if we have a set S:
1 X
avg(S) = x, (6.8.1)
|S|
x∈S

which can be coded as:

def avg(S):
sum = 0.0
for x in S: sum += x
return sum/len(S)

S = [11, 2, 9, 6]
print("The avg of S:", avg(S))

The avg of S: 7.0

This is quite elegant. As you see it works at a higher abstraction level. This can be done because the in
operator comes in very handy and when used as an iterator, it iterates over the elements of the list.

122 Chapter 6. Functions

Sometimes it is more desirable to work with the indices to reach out to each element. Remember that indexing
of container elements start at zero. So, a more index-based formulation of the same formula would be:
|S|−1
1 X
avg(S) = Si (6.8.2)
|S|
i=0

The following implements this definition:

def avg(S):
sum = 0.0
for i in range(0, len(S)): sum += S[i]
return sum/len(S)

S = [11, 2, 9, 6]
print("The avg of S:", avg(S))

This is almost okay but not perfect. len is a built-in function that returns the element count of any container
(string, list, tuple). As you have observed len(S) is calculated twice. This is inefficient. Function calls
have a price, time-wise and space-wise, and therefore, unnecessary calls should be avoided. To fix this, we
change the code to do computation once and store the result for further usages.

def avg(S):
sum = 0.0
length = len(S)
for i in range(0, length): sum += S[i]
return sum/length

S = [11, 2, 9, 6]
print("The avg of S:", avg(S))

The avg of S: 7.0

sum, length, i and the parameter S are all local variables of the function avg. In contrary to many other
programming languages, in Python we do not have to declare locals. They are automatically recognized.
This comes in very handy.
One point to note about the code is that the |S| − 1 is missing. It is not necessary because the range iterator
excludes the end value, as we discussed in the previous chapter.

Now let us also introduce the code for standard deviation. First the mathematical definition:

v
u
u 1 |S|−1
X
std(S) = t (Si − avg(S))2 (6.8.3)
|S|
i=0

This time we bear in mind that unnecessary function calls should be avoided:

6.8. Function Examples 123

def avg(S):
sum = 0.0
length = len(S)
for i in range(0, length): sum += S[i]
return sum/length

def std(S):
sum = 0.0
length = len(S)
average = avg(S)
for i in range(0, length): sum += (S[i] - average)**2
return (sum/length)**0.5

S = [11, 2, 9, 6]
print("The avg of S:", avg(S))
print("The std of S:", std(S))

The avg of S: 7.0

The std of S: 3.391164991562634

We gave the solution in terms of explicit indexing. As given below this particular problem can be coded
without an iteration over indices. Such problems are rare. Problems that require operations over vectors and
matrices or problems that aim optimizations usually require explicit index iterations.
For sake of completeness we state the index-less solution of the average and standard-deviation problem:

def avg(S):
sum = 0.0
length = len(S)
for x in S: sum += x
return sum/length

def std(S):
sum = 0.0
length = len(S)
average = avg(S)
for x in S: sum += (x - average)**2
return (sum/length)**0.5

S = [11, 2, 9, 6]
print("The avg of S:", avg(S))
print("The std of S:", std(S))

The avg of S: 7.0

The std of S: 3.391164991562634

Example 2: Computing check digit

A Standard Definition from Wikipedia24 :
“A check digit is a form of redundancy check used for error detection on identification num-
bers, such as bank account numbers, which are used in an application where they will at least
24
https://en.wikipedia.org/wiki/Check_digit

124 Chapter 6. Functions

sometimes be input manually. It consists of one or more digits computed by an algorithm from
the other digits (or letters) in the sequence input.
With a check digit, one can detect simple errors in the input of a series of characters (usually
digits) such as a single mistyped digit or some permutations of two successive digits.”
Middle East Technical University, our university, also implements a check digit algorithm for student num-
bers. A student number has exactly six digits. The algorithm to generate the checkdigit is as follows:
• sum ← 0
• For each digit in the student number (the leftmost digit is the named as the ‘first’) that has an odd
position (first, third, fifth), take the digit and add it to the sum.
• For each digit in the student number that has an even position (second, forth, sixth) take twice the digit.
If this result of doubling is a two digit number, then add each of these digits to the sum, otherwise add
the result of the doubling (which is a single digit number itself) to the sum.
• The one digit number, which when added to the sum results in a multiple of 10, is the check digit.
Take as example the student number 167912:

Fig. 6.8.1: How checksum is calculated for a student number.

The check digit of the student number 167912 is found to be 5. As you all well know, this is written as
167912-5. That is called the student-id.
Below you will find the code that computes the checkdigit for a student number, given as a string. We
encourage you to investigate it. A very good technique to understand how a code progresses when it runs is
to insert print statements to points of interest. You can start by inserting:

print(digit, sum)

right after the line of else: (left-aligned with the else).

def checksum(S):
sum = 0
for i in range(0, len(S)):
digit = int(S[i])
if i%2 == 0: sum += digit
else: sum += (9 if digit==9 else (2*digit) % 9)
n = sum % 10
return 0 if n == 0 else 10 - n

print(checksum("167912"))

6.8. Function Examples 125

A few notes about the code:

• Remember indexing starts in Python at zero. One group of digits, namely (first, third, fifth) are actually
at S[0], S[2] and S[4]. Their indices are even. Or in mathematical terms, they are (modulo 2) zero.
• Summing the digits of a number is a trick we learned in primary school. It is actually nothing else but
a quick way to do (modulo 9). So, we are actually taking the (modulo 9) of the doubled digit. With
one exception: If the digit is 9, the double is 18 where we have to add a 1+8=9 to the sum. But (18 %
9) would give us zero (not 9). So, here we have to insert an exception: For all digits except 9, we can
use (2*digit) % 9. For 9, we ought to add a 9 to the sum.
• The rule of constructing the check digit from the sum has a similar exception case for sum being a
multiple of 10 (i.e. n (modulo 10) = 0). In that case we would have a checkdigit of 10 (because 10-0 =
10) which is not a digit, but two digits. The designers decided to use a zero as the checkdigit for this
particular case.
Now let us see how the checkdigit can be used to recover a missing digit of the student id.
Let us assume you have a student ID as: 1�9503-7. Now, can we discover what the (�) digit is?
For this purpose we will write a function find_q_mark() that will take a string as an argument, in which the
missing digit is indicated by a (?) (question mark). We are expecting that a call of find_q_mark("1?9503-
7") will recover the missing digit and return the demystified string: “139503-7”.
Note that the checksum() definition of above is still effective.

# places digit in the place of '?' in the string S

def replace_q_mark(S, digit):
new_S =""
for c in S: new_S += str(digit) if c=='?' else c
return new_S

# finds the value for the position with '?'

def find_q_mark(MissingStdID): # MissingStdID is something like "1?7912-5"
six_digit_string = MissingStdID[:6] # "1?7912"
checksum_digit = int(MissingStdID[7])
for missing_digit in range(0, 10):
candidate = replace_q_mark(six_digit_string, missing_digit)
if checksum(candidate) == checksum_digit: return candidate + MissingStdID[6:]

print(find_q_mark("1?7912-5"))

167912-5

A few notes about the code:

• replace_q_mark() function is called by find_q_mark(). It is only called once, therefore its ac-
tions could have been migrated into find_q_mark(). Efficiency-wise there would be no change. But
seperating it out, making it into a function and naming it accordingly makes the code more readable.
• The first two lines that define find_q_mark() actions are merely dissecting the string into a student-id
and a checkdigit part.

126 Chapter 6. Functions

• What we do in the for loop in find_q_mark() (on Lines 11-13) is coined in Computer Science as
brute force. Practically, it is trying every possible case for being a solution. In this problem, the (?)
mark can be any digit in the range [0,9]. The for loop tries each of these possibilities out until the
computed checksum becomes the same as the checkdigit appearing in the input string.
Example 3: Sequential Search
Searching an item in a list or any other container is a frequently performed operation required while solving
world problems. For the sake of simplicity, let us just focus on situations where only numbers are involved;
i.e. the searched item is a number and the list includes only numbers. Extension to other data types is trivial.
In a search problem, we have a query item (let’s call it x) and a list of items (let’s call it L). If x is a member
of L, then we want the search algorithm to return True, optionally by returning its index in L. If x is not a
member of L, we expect the algorithm to return False.
Consider the example list below:

109 −48 25 4 −13 (6.8.4)

If x is 4, the search algorithm should return True. If it is 5, the answer should be False.
Let us start with the simplest algorithm we can use: Sequential Search. In sequential search, we start with
the beginning of the list, go over each item in the list and check if the query is equal to any item that we go
over. Here is the algorithm:

Algorithm: Sequential Search with While Iterations

----------------------------
Input: x -- the query number
L -- the list of numbers
Output: True (if x is in L) or False (if x is not in L)
----------------------------
Step 1: Create a variable named Index with initial value 0
Step 2: Create a variable named N with value length(L)
Step 3: While Index < N is True, execute Steps 4 and 5:
Step 4: If x is equal to L[Index], then return True as the result
Step 5: Index = Index + 1
Step 6: Return False as the result

Let us implement this simple algorithm in Python:

# Sequential Search with while statement

def seq_search(x, L):
index = 0 # Step 1
N = len(L) # Step 2
while index < N: # Step 3
if x == L[index]: return True # Step 4
index = index + 1 # Step 5
return False # Step 6

# Now let us test the function with examples:

L = [109, -48, 25, 4, -13]
x = 4
print(f"seq_search({x}, {L}): ", seq_search(x, L))
x = 5
print(f"seq_search({x}, {L}): ", seq_search(x, L))
(continues on next page)

6.8. Function Examples 127

(continued from previous page)
x = 5
L = []
print(f"seq_search({x}, {L}): ", seq_search(x, L))

seq_search(4, [109, -48, 25, 4, -13]): True

seq_search(5, [109, -48, 25, 4, -13]): False
seq_search(5, []): False

That looks nice. However, we can have a more compact solution using for loops:

# Sequential Search with for statement

def seq_search(x, L):
for y in L:
if x == y: return True
return False

# Now let us test the function with examples:

L = [109, -48, 25, 4, -13]
x = 4
print(f"seq_search({x}, {L}): ", seq_search(x, L))
x = 5
print(f"seq_search({x}, {L}): ", seq_search(x, L))
x = 5
L = []
print(f"seq_search({x}, {L}): ", seq_search(x, L))

seq_search(4, [109, -48, 25, 4, -13]): True

seq_search(5, [109, -48, 25, 4, -13]): False
seq_search(5, []): False

That’s the beauty of Python. These two are iterative solutions. Let us have a look at a recursive solution.
First, let us have a look at the algorithm. The idea is essentially the same: We compare with the first item
and return True if they match. Otherwise, we continue our search with the remaining items.

Algorithm: Sequential Search with Recursion

----------------------------
Input: x -- the query number
L -- the list of numbers
Output: True (if x is in L) or False (if x is not in L)
----------------------------
Step 1: If L is empty, return False
Step 2: If x is equal to the first item of L (L[0]), return True
Step 3: Make a recursive call on L[1:N], and return its answer

And here is the implementation in Python:

# Sequential Search with recursion

def seq_search(x, L):
if L == []: return False # Step 1: We have exhausted the list, x was not found
if x == L[0]: return True # Step 2: x was at the beginning of the list, yay!
return seq_search(x, L[1:]) # Step 3: Search in the remaining items since x != L[0]

(continues on next page)

128 Chapter 6. Functions

(continued from previous page)
# Now let us test the function with examples:
L = [109, -48, 25, 4, -13]
x = 4
print(f"seq_search({x}, {L}): ", seq_search(x, L))
x = 5
print(f"seq_search({x}, {L}): ", seq_search(x, L))
x = 5
L = []
print(f"seq_search({x}, {L}): ", seq_search(x, L))

seq_search(4, [109, -48, 25, 4, -13]): True

seq_search(5, [109, -48, 25, 4, -13]): False
seq_search(5, []): False

Exercise
Write down the pseudocode for the Python function seq_search with for statement.
Exercise
Extend the recursive implementation of seq_search to search for an item x in a list L in cases where L can
be nested. In other words, the extended search algorithm shoud return True for:
seq_search_nested(5, [10, [4, [5]], -84, [3]])
and False for:
seq_search_nested(8, [10, [4, [5]], -84, [3]])
Example 4: Binary Search
If the list of items (numbers in our case) is not sorted, sequential search is the only option for searching for
an item. If, however, the numbers are sorted (in an increasing or decreasing order), then we have a much
more efficient algorithm called binary search.
Assuming that L, our list of numbers, is sorted in an increasing order: In binary search, we make our first
comparison with the number that is at the center (middle) of the list. If the middle number is equal to x, then
we obviously return True. Otherwise, we compare x with the middle number: If x is smaller, then it has to
be to the left of the middle number – we can continue our search on the left part of L. If, on the other hand,
x is larger than the middle number, it has to be on the right side of the middle number – we can continue our
search on the right part of L. This is illustrated in the following figure.

Fig. 6.8.2: Binary search algorithm.

6.8. Function Examples 129

Here is the pseudocode that describes this algorithm:

Algorithm: Binary Search with Recursion

----------------------------
Input: x -- the query number
L -- the list of numbers
Output: True (if x is in L) or False (if x is not in L)
----------------------------
Step 1: If length(L) is zero then return False since the list is empty
Step 2: Create a variable Mid_index with value length(L)/2
Step 3: If x = L[Mid_Index], return True, we found the item
Step 4: If x < L[Mid_index] then make recursive call with L[0:Mid_index] and return its␣
,→answer
Step 5: Otherwise (x > L[Mid_index]) then make a recursive call with L[Mid_index+1:] and␣
,→return its answer

The following is the implementation in Python:

# Binary search with recursion

def bin_search(x, L):
if len(L) == 0: return False # Step 1: Terminating condition
Mid_index = len(L) // 2 # Step 2: Uses integer division
if x == L[Mid_index]: return True # Step 3
if x < L[Mid_index]: return bin_search(x, L[0:Mid_index]) # Step 4
if x > L[Mid_index]: return bin_search(x, L[Mid_index+1:]) # Step 5: Can drop if statement

# Now let us test the function with examples:

L = [-48, -13, 4, 25, 109]
x = 4
print(f"bin_search({x}, {L}): ", bin_search(x, L))
x = 5
print(f"bin_search({x}, {L}): ", bin_search(x, L))
x = 5
L = []
print(f"bin_search({x}, {L}): ", bin_search(x, L))

bin_search(4, [-48, -13, 4, 25, 109]): True

bin_search(5, [-48, -13, 4, 25, 109]): False
bin_search(5, []): False

Exercise
Implement binary search using while statements.
Exercise
Identify the complexity of sequential search and binary search in the following cases:
• Best case: When the item that we are making our first comparison with is the one we were looking
for.
• Worst case: The item we are looking for is the last item with which we performed the comparison.
• Average case.
Example 5: Sorting

130 Chapter 6. Functions

Sorting pertains to ordering a list of items (L) so that each item in L satisfies the following constraint:

Li < Li+1 , (6.8.5)

which leads to an ascending (increasing) order of items in L and this is called ascending sort. If the order
of the constraint is changed to Li > Li+1 , the outcome is a descending (decreasing) order of items in L and
this is called descending sort.
An algorithm that can perform ascending sort can easily be changed to descending sort; therefore, we will
just focus on one of them, namely, ascending sort. Moreover, we will focus on sorting a list of numbers;
however, the same algorithms can be applied on other data types as long as an ordering constraint (< or >)
can be defined.
Bubble Sort
Let us have a look at a very simple though inefficient sorting algorithm, called Bubble Sort. In Bubble Sort,
we start from the beginning of the list and check whether the constraint Li < Li+1 is violated for an i
value. If it is, we swap Li with Li+1 . When we hit the end of the list, we start from the beginning of the
list again and continue checking the constraint Li < Li+1 . Once at the end of the list, we again go back to
the beginning and continue checking the constraint for each i. This continues until no i value violates the
constraint Li < Li+1 , which means that the numbers are sorted.
Therefore, in Bubble Sort, we have two loops, one outer loop that continues until all numbers are sorted and
an inner loop that goes through the list and checks the constraint. Here is a pseudo-code for this:

Algorithm: Bubble Sort

----------------------------
Input: L -- the list of numbers (not sorted)
Output: L -- the sorted list of numbers (ascending order)
----------------------------
Step 1: Create a varible Is_sorted with value False
Step 2: Create a variable N with value length(L)
Step 3: Create a variable Index
Step 4: While Is_sorted is False, repeat Steps 5-11:
Step 5: Set Is_sorted to True
Step 6: Set Index to zero
Step 7: While Index < (N-1), repeat Steps 8-11
Step 8: If L[Index] > L[Index+1], then execute Steps 9 & 10:
Step 9: Swap L[Index] and L[Index+1]
Step 10: Set Is_sorted to False
Step 11: Index = Index + 1
Step 12: return L

One outer iteration of the algorithm is illustrated on a small list of numbers below:

6.8. Function Examples 131

Fig. 6.8.3: One outer iteration of the Bubble Sort algorithm.
132 Chapter 6. Functions
With one outer iteration, the list appears to be one step closer to a sorted list. In the example above, the
largest number has found its correct place after one outer iteration. With a second one, the second largest
number will find its place. With a third iteration, the third largest number will find its place etc.
And here is the implementation in Python:

# Bubble Sort

def bubble_sort(L):
Is_sorted = False # Step 1
N = len(L) # Step 2
Index = 0 # Step 3: Can be removed in Python

while Is_sorted is False: # Step 4

Is_sorted = True # Step 5
Index = 0 # Step 6
while Index < (N-1): # Step 7
if L[Index] > L[Index+1]: # Step 8
(L[Index], L[Index+1]) = (L[Index+1], L[Index]) # Step 9
Is_sorted = False # Step 10
Index = Index + 1 # Step 11
return L # Step 12

# Let us test our implementation with some sample lists

L = [109, -13, 4, -48, 25]
print(f"bubble_sort({L}): ", bubble_sort(L))

L = [-48, -13, 4, 25, 109]

print(f"bubble_sort({L}): ", bubble_sort(L))

L = []
print(f"bubble_sort({L}): ", bubble_sort(L))

bubble_sort([109, -13, 4, -48, 25]): [-48, -13, 4, 25, 109]

bubble_sort([-48, -13, 4, 25, 109]): [-48, -13, 4, 25, 109]
bubble_sort([]): []

6.9 Programming Style

With the topics covered in this chapter, we started working with longer and more elaborated Python codes.
Up to now, we have introduced crucial aspects that can be arranged differently among programmers:
• Naming variables and functions.
• Indentation.
• Function definition.
• Commenting.
Longing for brevity and readability, the creators of Python have defined some guidelines of style for those
interested: Python Style Guide25 .
25
https://www.python.org/dev/peps/pep-0008/

6.9. Programming Style 133

There are alternative styles and we strongly recommend you to choose one and stick to it as a programming
style.

6.10 Important Concepts

We would like our readers to have grasped the following crucial concepts and keywords from this chapter
(all related to Python):
• Defining functions.
• Function parameters and how to pass values to functions.
• Default parameters.
• Scopes of variables.
• Local, enclosing, global and built-in scopes.
• Higher-order functions.
• Differences of functions between programming and Mathematics.
• Recursion.

6.11 Further Reading

• Functional tools of Python: https://docs.python.org/3/howto/functional.html

• A more comprehensive coverage on recursion: https://link.springer.com/chapter/10.1007/
978-3-7091-1343-1_4
• Python style guide: https://www.python.org/dev/peps/pep-0008/

6.12 Exercises

• Write a Python function named only_numbers() that removes items in a list that are not numbers. E.g.
only_numbers([10, "ali", [20], True, 4]) should return [10, 4]. Implement two versions:
One using iteration and another using recursion.
• Write a Python function named flatten() that removes nesting in a nested list and lists all elements
in a single non-nested list. E.g. flatten([1, [4, 5, [6]], [["test"]]]) should return [1, 4,
5, 6, "test"].
• Write a Python function named reverse() that reverses the order of elements in a list. E.g.
reverse([10, 5, 0]) should return [0, 5, 10]. Implement two versions: One using iteration
and another using recursion.
• Write a Python function named greatest_divisor() that finds the greatest divisor of an integer. The
greatest divisor of an integer n is an integer x < n that satisfies n = k · x for some integer k. E.g.
greatest_divisor(12) should return 6.

134 Chapter 6. Functions

7 | A Gentle Introduction to Object-
Oriented Programming

(C) Copyright Notice: This chapter is part of the book available athttps://pp4e-book.github.io/and copying,
distributing, modifying it requires explicit permission from the authors. See the book page for details:https:
//pp4e-book.github.io/
At this stage of programming, you must have realized that programming possesses some idiosyncrasies:
• Unless a randomization is explicitly build in, all computations are deterministic; i.e., the result is
always the same for the same input.
• The logic involved is binary; i.e., two truth values exist: True and False.
• There is a clear distinction of actions and data. Actions are coded into expressions, statements and
functions. Data is coded into integers, floating points and containers (strings, tuples and lists).
The first two can be dealt with in a controlled manner. Furthermore, mostly it is very much preferred to have
a crisp and deterministic outcome. But, the third is not so natural. The world we live in is not made of data
and independent actions acting on that data. This is merely an abstraction. The need for this abstraction stems
from the very nature of the computing device mankind has manufactured, the von Neumann Machine. What
you store in the memory is either some data (integer or floating point) or some instruction. The processor
processes the data based on a series of instructions. Therefore, we have a clear separation of data and action
in computers.
But when we look around, we don’t see such a distinction. We see objects. We see a tree, a house, a table,
a computer, a notebook, a pencil, a lecturer and a student. Objects have some properties which would be
quantified as data, but they also have some capabilities that would correspond to some actions. What about
reuniting data and action under the natural concept of “object”? Object Oriented Programming, abbreviated
as OOP, is the answer to this question.

7.1 Properties of Object-Oriented Programming

OOP is a paradigm that comes with some properties:

• Encapsulation: Combining data and functions that manipulate that data under a concept that we name
as ‘object’ so that a rule of “need-to-know” and “maximal-privacy” is satisfied.
• Inheritance: Defining an object and then using it to create “descendant” objects so that the descendant
objects inherit all functions and data of their ancestors.
• Polymorphism: A mechanism allowing a descendant object to appear and function like its ancestor
object when necessary.

135
7.1.1 Encapsulation

Encapsulation is the property that data and actions are glued together in a data-action structure called ‘object’
that conforms to a rule of “need-to-know” and “maximal-privacy”. In other words, an object should provide
access only to data and actions that are needed by other objects and other data & actions that are not needed
should be hidden and used by the object itself for its own merit.
This is important especially to keep implementation modular and manageable: An object stores some data
and implements certain actions. Some of these are private and hidden from other objects whereas others are
public to other objects so that they can access such public data and actions to suit their needs.
The public data and actions function as the interface of the object to the outside world. In this way, objects
interact with each other’s interfaces by accessing public data and actions. This can be considered as a message
passing mechanism: Object1 calls Object2’s action f, which calls Object3’s function g, which returns a
message (a value) back to Object2, which, after some calculation returns another value to Object 1. As
a realistic example of a university registration system, assume Student object calls register action of a
Course object and it calls checkPrerequisite action of a Curriculum object. checkPrerequisite checks
if course can be taken by student and returns the result. register action does additional controls and returns
the success status of the registration to the Student.
In this modular approach, Object1 does not need to know how Object2 implements its actions or how it
stores data. All Object1 needs to know is the public interface via which ‘messages’ are passed to compute
the solution.
Assume you need to implement a simple weather forecasting system. This hypothetical system gets a set of
meteorological sensor data like humidity, pressure, temperature from various levels of atmosphere and try
to estimate the weather conditions for the next couple of days. The data of such a system may have a time
series of sensor values. The actions of such system would be a group of functions adding sensor data as they
are measured and forecasting functions for getting the future estimate of weather conditions. For example:

sensors = [{'datetime':'20201011 10:00','temperature':12.3,

'humidity': 32.2, 'pressure':1.2,
'altitute':1010.0},
{'datetime':'20201011 12:00','temperature':14.2,
'humidity': 31.2, 'pressure':1.22,
'altitute':1010.0},
....]

def addSensorData(sensorarr, temp, hum, press, alt):

'''Add sensor data to sensor array with
current time and date'''
....

def estimate(sensorarr, offset):

'''return whether forecast for given
offset days in future'''
...
...
addSensorData(sensors, 20.3, 15.4, 0.82, 10000)
...
print(estimate(sensors, 1))
...

In the implementation above, the data and actions are separated. The programmer should maintain the list
containing the data and make sure actions are available and called as needed with the correct data. This

136 Chapter 7. A Gentle Introduction to Object-Oriented Programming

approach has a couple of disadvantages:
1. There is no way to make sure actions are called with the correct sensor data format and values (i.e. es-
timate('Hello world', 1,1,1,1)).
2. addSensorData can make sure that sensor list contains correct data, however, since sensors data can
be directly modified, its integrity can be violated later on (i.e. sensors[0]='Hello World').
3. When you need to have forecast of more than one location, you need to duplicate all data and maintain
them separately. Keeping track of which list contains which location requires extra special care.
4. When you need to improve your code and change data representation like storing each sensor type
on a separate sorted list by time, you need to change the action functions. However, if some code
directly accesses the sensor data, it may conflict with the changes you made on data representation.
For example, if the new data representation is as follows:

sensors = {'temperature': [('202010101000',23),...],

'humidity': [('2020101000',45.3),...],
'pressure': [('2020100243',1.02),...]}

Any access to sensors[0] as a dictionary directly by code segments will be incorrect.

With encapsulation, sensor data and actions are put into the same body of definition so that the only way to
interact with the data would be through the actions. In this way:
1. Data and actions maintained together. Encapsulation mechanism guarantees that data exists and it has
correct format and values.
2. Multiple instances can be created for forecasting for multiple locations, and each location is maintained
in its object as if it was a simple variable.
3. Since no code part accesses the data directly but calls the actions, changing internal representation and
implementation of functions will not cause any problem.
The following is an example OOP implementation for the problem at hand:

class WhetherForecast:
# Data
__sensors = None

# Actions acting on the data

def __init__(self):
self.__sensors = [] # this will create initial sensor data

def addSensorData(self, temp, hum, press, alt):

....

def estimate(self, offset):

...
return {'lowest':elow, 'highest':ehigh,...}

ankara = WhetherForecast() # Create an instance for location Ankara

ankara.addSensorData(...)
....

izmir = WhetherForecast() # Create an instance for location Izmir

(continues on next page)

7.1. Properties of Object-Oriented Programming 137

(continued from previous page)
izmir.addSensordata(...)
....

print(ankara.estimate(1)) # Work with the data for Ankara

print(izmir.estimate(2)) # Work with the data for Izmir

The above syntax will be more clear in the following sections; however, please note how the newly created
objects ankara and izmir behave. They contain their sensor data internally, and the programmer does not
need to care about their internals. The resulting object will syntactically behave like a built-in data type of
Python.

7.1.2 Inheritance

In many applications, the objects we are going to work with are going to be related. For example, in a
drawing program, we are going to work with shapes such as rectangles, circles, triangles which have some
common data and actions, e.g.:
• Data:
– Position
– Area
– Color
– Circumference
• and actions:
– draw()
– move()
– rotate()
What kind of data structure we use for these data and how we implement the actions are important. For
example, if one shape is using Cartesian coordinates (x, y) for position and another is using Polar coordinates
(r, θ), a programmer can easily make a mistake by providing (x, y) to a shape using Polar coordinates.
As for actions, implementing such overlapping actions in each shape from scratch is redundant and ineffi-
cient. In the case of separate implementations of overlapping actions in each shape, we would have to update
all overlapping actions if we want to correct an error in our implementation or switch to a more efficient al-
gorithm for the overlapping actions. Therefore, it makes sense to implement the common functionalities in
another object and reuse them whenever needed.
These two issues are handled in OOP via inheritance. We place common data and functionalities into an
ancestor object (e.g. Shape object for our example) and other objects (Rectangle, Triangle, Circle) can
inherit (reuse) these data and definitions in their definitions as if those data and actions were defined in their
object definitions.
In real life entities, you can observe many similar relations. For example:
• A Student is a Person and an Instructor is a Person. Updating personal records of a Student is
no different than that of an Instructor.

138 Chapter 7. A Gentle Introduction to Object-Oriented Programming

• An DCEngine, a DieselEngine, and a StreamEngine are all Engines. They have the same charac-
teristic features like horse power, torque etc. However, DCEngine has power consumption in units of
Watt whereas DieselEngine consumption can be measured as litres per km.
• In a transportation problem, a Ship, a Cargo_Plane and a Truck are all Vehicles. They have the
same behaviour of carrying a load; however, they have different capacities, speeds, costs and ranges.
Assume we like to improve the forecasting accuracy through adding radar information in our Whether-
Forecast example above. We need to get our traditional estimate and combine it with the radar image data.
Instead of duplicating the traditional estimator, it is wiser to use existing implementation and extend its func-
tionality with the newly introduced features. This way, we avoid code duplication and when we improve our
traditional estimator, our new estimator will automatically use it.
Inheritance is a very useful and important concept in OOP. Together with encapsulation, it improves reusabil-
ity, maintenance, and reduces redundancy.

7.1.3 Polymorphism

Polymorphism is a property that enables a programmer to write functions that can operate on different data
types uniformly. For example, calculating the sum of elements of a list is actually the same for a list of
integers, a list of floats and a list of complex numbers. As long as the addition operation is defined among
the members of the list, the summation operation would be the same. If we can implement a polymorphic
sum function, it will be able to calculate the summation of distinct datatypes, hence it will be polymorphic.
In OOP, all descendants of a parent object can act as objects of more than one types. Consider our example
on shapes above: The Rectangle object that inherits from the Shape object can also be used as a Shape
object since it bears data and actions defined in a Shape object. In other words, a Rectangle object can
be assumed to have two data types: Rectangle and Shape. We can exploit this for writing polymorphic
functions. If we write functions or classes that operate on Shape with well-defined actions, they can operate
on all descendants of it including, Rectangle, Circle, and all objects inheriting Shape. Similarly, actions
of a parent object can operate on all its descendants if it uses a well-defined interface.
Polymorphism improves modularity, code reusability and expandability of a program.

7.2 Basic OOP in Python

The way Python implements OOP is not to the full extent in terms of the properties listed in the previous
section. Encapsulation, for example, is not implemented strongly. But inheritance and polymorphism are
there. Also, operator overloading, a feature that is much demanded in OOP, is present.
In the last decade, Python started to become a standard for Science and Engineering computation. For various
computational purposes, software packages were already there. Packages to do numerical computations,
statistical computations, symbolic computations, computational chemistry, computational physics, all sorts
of simulations were developed over four decades. Now many such packages, free or proprietary, are wrapped
to be called through Python. This packaging is done mostly in an OOP manner. Therefore, it is vital to know
some basics of OOP in Python.

7.2. Basic OOP in Python 139

7.2.1 The Class Syntax

In Python, an object is a code structure that is like in Fig. 7.2.1:

Fig. 7.2.1: An object includes both data and actions (methods and special methods) as one data item.

First, a piece of jargon:

• Class: A prescription that defines a particular object. The blueprint of an object.
• Class Instance$;:raw-latex:equiv; $Object: A computational structure that has functions and data
fields built according the blueprint, namely the class. Similar to the construction of buildings accord-
ing to an architectural blueprint, in Python we can create objects (more than one) conforming to a
class definition. Each of these objects will have their own data space and in some cases customized
functions. Objects are equivalently called Class instances. Each object provides the following:
– Methods: Functions that belong to the object.
– Sending a message to an object: Calling a method of the object.
– Member: Any data or method that is defined in the class.
So, as you would guess, we start with a structural plan, using the jargon, the ‘class definition’. In Python
this is done by the keyword class:

class ClassN ame :

Statement block

Here is an example:

class shape:
color = None
x = None
y = None

def set_color(self, red, green, blue):

self.color = (red, green, blue)

def move_to(self, x, y):

self.x = x
self.y = y

140 Chapter 7. A Gentle Introduction to Object-Oriented Programming

This blueprint tells Python that:
1. The name of this class is shape.
2. Any object that will be created according to this blueprint has three data fields, named color, x and y.
At the moment of creation, these fields are set to None (a special value of Python indicating that there
is a variable here but no value is assigned yet).
3. Two member functions, the so-called methods, are defined: set_color and move_to. The first takes
four arguments, constructs a tuple of the last three values and stores it into the color data field of the
object. The second, move_to, takes three arguments and assign the last two of them to the x and y
data_fields, respectively.
The peculiar keyword self in the blueprint refers to the particular instance (when an object is created based
on this blueprint). The first argument to all methods (the member functions) have to be coded as self. That
is a rule. The Python system will fill it out when that function is activated.
To refer to any function or any data field of an object, we use the (.) dot notation. Inside the class definition,
it is self.� . Outside of the object, the object is certainly stored somewhere (a variable or a container).
The way (syntax) to access the stored object is followed. Then, this syntax is appended by the (.) dot which
is then followed by the data field name or the method name.
For our example shape class, let us create two objects and assign them to two global variables p and s,
respectively:

p = shape()
s = shape()
p.move_to(22, 55)
p.set_color(255, 0, 0)
s.move_to(49, 71)
s.set_color(0, 127, 0)

The object creation is triggered by calling the class name as if it is a function (i.e. shape()). This creates an
instance of the class. Each instance has its private data space. In the example, two shape objects are created
and stored in the variables p and s. As said, the object stored in p has its private data space and so does s.
We can verify this by:

print(p.x, p.y)
print(s.x, s.y)

When a class is defined, there are a bunch of methods, which are automatically created, and they serve the
integration of the object with the Python language. For example, what if we issue a print statement on the
object? What will

print(s)

print?
These default methods can be overwritten (redefined). Let us do it for two of them: __str__ is the method
that is automatically activated when a print function has an object to be printed. The built-in print function
sends to the object an __str__ message (that was the OOP jargon, i.e. calls the __str__ member function
(method)). All objects, when created, have a some special methods predefined. Many of them are out of the
scope of this course, but __str__ and __init__ are among these special methods.
It is possible that the programmer, in the class definition, overwrites (redefines) these predefinitions.

7.2. Basic OOP in Python 141

__str__ is set to a default definition so that when an object is printed such an internal location informa-
tion is printed:

<main.shape object at 0x7f295325a6a0>

Not very informative, is it? We will overwrite this function to output the color and coordinate information,
which will look like:

shape object: color=(0,127,0) coordinates=(47,71)

The second special method that we will overwrite is the __init__ method. __init__ is the method that
is automatically activated when the object is first created. As default, it will do nothing, but can also be
overwritten. Observe the following statement in the code above:

s = shape()

The object creation is triggered by calling the class name as if it is a function. Python (and many other OOP
languages) adopt this syntax for object creation. What is done is that the arguments passed to the class name
is sent ‘internally’ to the special member function __init__. We will overwrite it to take two arguments at
object creation, and these arguments will become the initial values for the x and y coordinates.
Now, let us switch to the real interpreter and give it a go:

class shape:
color = None
x = None
y = None

def set_color(self, red, green, blue):

self.color = (red, green, blue)

def move_to(self, x, y):

self.x = x
self.y = y

def __str__(self):
return "shape object: color=%s coordinates=%s" % (self.color, (self.x,self.y))

def init(self, x, y):

self.x = x
self.y = y

def lt(self, other):

return self.x + self.y < other.x + other.y

p = shape(22,55)
s = shape(12,124)
p.set_color(255,0,0)
s.set_color(0,127,0)

print(s)
s.move_to(49,71)
print(s)

(continues on next page)

142 Chapter 7. A Gentle Introduction to Object-Oriented Programming

(continued from previous page)
print(p.__lt__(s))
print(p < s) # just the same as above but now infix

print(s.__dir__())

shape object: color=(0, 127, 0) coordinates=(12, 124)

shape object: color=(0, 127, 0) coordinates=(49, 71)
True
True
['x', 'y', 'color', '__module__', 'set_color', 'move_to', '__str__', '__init__', '__lt__', '_
,→_dict__', '__weakref__', '__doc__', '__repr__', '__hash__', '__getattribute__', '__setattr_
,→_', '__delattr__', '__le__', '__eq__', '__ne__', '__gt__', '__ge__', '__new__', '__reduce_
,→ex__', '__reduce__', '__subclasshook__', '__init_subclass__', '__format__', '__sizeof__',
,→'__dir__', '__class__']

7.2.2 Special Methods/Operator Overloading

There are many more special methods than the ones we described above. For a complete reference, we refer
you to “Section 3.3 of the Python Language Reference”26 .
Last, but not least, a special method to mention is the magnitude comparison for an object. In other words,
if you have an object, how will it behave under comparison? For example, the following rich comparisons
are possible:
• x<y calls x.__lt__(y),
• x<=y calls x.__le__(y),
• x==y calls x.__eq__(y),
• x!=y calls x.__ne__(y),
• x>y calls x.__gt__(y), and
• x>=y calls x.__ge__(y).
Having learned this, please copy-and-paste the following definition to the class definition of shape above
in to the code box. Now, you have the (<) comparison operator available.

def lt(self, other):

return self.x + self.y < other.x + other.y

As you can see, the comparison result is based on the Manhattan distance from the origin. You can give it a
test right away and compare the two objects s and p as follows:

print(s<p)

How would you modify the comparison method so that it compares the Euclidean distances from the origin?
As we have created our first object in Python, we strongly advise that the ‘encapsulation’ property of OOP
is followed by the programmer, i.e. you. This property is that the data of an object is private to the object
and not open to modification (or not even to inspection if ‘encapsulation’ is taken extremely strictly) by a
26
https://docs.python.org/3/reference/

7.2. Basic OOP in Python 143

piece of code external to the object. Only member functions, the so-called methods, of that object can do
this. Therefore, if you want to update the color value, for example, of an shape object, though you can do
it ‘brutally’ by e.g.:

p.color = (127,64,5)

However, you should not to do so. In contrary to some other OOP programming languages, Python does not
forbid this by a brutal force. Therefore, it is not a you cannot but a you should not type of action. All data
access should be done through messages (functions).

7.2.3 Example 1: Counter

Now, let us implement a very simple object called Counter. Counter has two important restrictions: it starts
with zero and it can only be incremented. If you use a simple integer variable instead of Counter, its value
can be initialized to any value and it can directly be assigned to an arbitrary value. Implementing a Python
class for a Counter will let you enforce those restrictions since you define the initialization and the actions.

class Counter:
def __init__(self):
self.value = 0 # this is the initialization

def increment(self):
'''increment the inner counter'''
self.value += 1

def get(self):
'''return the counter as value'''
return self.value

def __str__(self):
'''define how your counter is displayed'''
return 'Counter:{}'.format(self.value)

stcnt = Counter() # create the counter

stcnt.increment()
stcnt.increment()
print("# of students", stcnt)

sheep = Counter()
while sheep.get() < 1000:
sheep.increment()

print("# of sheep",sheep)

# of students Counter:2
# of sheep Counter:1000

144 Chapter 7. A Gentle Introduction to Object-Oriented Programming

7.2.4 Example 2: Rational Number

As our next example, let us try to define a new data type, Rational Number, as a Python class. Representing
a rational number simply as a tuple of integers as (numerator, denominator) is possible. However, this
representation has a couple of issues. First, the denominator can be 0, which leads to an invalid value.
Second, many distinct tuples represent practically the same value and they need to be interpreted in a special
way in operators like comparison. We need to either normalize all into their simplest form or implement
comparison operators to respect the values they represent. In the following implementation, we choose the
first approach: We find greatest common divisor of numerator and denominator and normalize them.

import math

class Rational:
def __init__(self, n, d):
'''initialize from numerator and denominator values'''
if d == 0:
raise ZeroDivisionError # raise an error. Explained in following chapters
self.num = n
self.den = d
self.simplify()

def simplify(self):
if self.num == 0:
self.den = 1
return
gcd = math.gcd(self.num,self.den)
if self.den < 0: # flip their signs if denominator is negative
self.den *= -1
self.num *= -1
self.num = self.num // gcd
self.den = self.den // gcd # normalize them by dividing by greatest common divisor

def __str__(self):
return '{}/{}'.format(self.num, self.den)

def __mul__(self, rhs): # this special method is called when * operator is used
''' (a/b)*(c/d) -> a*c/b*d in a new object '''
retval = Rational(self.num * rhs.num, self.den * rhs.den) # create a new object
return retval

def __add__(self, rhs): # this special method is called when * operator is used
''' (a/b)+(c/d) -> a*d+b*c/d*b in a new object '''
retval = Rational(self.num * rhs.den + rhs.num * self.den,
self.den * rhs.den) # create a new object with sum
return retval

# -, /, and other operators left as exercise

def eq(self, rhs): # called when == operator is used

'''a*d == b*c '''
return self.num*rhs.den == self.den*rhs.num

def lt(self, rhs): # called when < operator is used

'''a*d < b*c '''
return self.num*rhs.den < self.den*rhs.num
(continues on next page)

7.2. Basic OOP in Python 145

(continued from previous page)

# rest can be defined in terms of the first two

def __ne__(self, rhs): return not self == rhs

def le(self, rhs): return self < rhs or self == rhs

def gt(self, rhs): return not self <= rhs

def ge(self, rhs): return not self < rhs

# Let us play with our Rational class

a = Rational(3, 9)
b = Rational(16, 24)
print(a, b, a*b+b*a)
print(a<b, a+b == Rational(1, 1))

1/3 2/3 4/9

True True

This class definition only implements *, +, and comparison operators. The remaining operators are left
as an exercise. Our new class Rational behaves like a built-in data type in Python, thanks to the special
methods implemented. User-defined data types implementing integrity restrictions through encapsulation
are called Abstract Data Types.

7.2.5 Inheritance with Python

Now let’s have a look on the second property on OOP, namely ‘inheritance’, in Python. We will do that by
extending our shape example.
A shape is relatively a general term. We have many types of shapes: Triangles, circles, ovals, rectangles,
even polygons. If we incorporated them into a software, for example, a simple drawing and painting appli-
cation for children, each of them would have a position on the screen and a color. This would be common
to all shapes. But then, a circle would be identified (additionally) by a radius; a triangle by two additional
corner coordinates etc.
Therefore, a triangle object should inherit all the properties and member functions of a shape object plus
some other.
Starting to define a triangle object, for example, with all properties and member functions copied from
shape is done by a first line:

class triangle(shape):

following the general syntax below:

class ClassN ame ( BaseClass1 , BaseClass2 , .., ):

Statement block

Though multiple inheritance (inheriting form more than one class) is possible, it is seldom used and is not

146 Chapter 7. A Gentle Introduction to Object-Oriented Programming

preferred. It has elaborated rules for resolving member function name clashes, which is beyond the scope of
an introductory book. We will continue with an example that does inheritance from a single base class.

7.2.6 Interactive Example: A Simple Shape Drawing Program

Below, you see a simple object-oriented Python code that draws a red circle and a green rectangle, and then
moves the green rectangle towards the circe. When you hover your mouse over the code below, a popup
window will explain that line of code.
Following this interactive code display, you can find a code box, the one with a small triangle on the left,
which is the same code. So, you can run it, and observe the code code live, running. In the exercise part,
you will be asked to modify the code.
The dark brownish sections in the code contain calls to a drawing library (named calysto) which is also
object oriented but has nothing to do with you. Not to distract the reader intentionally, we have darkened it
out.
Our object-oriented programming (OOP) intention in this example is to define a general class that we will
name shape. shape contains all the properties and functions that a drawable geometric object (circle, rect-
angle, triangle, …) possesses. All of them are drawn on an imaginary canvas at a certain coordinate with
some color. So, all shapes have a coordinate at which they are drawn and a color. The first lines of the class
shape definition starts with these variables.
[You can get additional information by hovering your mouse over any piece of code.]
In this example, we implement just two geometric shapes, the circle and the rectangle. This piece of code
with the green background is the definition of the circle class. It is derived from the base class shape.
So, it inherits all the definitions of shape (but these definitions can be overwritten, in other words, can be
redefined). As you observe, it defines a radius variable (in addition to the color, x, y variables that were
inherited form the base class). Also, it defines, for the first time, a draw member function which has the duty
to draw a circle on the canvas, based on the parameters x, y, radius and color using some functions of
the library calysto. The details of this drawing implementation is not the reader’s concern (therefore it is
darkened out).
[You can get additional information by hovering your mouse over any piece of code.]
The second geometric shape, which is implemented, is the rectangle. The orange backgrounded code defines
this class. Again, it is a descendent of the base class shape. Hence, it inherits all the definitions of shape
(just as circle did). This time, it does not add a radius but a width and a height variables to the class. In
addition to the redefinition of the initializer __init__, which is activated the moment an instance is created,
this class has its own definition of the draw member function.
[You can get additional information by hovering your mouse over any piece of code.]
Now, for an instance, return to the class shape definition (the yellow box). There, you will see a displace
member function being defined. It is a function that displaces a geometric shape by an amount of (∆x, ∆y).
Please inspect the function now. You will discover that the function:
• memorizes the color of the shape,
• changes the color to white,
• calls the draw member function and draws a white shape exactly on top of the old one,
• restores the drawing color (from white to the memorized color),
• changes the x, y values by amounts delta_x and delta_y, respectively,

7.2. Basic OOP in Python 147

• calls the draw member function and draws at the new x,y coordinates the shape with the original color.
Now we have an interesting phenomenon. draw is even not defined in shape. It will be defined differently
for each class that is derived from the shape base class. How can a function of shape make a reference to a
function which does not exist for shape and will only be defined in classes that are descendants of this base
class? Also note that the each of the descendant classes (circle and rectangle) are going to define it (the
draw function) their own way. This is called forward-referencing and polymorphism, a nice and powerful
feature of OOP.
Note that the class definitions (class shape, class circle, class rectangle) are merely blueprints.
They do not create any data. They define how they will be created and which functions will act how. The
actual creation (based on these blueprints) are done in the code in the last (turquoise colored) box.
In summary, the code below performs the following:
• An instance of the circle class is created (by calling its initializer) at coordinates (100,100) with a
radius of 50. This instance is stored into the variable mycircle.
• The color of this freshly created object is changed to the brightest red.
• The object’s draw function is called (in OOP jargon: a draw message is sent to the object stored in
mycircle)
• Similarly, an instance of rectangle class is created at coordinates (200,100) and with width=50,
height=80. This instance is stored into the variable myrectangle.
• The color of the rectangle is changed to the brightest green.
• The rectangles draw function is called.
• The program pauses (waits) for 1 second.
• Finally, it calls the displace function of the rectangle. Though it was defined in the base class, dis-
place locates the correct draw (the draw of the rectangle) and performs the job.
Please use the Colab link at the top to see the interactive demo.

7.2.7 Useful Short Notes on Python’s OOP

These notes are provided for completeness and a possible need in your further studies. These are out of the
introductory scope of the book.
• It is possible that the derived class overrides the base class’s member function. But, still want to access
the (former) definition in the base class. One can access that definition by prefixing the function call
by super(). (no space after the dot).
• There is no proper destructor in Python. This is because the Python engine does all the memory
allocation and book-keeping. Though entirely under the control of Python, sometimes the so-called
Garbage Collection is carried out. At that moment, all unused object instances are wiped out of the
memory. Before that, a special member function __del__ is called. If you want to do a special
treatment of an object before it is wiped forever, you can define the __del__ function. The concept of
garbage collection is complex and it is wrong to assume that __del__ will right away be called even
if an object instance is deleted by the del statement (del will mark the object as unused but it will not
necessarily trigger a garbage collection phase).
• Infix operators have special, associated member functions (those that start and end with double under-
scores). If you want your objects to participate in infix expressions, then you have to define those. For

148 Chapter 7. A Gentle Introduction to Object-Oriented Programming

example ‘+’ has the associated special member function __add__. For a complete list and how-to-do’s,
search for “special functions of Python” and “operator overloading”.
• You can restrict the accessibility of variables defined in a class. There are three types of accessibility
modifications you can perform: public, private, and protected. The default is public.
– Public access variables can be accessed anywhere inside or outside the class.
– Private variables can only be accessed inside the class definitions. A variable which starts with
two underscores is recognized by programmers as private.
– Protected variables can be accessed within the same package (file). A variable which starts with
a single underscore is recognized by programmers as protected.

7.3 Widely-used Member Functions of Containers

Being acquainted with the OOP concepts, it is time to reveal the ‘object’ properties of some Python compo-
nents. In Python, every data is an object. However, here we will look at the OOP components of containers.
Strings
Assume S is a string. In the table below, you will find some of the very frequently used member functions of
strings (in the Operation column, anything in square brackets denotes that the content is optional – if you
enter the optional content, do no type in the square brackets):

7.3. Widely-used Member Functions of Containers 149

Operation Result
S.capitalize() Returns a copy of S with its first character capitalized, and the rest of the
characters lowercased.
S .count(sub [, start Returns the number of occurrences of substring sub in string S.
[, end]])
`` S.find(sub [, start [, Returns the lowest index in S where substring sub is found. Returns -1 if sub
end]])`` is not found.
S.isalnum() Returns True if all characters in S are alphanumeric, False otherwise.
S.isalpha() Returns True if all characters in S are alphabetic, False otherwise.
S.isdigit() Returns True if all characters in S are digit characters, False otherwise.
S.islower() Returns True if all characters in S are lowercase, False otherwise.
S.isspace() Returns True if all characters in S are whitespace characters, False other-
wise.
S.isupper() Returns True if all characters in S are uppercase, False otherwise.
separator.join(seq) Returns a concatenation of the strings in the sequence seq, separated by string
separator, e.g.: "#".join(["a","bb","ccc"]) returns "a#bb#ccc"
S.ljust/rjust/c Returns S, left/right justified/centered in a string of length width. surrounded
enter(width [, by the appropriate number of fillChar characters.
fillChar=' '])
S.lower() Returns a copy of S converted to lowercase.
S.lstrip([chars]) Returns a copy of S with leading chars (default: blank chars) removed.
S.partition(separ) Searches for the separator separ in S, and returns a tuple (head, sep, tail)
containing the part before it, the separator itself, and the part after it.
S.replac e(old, new Returns a copy of S with the first maxCount (-1: unlimited) occurrences of
[, maxCount = -1]) substring old replaced by new.
S.sp lit([separator Returns a list of the words in S, using separator as the delimiter string.
[, maxsplit]])
S. Returns a list of the lines in S, breaking at line boundaries.
splitlines([keepends])
S.startsw ith(prefix Returns True if S starts with the specified prefix, otherwise returns False.
[, start [, end]]) Negative numbers may be used for start and end. Prefix can also be a tuple
of strings to try.
S.strip([chars]) Returns a copy of S with leading and trailing chars (default: blank chars)
removed.
S.swapcase() Returns a copy of S with uppercase characters converted to lowercase and
vice versa.
S.upper() Returns a copy of S converted to uppercase.

Lists
Assume L is a list. In the table below, you will find some of the very frequently used member functions of
lists (in the Operation column, anything in square brackets denotes that the content is optional – if you enter
the optional content, do no type in the square brackets):

150 Chapter 7. A Gentle Introduction to Object-Oriented Programming

Operation Result
L.append(x) same as L[len(L) : len(L)] = [x]
L.extend(x) same as L[len(L) : len(L)] = x
L.count(x) returns number of i’s for which L[i] == x
` L.index(x [, start [, stop ]])` returns smallest i such that L[i] == x. start and
stop limit search to only part of the list.
L.insert(i, x) same as L[i:i] = [x] if i_� 0. if i � -1,
inserts before the last element.
L.remove(x) same as del L[L.index(x)]
L.pop([i]) same as x = L[i]; del L[i]; return x
L.reverse() reverses the items of L in place
L.sort([cmp]) OR L.sort([ cmp=cmpFct ] [, sorts the items of L in place
key =keyGetter ] [, reverse=bool ])

Dictionaries
Assume D is a dictionary. In the table below, you will find some of the very frequently used member functions
of dictionaries (in the Operation column anything in square brackets denotes that the content is optional –
if you enter the optional content, do no type in the square brackets):

Operation Result
D.from Class method to create a dictionary with keys provided by iterator, and all
keys(iterable, values set to value
value=None)
D.clear() Removes all items from D
D.copy() A shallow copy of D
D.has_key(k) OR k in True if D has key k, else False
D
D.items() A copy of D’s list of (key, item) pairs
D.keys() A copy of D’s list of keys
D1.update(D2) for k, v in D2.items(): D2[k] = v
D.values() A copy of D’s list of values
D.get(k, default- The item of D with key k
val)
D.se tdefault(k [, D[k] if k in D, else defaultval (also setting it)
defaultval])
D.iteritems() Returns an iterator over (key, value) pairs
D.iterkeys() Returns an iterator over the mapping’s keys
D.itervalues() Returns an iterator over the mapping’s values.
D.pop(k [, default Removes key k and returns the corresponding value. If key is not found, de-
]) fault is returned if given, otherwise KeyError is raised.
D.popitem() Removes and returns an arbitrary (key, value) pair from D

Sets
Assume T, T1, T2 are sets (unless otherwise stated). In the table below you will find some of the very
frequently used member functions of sets (in the Operation column anything in square brackets denotes that
the content is optional – if you enter the optional content, do no type in the square brackets):

7.3. Widely-used Member Functions of Containers 151

Operation Result
T1.issubset(T2) True if every element in T1 is in iterable T2
T1.issuperset(T2) True if every element in T2 is in iterable T1
T.add(elt) Adds element elt to set T (if it doesn’t already exist)
T.remove(elt) Removes element elt from set T. KeyError if element not found
T.discard(elt) Removes element elt from set T if present
T.pop() Removes and returns an arbitrary element from set T; raises KeyError if empty
T.clear() Removes all elements from this set
T1 . Synonym to (T1 & T2). Returns a new Set with elements common to all sets (in
intersection(T2 the method T2, T3,… can be any iterable)
[, T3 ...])
T1.union(T2 [, T3 Synonym to (T1 | T2). Returns a new Set with elements from either set (in the
...]) method T2, T3, ... can be any iterable)
`` T1.difference(T2 [, Synonym to (T1 - T2). Returns a new Set with elements in T1 but not in any of
T3 …])`` T2, T3, .. (in the method T2, T3, ... can be any iterable)
` Synonym to (T1 ^ T2). Returns a new Set with elements from either of two sets
T1.symmetric_difference(T2)`
but not in their intersection
T.copy() Returns a shallow copy of set T

7.4 Important Concepts

We would like our readers to have grasped the following crucial concepts and keywords from this chapter:
• Encapsulation, inheritance and polymorphism.
• Benefits of the Object-Oriented Paradigm.
• Concepts such as class, instance, object, member, method, message passing.
• Concepts such as base class, ancestor, descendant.

7.5 Further Reading

• Special Methods in Python: https://docs.python.org/3/reference/datamodel.html#

special-method-names
• Object-oriented Programming chapter: https://link.springer.com/chapter/10.1007/
978-3-7091-1343-1_7

7.6 Exercises

Go to the interactive example and modify it as follows:

1. Add a triange class. It should take, at creation time, (x,y,dx2,dy2,dx3,dy3) as parameter, where
x,y are the coordinates of one of the corners of the triangle on the canvas, and dx2,dy2 are the incre-
ments relative to x1,y1 to reach the second corner of the triangle (so, the absolute coordinate on the
canvas of the second corner becomes (x+dx2,y+dy2)), dx3,dy3 have a similar meaning to reach the
third corner. You are expected to define all member functions present in circle or rectangle, this

152 Chapter 7. A Gentle Introduction to Object-Oriented Programming

time for triangle. For the draw member function, you can investigate the content of the draw of the
rectangle class.
2. Implement a move_on_line(x_fin, y_fin, count_of_steps) that will work for all geometric
shapes and displace a geometric shape starting from its current position on the canvas and ending at
(x_fin,y_fin) coordinate in count_of_steps many displacements.
3. Create a class car, which is a child-drawing style car, that consists of a rectangle and two circles
below it. It should implement all member functions of circle or rectangle. Make use of the existing
rectangle and circle classes.

7.6. Exercises 153

154 Chapter 7. A Gentle Introduction to Object-Oriented Programming
8 | File Handling

(C) Copyright Notice: This chapter is part of the book available athttps://pp4e-book.github.io/and copying,
distributing, modifying it requires explicit permission from the authors. See the book page for details:https:
//pp4e-book.github.io/
All variables used in a program are kept in the main memory and they are volatile, i.e., their values are lost
when the program ends. Even if you write a program running forever, your data will be lost in case of a
shutdown or a power failure.
Another drawback of the main memory is the capacity limitation. In the extreme case, when you need more
than a couple of gigabytes for your variables, it will be difficult to keep all of them in the main memory.
Especially, infrequently required variables are better kept on an external storage device instead of the main
memory.
Files provide a mechanism for storing data persistently in hard drives which provide significantly larger
storage than the main memory. These devices are also called secondary storage devices. The data you put
in a file will stay on the hard drive until someone overwrites or deletes the file (or when the hard drive fails,
which is a sad but rare case).
A File is a sequence of bytes stored on the secondary storage, typically hard drive (alternative secondary
storage devices include CD, DVD, USB disk, tape drive). Data on a file has the following differences from
data in memory (variables):
1. A file is just a sequence of bytes. Therefore, data in a file is unorganized, there is no data type, no
variable boundaries.
2. Data needs to be be accessed indirectly, using I/O functions. E.g. updating a value in a file requires
reading it in memory, updating in memory, then writing it into the file back.
3. Accessing and updating data is significantly slower since it is on an external device.
Keeping data on a file instead of the main memory has the following use cases:
1. Data needs to be persistent. Data will be in the file when you restart your program, reboot your machine
or when you find your ancient laptop in the basement 30 years later (Probably it will not be there when
a 3000BC archeologist finds your laptop on an excavation site. Hard disks are not that durable. So,
persistency is bounded).
2. You need to exchange data with another program. Examples:
• You download data from web and your program gets it as input.
• You like to generate data in your program and put it on a spreadsheet for further processing.
3. You have large amount of data which does not fit in the main memory. In this case, you will probably
use a library or software like a database management system to access data in a faster and organized
way. Files are the most primitive, basic way of achieving it.

155
In this chapter, we will talk about simple file access so that you will learn about simple file operations like
open, close, read, write. The examples of the chapter will create and modify files when run – we strongly
encourage you to check the contents of the created files using the file access mechanism at the left-hand side.

8.1 First Example

Let us quickly look at a simple example to get a feeling for the different steps involved in working with files.

fpointer = open('firstexample.txt',"w")
fpointer.write("hello\n")
fpointer.write("how are\n")
fpointer.write("you?\n")
fpointer.close()

The program above will create a file in the current directory with filename firstexample.txt. You can open
it with your favourite text editor (there are plenty of text editors for operating systems: notepad, wordpad,
textedit, nano, vim) to see and edit it. The content will look like this:

hello
how are
you?

First line of the program is fpointer = open('firstexample.txt',"w"). This line opens the file named
firstexample.txt for writing to it. If the file exists, its content will be erased (it will be an empty file
afterwards). The result of open() is a file object that we will use in the following lines. This object is
assigned to variable fpointer.
In the following lines, all functions we call with this file object fpointer will work on this file
(i.e. firstexample.txt). This special dot notation helps us with calling functions in scope of the file.
fpointer.functionname() will call the functionname function for this file. write(string) function will
write the string content to the file. Each call to write(string) will append the string to the file and the
file will grow. At the end, when we are done, we call close() to finish accessing the file so that your oper-
ating system will know and do necessary actions about it. All open files will be closed when your program
terminates. However, calling close() after finishing writing is a good programming practice.
Now, let us read this file:

fp = open("firstexample.txt","r")
content = fp.read()
fp.close()
print(content)

hello
how are
you?

In this case, we called open() with argument "r" which tells that we are going to read the file (or use it as an
input source). If you skip the second argument in open(), it is assumed to be "r", so open("firstexample.
txt") will be equivalent. The read() call gets an optional argument, which is the number of bytes to read.
If you skip it, it will read the whole file content and return it as a string. Therefore, after the call, the content
variable will be a string with the file content.

156 Chapter 8. File Handling

8.2 Files and Sequential Access

A file consists of bytes and read/write operations access those bytes sequentially. In sequential access,
the current I/O operation updates the file state so that next I/O operation will resume from the end of the
current I/O operation.
Assume you have an old MP3 player that supports only play me next 10 seconds operation on a button.
Pressing it will play the next 10 seconds of the song. When you press again, it will resume from where it is
left and play another 10 seconds. This follows until the song is over. The sequential access is like this. A
file pointer keeps the current offset of the file and each I/O operation advances it so that next call will read
or write from this new offset – see Fig. 8.2.1.

Fig. 8.2.1: Sequential reading of a file.

The following is a sample program illustrating sequential access:

fp = open("firstexample.txt","r") # the example file we created above

for i in range(3): # repeat 3 times

content = fp.read(4) # read 4 bytes in each step
print("> ", content) # output 4 bytes preceded by >

fp.close()

> hell
> o
ho
> w ar

The text in the file was:

hello
how are
you?

The first read() reads 'hell', the second reads 'o\nho' (note that \n stands for a new line so that ho is
printed on a new line), and the third reads 'w ar'. After these operations, the file offset is left at a position
so that the following reads will resume from content 'e\nyou\n'.
We provided the example with 4-byte read operations. However, for text files, the typical scenario is reading
characters line by line instead of fixed size strings.

8.2. Files and Sequential Access 157

8.3 Data Conversion and Parsing

A file, specifically a text file, consists of strings. However, especially in engineering and science, we work
with numbers. A number is represented in a text file as a sequence of characters including digits, a sign
prefix ('-' and '+') and at most one occurrence of a dot ('.'). That means in your Python program, you
may use π as 3.1416 however, in the text file, you store '3.1416', which is a string consisting of chars
'3','.','1','4','1','6'.

pi = 3.1416
pistr = '3.1416'
print(pi+pi,':', pi * 3)
print(pistr+pistr,':', pistr *3)

6.2832 : 9.4248
3.14163.1416 : 3.14163.14163.1416

Funny, the second line of output above is a result of Python interpreting + operator as string concatenation,
and * as adjoining multiple copies of a string. If we need to treat numbers as numbers, we need to convert
them from string. There are two handy functions for this: int() and float() convert a string into an integer
and a floating point value, respectively. Here is an illustration:

pistr = ' 0.31416E01 '

nstr = ' 47 '

# Convert numbers in the strings into numerical data types:

piflt = float(pistr)
nint = int(nstr)

print(piflt*2, nint*2)

6.2832 94

Note that we cannot call int('3.1416') since the string is not a valid integer. That brings us another
challenge of making sure that strings we need to convert are actually numbers. Obviously int('hello')
and float('one point five') will not work. The ways of dealing with such errors are left for the next
chapter. In this chapter, we assume that we have our data carefully created and all conversions work without
any error.
Our next challenge is having multiple numbers on a string separated by special characters or simply spaces
as '10.0 5.0 5.0'. In this case, we need to decompose a string into string pieces representing numbers,
so that we will have '10.0','5.0','5.0' for the above string. The next step will be converting them into
numbers:
Step 1 Step 2
'10.0 5.0 5.0' −→ ['10.0','5.0','5.0'] −→ [10.0, 5.0, 5.0]
For the first step, we will use the split() method of a string. String, or the variable containing the string,
is followed by .split(delimiter), which returns a list of strings separated by the given delimiters. The
delimiters are removed and all values in between are put in the list – for example:

print('a:b:c'.split(':'))
print('hello darkness, my old friend'.split(' '))
print('a <=> b <=> c'.split(' <=> '))
(continues on next page)

158 Chapter 8. File Handling

(continued from previous page)
print('multiple spaces are tricky'.split(' '))
a = '10.0 5.0 5.0'
print(a.split(' '))

['a', 'b', 'c']

['hello', 'darkness,', 'my', 'old', 'friend']
['a', 'b', 'c']
['multiple', '', '', '', '', '', '', 'spaces', '', '', '', '', '', '', '', '', '', 'are', '',
,→ '', '', '', '', '', '', '', 'tricky']

['10.0', '5.0', '5.0']

For the second step, we will use the float() function on a list (or the int() function if you have a list of
integers). We have couple of options for this. One is to start from an empty list and append the converted
value at each step:

instr = '10.0 5.0 5.0'

outlst = []

# Go over each substring

for substr in instr.split(' '):
outlst += [float(substr)] # Convert each element to float and append it to the list

print(outlst)

[10.0, 5.0, 5.0]

A more practical and faster version will be list comprehension, which is the compact version of mapping a
value into another as:

instr = '10.0 5.0 5.0'

outlst = [ float(substr) for substr in instr.split(' ')]

print(outlst)

[10.0, 5.0, 5.0]

As we have explained in Chapter 3, the syntax is similar to set/list notation in Math:

{f loat(s) | s ∈ S} vs. [float(s) for s in S]
If you need to have multiple spaces within the values, you can use “import re” and call “re.split(' +',
inputstr)” instead of “inputstr.split(' ')”. This will split the 'multiple spaces are tricky'
example above into 4 words without spaces. How it works is beyond the scope of the book. Curious readers
can refer to “re” and “parse” modules for more advanced forms of input parsing. These are not trivial
modules for beginners.
Now, let us consider the reverse of the operation: Assume we have a list of integers and we like to convert
that into string that can be written in a file:
Step 1 Step 2
[10.0, 5.0, 5.0] −→ ["10.0","5.0","5.0"] −→ "10.0 5.0 5.0"

8.3. Data Conversion and Parsing 159

The first step will be handled with the str() function which converts any Python value into a human readable
string:

inlst = [10.0, 5.0, 5.0]

outlst = [str(num) for num in inlst]

print(outlst)

['10.0', '5.0', '5.0']

The next step is to join those elements with a delimiter, which is reverse of the split() operation. Not by
accident, name of this operation is join(). join() is a method of the delimiter string and list is the argument
of it. ':'.join(['hello','how','are','you?']) returns 'hello:how:are:you?'.

inlst = [10.0, 5.0, 5.0]

outlst = [str(num) for num in inlst]

print(' '.join(outlst))

10.0 5.0 5.0

A more advanced way of converting values into strings is called formatted output and briefly introduced in
a section below.

8.4 Accessing Text Files Line by Line

Files consisting of human readable strings are called text files. Text files consist of strings separated by
the end of line character '\n', also known as new line. The sequence of characters in a file contains the
end-of-line characters so that a text editor will end the current line and show following characters on a new
line. We use end-of-line characters so that logically relevant data is on the same line. For example:

4
10.0 20.0
15.5 22.2
3 44
10 10.5

Let us assume the integer value 4 on the first line denotes how many lines will follow. Assume also that
each of the following 4 lines have two real values denoting x and y values of a point. In this way, we can
represent our input separated by end-of-line characters for each point and by space character for each value
in a line.
Let us create such a text file from a Python list. Please note that the file read function returns a string,
the write function expects a string argument. I.e., calling write(3.14) will fail. In order to make the
conversion, we use the str() function for numeric values and call write(str(3.14)) instead. Another
tricky point is that write() does not put end of line character automatically. You need to put it in the output
string or call an extra write("\n").

160 Chapter 8. File Handling

pointlist = [(0,0), (10,0), (10,10), (0,10)]

fp = open("pointlist.txt", "w") # open file for writing

fp.write(str(len(pointlist))) # write list length
fp.write('\n')

# Go over each point in the list

for (x,y) in pointlist: # for each x,y value in the list
fp.write(str(x)) # write x
fp.write(' ') # space as number separator
fp.write(str(y)) # write y
fp.write('\n') # \n as line separator

fp.close()

# let us read the content to verify what we wrote

fp = open("pointlist.txt") # open for reading
content = fp.read()
print(content)
fp.close()

4
0 0
10 0
10 10
0 10

Using read() will get the whole content of the file; if the file is large, your program would use too much
memory and processing the data will be difficult. Instead of that, we can access a text file line by line using
the readline() function.
Let us write a program to read and output the content of a text file. We need a loop to read the file line by line
and output. But, when we are going to stop is another problem. Python’s read() and readline() functions
return an empty string '' when there is nothing left to read. We can use this to stop reading:

fp = open("pointlist.txt") # open file for reading

nextline = fp.readline() # read the first line

while nextline != '': # while read is successful
print(nextline) # output the line
nextline = fp.readline() # read the nextline

fp.close() # when nextline == '' loop terminates

0 0

10 0

10 10

0 10

8.4. Accessing Text Files Line by Line 161

Please note the empty lines between the each output line. This is due to '\n' character at the end of the string
that readline() returns. In other words, readline() keeps the new line character it reads. print() puts
an end of line after the output (this can bu suppressed by adding an end='' argument). As a result, we have
extra end-of-line at the end of each line. In order to avoid it, you can call rstrip('\n') on the returned
string to remove end of line. The new code will be:

fp = open("pointlist.txt") # open file for reading

nextline = fp.readline() # read the first line

while nextline != '': # while read is successful
nextline = nextline.rstrip('\n') # remove occurrences of '\n' at the end
print(nextline) # output the line
nextline = fp.readline() # read the nextline

fp.close()

4
0 0
10 0
10 10
0 10

Converting this file into the initial Python list [(0,0), (10,0), (10,10), (0,10)] is our next challenge.
This requires conversion of a string as "0 0\n" into (0,0). The first one is of type str whereas the second
is a tuple of numeric values. We can use int() or float() functions to convert strings into numbers. Note
that the string should contain a valid representation of a Python numeric value: int("hello") will raise an
error.
The second issue is separating two numbers in the same string. We can use split() function followed by
the separator string as in nextline.split(' '). This call will return a sequence of strings from a string.
If the separator does not occur in the string, it will return a list with one element, if there is one separator, it
will return two elements. For n occurrences of the separator, it will return a list with n − 1 elements.
Here is the solution in Python:

fp = open("pointlist.txt") # open file for reading

pointlist = [] # start with empty list

nextline = fp.readline() # read the first line

n = int(nextline) # find number of lines to read

for i in range(n): # repeat n times

nextline = fp.readline() # read the nextline
nextline = nextline.rstrip('\n') # remove occurrences of '\n' at the end
(x, y) = nextline.split(' ') # get x and y (note that they are still strings)
x = float(x) # convert them into real values
y = float(y)
pointlist.append( (x,y) ) # add tuple at the end

fp.close()
print(pointlist) # output the resulting list

162 Chapter 8. File Handling

[(0.0, 0.0), (10.0, 0.0), (10.0, 10.0), (0.0, 10.0)]

8.5 Termination of Input

There are two ways to stop reading input:

1. By reading a definite number of items.
2. By the end of the file.
In our previous examples, we read an integer that told us how many lines followed in the file. Then, we
called readline() in a for loop with the given number of lines. This is an example of the first case which
provides a definite number of items.
The alternative is to read lines in a while loop until a termination condition arises. The termination condition
is usually the end of file, the case where functions like read() and readline() return an empty string ''.

fp = open("pointlist.txt") # open file for reading

pointlist = [] # start with empty list

nextline = fp.readline() # skip the first line (4) since we don't need it

nextline = fp.readline() # read the first line

while nextline != '': # until end of file
nextline = nextline.rstrip('\n') # remove occurrences of '\n' at the end
(x, y) = nextline.split(' ') # get x and y (note that they are still strings)
x = float(x) # convert them into real values
y = float(y)
pointlist.append( (x,y) ) # add tuple at the end
nextline = fp.readline() # read the nextline

fp.close()
print(pointlist)

[(0.0, 0.0), (10.0, 0.0), (10.0, 10.0), (0.0, 10.0)]

Note that the example above skips (reads and throws away) the first line so that the integer on the first line
is ignored. When your input does not contain such an unnecessary value, you can delete this line.
Sometimes termination can be marked explicitly by a sentinel value which is a value marking the end of
values. This is especially useful when you have multiple objects to read:

# First, create a file named `twopointlists.txt`

fp = open("twopointlists.txt", "w")
fp.write("""3 0
3.4 2.1
5.1 3.2
EOLIST
1 1.5
2.0 2.5""")
fp.close()

This will create a sample file content as:

8.5. Termination of Input 163

3 0
3.4 2.1
5.1 3.2
EOLIST
1 1.5
2.0 2.5

fp = open("twopointlists.txt")
pntlst1 = [] # start with empty list
pntlst2 = [] # start with empty list

nextline = fp.readline() # read the first line

while nextline != 'EOLIST\n': # sentinel value
nextline = nextline.rstrip('\n') # remove occurrences of '\n' at the end
(x, y) = nextline.split(' ') # get x and y (note that they are still strings)
x = float(x) # convert them into real values
y = float(y)
pntlst1.append( (x,y) ) # add tuple at the end
nextline = fp.readline() # read the nextline

# first list has been read, now continue with the second list from the same file
nextline = fp.readline()
while nextline != '': # until end of file
nextline = nextline.rstrip('\n') # remove occurrences of '\n' at the end
(x, y) = nextline.split(' ') # get x and y (note that they are still strings)
x = float(x) # convert them into real values
y = float(y)
pntlst2.append( (x,y) ) # add tuple at the end
nextline = fp.readline() # read the nextline

fp.close()
print('List 1:', pntlst1)
print('List 2:', pntlst2)

List 1: [(3.0, 0.0), (3.4, 2.1), (5.1, 3.2)]

List 2: [(1.0, 1.5), (2.0, 2.5)]

The output will be:

List 1: [(3.0, 0.0), (3.4, 2.1), (5.1, 3.2)]

List 2: [(1.0, 1.5), (2.0, 2.5)]

8.6 Example: Processing CSV Files

CSV stands for Comma Separated Value; it is a text-based format for exporting/importing spreadsheet
(i.e. Excel) data. Each row in a CSV file is separated by newlines and each column is separated by a comma
,. Actually, the format is more complex but for the time being, let us ignore comma that might be appearing
in strings and focus on a simple form as follows:

164 Chapter 8. File Handling

Name,Surname,Grade
Han,Solo,80
Luke,Skywalker,90
Obi,Van Kenobi,88
Leya,Skywalker,91
Anakin,Skywalker,55

Usually first line is the names of the columns in a spreadsheet. Now, let us create this file:

content = '''Name,Surname,Grade
Han,Solo,80
Luke,Skywalker,90
Obi,Van Kenobi,88
Leya,Skywalker,91
Anakin,Skywalker,55
'''
fp = open("first.csv", "w") # open for writing
fp.write(content) # write in a single operation, practical for small files
fp.close()

Our next task is to read this file in memory as a list of dictionary form, as: [{"Name":"Han",
"Surname":"Solo","Grade":"80"},...]
We need to read the file line by line, extract the components using the split() function, then create the
dictionary. Then, we can append it to resulting list. For example:

fp = open("first.csv","r") # open for reading

line = fp.readline() # read column names

line = line.rstrip('\n') # get rid of new line
colnames = line.split(',') # list of column names

result = [] # resulting list of dictionaries

line = fp.readline()
while line != '': # end-of-file check
line = line.rstrip('\n')
entry = {} # start with empty dictionary
c = 0 # a counter to address column number
for v in line.split(','): # in a loop process each column of the row
entry[colnames[c]] = v # column name is index, value is from current row
c += 1
result.append(entry) # add dictionary to result
line = fp.readline() # read next line

fp.close()
print(type(result))
print(result)

<class 'list'>
[{'Name': 'Han', 'Surname': 'Solo', 'Grade': '80'}, {'Name': 'Luke', 'Surname': 'Skywalker',
,→ 'Grade': '90'}, {'Name': 'Obi', 'Surname': 'Van Kenobi', 'Grade': '88'}, {'Name': 'Leya',

,→ 'Surname': 'Skywalker', 'Grade': '91'}, {'Name': 'Anakin', 'Surname': 'Skywalker', 'Grade

,→': '55'}]

Let us improve this example by adding a column as a result of a computation. Let us calculate the grade

8.6. Example: Processing CSV Files 165

average and show the difference from the average as a new column. We need to go over all grade values in
the list, convert to them real values (so that we can do arithmetic on them), calculate the average, then go
over all rows to add a new column. Then, go over the list again to export/write it into a new CSV file.

n = 0

# Calculate the average

sum = 0
for entry in result:
sum += float(entry['Grade'])
n += 1
average = sum / n

# Calculate the difference of each grade from the average

for entry in result:
entry['Avgdiff'] = str(float(entry['Grade']) - average)

# Write the updated content into another CSV file

fp = open('second.csv', 'w')
colnames = entry.keys() # this returns the keys (column names) of the CSV file
fp.write(','.join(colnames) + '\n') # write this as the first line with comma separated␣
,→values

for entry in result: # Go over each row

vals = []
for key in colnames: # Write each column on this row
vals.append(entry[key]) # extract values of entry, entry.values() is a short␣
,→version of this
fp.write(','.join(vals) + '\n')

# Finished, close the file

fp.close()

%cat second.csv

Name,Surname,Grade,Avgdiff
Han,Solo,80,-0.7999999999999972
Luke,Skywalker,90,9.200000000000003
Obi,Van Kenobi,88,7.200000000000003
Leya,Skywalker,91,10.200000000000003
Anakin,Skywalker,55,-25.799999999999997

8.7 Formatting Files

Sometimes readability is important for text files, especially if data is in a tabular form. For example, seeing
all related data in a column start at the same position can improve readability significantly. The following
shows the unformatted and formatted versions of the same data side by side:

Name,Surname,Grade,Avgdiff Name , Surname , Grade, Avgdiff

Han,Solo,80,-0.7999999999999972 Han , Solo , 80, -0.800
Luke,Skywalker,90,9.200000000000003 Luke , Skywalker, , 90, 9.200
Obi,Van Kenobi,88,7.200000000000003 Obi , Van Kenobi , 88, 7.200
(continues on next page)

166 Chapter 8. File Handling

(continued from previous page)
Leya,Skywalker,91,10.200000000000003 Leya , Skywalker , 91, 10.200
Anakin,Skywalker,55,-25.799999999999997 Anakin , Skywalker , 55,-25.800

In order to achieve this, you can use the format() method of a template string as in
'{:10}, {:20}, {:3d}, {:7.3f}'.format('Han', 'Solo', 80, -0.2)'.
Each {} in the template matches a data value in the arguments. The value after : denotes the (minimum)
width of the data. If data fits in a smaller number of characters, spaces are inserted on the right to make it
have exactly given size (left-aligned). For integers, the number is followed by a d to format it as a decimal
value spaced padded on the left (right aligned). For floating point values, this value can be followed by a ‘.’
and another number and an f. The second number denotes the size of the fraction, f marks this value as a
float, and the fraction part is rounded to given number of digits.
The detailed description of format() is out of the scope of this course and the document. For detailed
description, please refer to Python reference manuals.
Let us rewrite the output part of the code using formatted output:

template = '{:10}, {:20}, {:5d}, {:7.3f}\n'

fp = open('third.csv', 'w')
colnames = entry.keys() # this returns the␣
,→keys of the CSV file
fp.write('{:10}, {:20}, {:5}, {:7}\n'.format(*colnames) ) # write this␣
,→as the first line with comma separated values
for entry in result:
fp.write(template.format(entry['Name'],entry['Surname'], # convert strings␣
,→to numbers to respect number formatting
int(entry['Grade']), float(entry['Avgdiff'])))
fp.close()

# Let us display the content of the file after formatting:

%cat third.csv

Name , Surname , Grade, Avgdiff

Han , Solo , 80, -0.800
Luke , Skywalker , 90, 9.200
Obi , Van Kenobi , 88, 7.200
Leya , Skywalker , 91, 10.200
Anakin , Skywalker , 55, -25.800

8.8 Binary Files

So far, we have only looked at text files where all values are represented as human readable text where all
numerical values are represented as decimal strings. However, if you remember our early chapters, comput-
ers do not store and process numbers as decimal digit sequences. They store variables in binary format like
Two’s Complement and the IEEE754 floating point standard. In order to process, read and write decimal
data in text, programming language and libraries convert data. Even though you won’t notice the time spent
in conversion, if you read 10 millions of numbers, you start spending significant amount of CPU time for
converting data.

8.8. Binary Files 167

Binary files, on the other hand, store numbers as they are stored in computer’s memory. They are still
sequences of bytes, but in a more structured way. For example, a 4-byte integer is kept as a sequence of 4
bytes, each byte is a part of the number in Two’s Complement form. Reading a binary file is simply copying
data to memory, either no conversion is performed or only the order of bytes is changed.
A floating point number 0 takes 1 byte in a text file, but the number 3.
1415926535897932384626433832795028 takes 34 bytes. In a binary file, the total size of a number
is fixed as the size of IEEE 754 format, i.e. 4 bytes on a 32-bit computer. Both 0 and the π are stored in 4
bytes for single precision, 8 bytes for double precision, in a binary file.
Keeping values in binary files has the following advantages:
1. It is more compact: Data occupies less space in the file.
2. No decimal to binary conversion is required. More efficient in terms of CPU usage.
3. Since sizes are fixed, randomly jumping to a location and reading relevant data is possible. In a text
file, you have to start from the beginning and read all lines up to the relevant data. This kind of usage
is a more advanced case and harder to understand for beginners.
On the other hand, using text files has the following advantages:
1. Files are human readable and editable. User can change data using a standard editor. In binary files,
special software should be used.
2. File format is more flexible, using variablename: value patterns in the file, data can be stored in
any order in a flexible way. This is why text files are often used as configuration files.
Most of the special formats with .exe, .xls, .zip, .pdf extensions are binary file formats.
Note: Binary files are kept out of scope of this book. The following paragraphs give couple of pointers for
curious readers.
In order to use binary files:
1. You need to add 'b' character in the second argument of open() method as: open('test.bin',
'rb') or open('test.bin','wb').
2. Binary I/O requires bytes typed values instead of str typed values. bytes is a sequence of bytes.
Elements of a byte sequence are not printable in contrast to str.
3. Python has struct module for converting any value into bytes value. struct.pack(format, val-
ues) converts values into bytes. However, this conversion is much more cheaper computationally
than decimal to binary conversion.
4. struct.unpack(format, bytesval) can be used to convert bytes value into Python values. It is
much cheaper than binary to decimal conversion.
5. read(), write() can be used as usual. In read(nbytes), data size should be given. struct.
calcsize(format) can be used to calculate data size from format.
The following is an example of binary I/O. Assume the binary file contains an integer N , for the number of
points, and followed by 2 × N floating point values. Let us write and then read this data:

import struct

points = [(1,1), (2.5, 3.4), (5.4,3.3), (2.2, 1.121)]

# 1- Open and write the binary file

(continues on next page)

168 Chapter 8. File Handling

(continued from previous page)
fp = open("points.bin", "wb")
fp.write(struct.pack('i', len(points))) # 'i' denotes a single integer value is converted␣
,→into bytes

for (x,y) in points:

fp.write(struct.pack('dd', x, y)) # 'dd' denotes two double precision floating point␣
,→values are converted into bytes

fp.close()

# 2- Open and read the binary file

fp = open("points.bin", "rb") # open same file for reading
content = fp.read(struct.calcsize('i')) # read binary data with length sizeof integer bytes
(n,) = struct.unpack('i', content) # unpack returns a tuple, 1tuple in this case

newpoints = []
for i in range(n): # n times
content = fp.read(struct.calcsize('dd'))
(x,y) = struct.unpack('dd', content) # read two double precision floats
newpoints += [(x,y)] # append value at the end

fp.close()

# 3- Print the read and converted values

print("The read & converted points are:", newpoints)

print("This is what binary data looks like:")

fp = open("points.bin", "rb")
print(fp.read())
fp.close()

The read & converted points are: [(1.0, 1.0), (2.5, 3.4), (5.4, 3.3), (2.2, 1.121)]
This is what binary data looks like:
b'x04x00x00x00x00x00x00x00x00x00xf0?x00x00x00x00x00x00xf0?x00x00x00x00x00x00x04@333333x0b@x9ax99x99x
xb2x9dxefxf1?'

8.9 Note on Files, Directory Organization and Paths

Files are organized under directories so that you can put relevant files under same category together. Op-
erating systems provide a filesystem hierarchy consisting of directories (some prefer word folder instead)
and regular files.
Directories can be arbitrarily nested. You may need to traverse N levels of directories to find your file.
For example, your program can be in the Homeworks directory under the Desktop directory under the user
directory under the Desktop Users directory under the root directory (/), which is the topmost level on
your filesystem. The top-level directory and the seperator is backslash,\ in MS Windows operating systems.
However, ‘/’ works in Python for Windows too.
In order to address a file, we use a path which is a sequence of directory names separated by /, ended by the
name of the file. For example, "/Desktop Users/user/Desktop/Homeworks/homework1.py" is a path
for the file named "homework1.py".
A path can be either full (absolute) or relative. In the former case, it starts with a slash (/). In the relative

8.9. Note on Files, Directory Organization and Paths 169

case, it is considered relative to the current working directory, i.e. the directory where you started your
program. Full paths ignore your current directory whereas relative paths depend on it. For example, if you
are currently in the Desktop directory, the path "Homeworks/homework1.py" will address the same full path
above.

8.10 List of File Class Member Functions

For completeness, below you can find commonly used member functions of the file class.
Assume F is a file. In the table below you will find some of the very frequently used member functions of
files (in the Operation column anything in square brackets denotes that the content is optional – if you enter
the optional content, do no type in the square brackets):

Operation Result
F.se Set file F’s position, like stdio’s fseek(). whence � 0 then use absolute indexing
ek(offset[, (using offset). whence � 1 then offset relative to current pos. whence � 2 then offset
whence=0]) relative to file end.
F.tell() Return file F’s current position (byte offset).
F. Truncate F’s size. If size is present, F is truncated to (at most) that size, otherwise F is
truncate([size])
truncated at current position (which remains unchanged).
F.write(str) Write string str to file F.
F. Write list of strings to file F. No EOL are added.
writelines(list)
F.close() Close file F.
F.fileno() Get fileno (fd) for file F.
F.flush() Flush file F’s internal buffer.
F.isatty() 1 if file F is connected to a tty-like dev, else 0.
F.next() Returns the next input line of file F, or raises StopIteration when EOF is hit.
F. Read at most size bytes from file F and return as a string object. If size omitted, read
read([size]) to EOF.
F.readline() Read one entire line from file F. The returned line has a trailing \n, except possibly at
EOF. Return "" on EOF.
F. Read until EOF with readline() and return a list of lines read.
readlines()

8.11 Important Concepts

We would like our readers to have grasped the following crucial concepts and keywords from this chapter:
• Sequential access. File access.
• Text files. Reading and writing text files. Parsing a text file.
• End of file, new line.
• Formatting files.
• Binary files and binary file access.

170 Chapter 8. File Handling

8.12 Further Reading

• String formatting in Python: https://docs.python.org/3.4/library/string.html#formatspec

• Working with binary data and files in Python: https://docs.python.org/3/library/binary.html
• Comma-Separated Values (CSV) file format: https://en.wikipedia.org/wiki/Comma-separated_values

8.13 Exercises

• Write a function that reads a text file with the following format (ignore characters following #):

N # Number of students
# Empty line
Name Surname # Fist student
M # Number of courses that the student has taken
Coursename1: Grade # Grade is a real number
Coursename2: Grade
Coursename3: Grade
...
CoursenameM: Grade

# Empty line
Name Surname # Second student
P # Number of courses that the student has taken
Coursename1: Grade # Grade is a real number
Coursename2: Grade
Coursename3: Grade
...
CoursenameP: Grade
...
...
...
# Empty line
Name Surname # Last student
Z # Number of courses that the student has taken
Coursename1: Grade # Grade is a real number
Coursename2: Grade
Coursename3: Grade
...
CoursenameZ: Grade

• Write a function that writes a list of dictionaries that have the following format into a text file. You
may choose to write the number of elements at the top of the file.

{ "city": "Ankara",
"plate code": "06",
"max temperature (C)": 40,
"min temperature (C)": -20,
"population": 5700000
}

• Write a function that reads a list of dictionaries from a file that you have written in the previous
question.

8.12. Further Reading 171

• Write a function that read the text file given below, represent the same content in binary, save it in a
binary file and read it back.
3 0 3.4 2.1 5.1 3.2 EOLIST 1 1.5 2.0 2.5

172 Chapter 8. File Handling

9 | Error Handling and Debugging

9.1 Types of Errors

9.1.1 Syntax Errors

Syntax errors are due to strict wellformedness requirements of programming languages. In a natural language
essay, we can use no punctuation at all, we can use silly abbreviations, mix the ordering of words, make typing
mistakes and still it will make sense to a reader (though your English teacher might reduce some points).
Computer programs are entirely different. When a programming language specification tells you to define
blocks based on indentation, to match parentheses properly, to follow certain statements with some certain
punctuation (i.e. loops and functions and :), you have to obey that. Because our compilers and interpreters
cannot convert a program with syntax errors into a machine understandable form.
The first step of the Python interpreter reading your program is to break it up into parts and construct a
machine-readable structure out of it. If it fails, you will get a syntax error, for example:

>>> for i in range(10)

... print(i)

File "<ipython-input-1-12d72cac235a>", line 1

(continues on next page)

173
(continued from previous page)
for i in range(10)
^
SyntaxError: invalid syntax

>>> x = float(input())
>>> a = ((x+5)*12+4

File "<ipython-input-2-dead5b360d91>", line 2

a = ((x+5)*12+4
^
SyntaxError: invalid syntax

>>> s = 0
>>> for i in range(10):
... s += i
... print(i)

File "<ipython-input-3-c3ef5d622e47>", line 4

print(i)
^
IndentationError: unexpected indent

>>> while x = 4:
... s += x

File "<ipython-input-4-befcf7769cec>", line 1

while x = 4:
^
SyntaxError: invalid syntax

In the examples above, syntax errors are:

1. : is missing at the end of the for loop.
2. The first parenthesis does not have a matching closing parenthesis.
3. Different indentation levels are used in the loop body.
4. while expects a boolean expression but an assignment is given (= is used instead of ==).
These are only a small sample of large number of possible syntax errors one can do.
Syntax errors are the most innocent errors since you are notified of the error immediately when you start run-
ning your program. Running your program once will give you the exact spot (though, sometimes parentheses
and quote matching can be non-trivial) in the error output. If you have learnt the syntax of your language,
you can fix it with a small effort.

174 Chapter 9. Error Handling and Debugging

9.1.2 Type errors

Python is an interpreted language and it does not make strict type checks like a compiler would check type
compatibility at compile time. Though, when it comes to performing an operation that is not compatible with
the current type of a variable/data, Python will complain. For example, when you try to override a string
element with an integer value, use an integer on a string context, select an element out of a float variables
etc., you will get an error:

astr = 'hello'
bflt = 4.56
cdict = {'a':4, 'b':5}

print(astr ** 3) # second power of a string

print(bflt[1]) # select first member of a float
print(cdict * 2) # multiply a dictionary by two
cdict < astr # compare a dictionary with a string

These will lead to the following errors:

>>> print(astr ** 3)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: unsupported operand type(s) for ** or pow(): 'str' and 'int'
>>> print(bflt[1])
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: 'float' object is not subscriptable
>>> print(cdict * 2)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: unsupported operand type(s) for *: 'dict' and 'int'
>>> cdict < astr
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: '<' not supported between instances of 'dict' and 'str'

Compiled languages and some interpreters enforce type compatibility at compile time and treat all such errors
as if they are syntax errors, providing a safer programming experience at run time. However, Python and
most other interpreters wait until command is first executed and raise the error at the execution time, causing
a run-time error.

9.1.3 Run-Time Errors

Run-time errors are more sneaky compared to syntax and other compile-time errors. Your program starts
to execute silently, visit many loops and functions without an error, produce some intermediate output, and
suddenly, in the most unexpected moment, it raises an error and you get a disappointing error text instead of
the decent result output that you expected.
The following example gets an input value and counts how many of the integers in the range [1, 1000] are
divisible by the input value.

def divisible(m, n):

return m % n == 0
(continues on next page)

9.1. Types of Errors 175

(continued from previous page)

def count(m):
sum = 0
for i in range(1,1000):
if divisible(i, m):
sum += 1

return sum

value = int(input())
print('input value is:', value)
print(value,' divides ', count(value), ' many integers in range [1, 1000]')

It works without any problem for non-zero input values. However if you enter 0, it outputs:

Fig. 9.1.1: A traceback example.

This output is called error Traceback. When a run-time error occurs, you will get one of these and the
execution of the program will be terminated. Depending on the settings, the output may be more brief but
you will get line numbers and the line leading to the error.
The reason for this example error is called an Exception and colored red in the above output. It is given in
the first line and repeated with explanation in the last line. Traceback also reports the functions involving
the error, from the outermost to the innermost. Each arrow (colored green) marks the program line causing
the error. The last item in the traceback is the actual position where the error occurred; the 2nd line having
return m % n == 0 in this case. The divisible() function containing the error line is called by the 7th
line, the count() function, which is called by 14th line of the program. By looking at the error output, you
will see the whole chain of calls that lead your program to the error state. The last one is the most important
one, where Python tried to execute an operation and failed.
The reason for failure is ZeroDivisionError in this case. If you trace the program with input value 0, you
will see that Python tried to evaluated m % 0 == 0 for some m value. The remainder operator is division

176 Chapter 9. Error Handling and Debugging

based and dividing something by 0 is impossible mathematically. All programming languages reject this
operation by a run-time error, and this is the Python style.
The following are the most common exceptions and run-time errors in Python:

Exception Reason
KeyboardInter- User presses Ctrl-C; not an error but user intervention
rupt
ZeroDivision- Right-hand side of / or % is 0
Error
AttributeError Object/class does not have a member
EOFError input() function gets End-of-Input by user
IndexError Container index is not valid (negative or larger than length)
KeyError dict has no such key
FileNotFoundEr- The target file of open() does not exist
ror
TypeError Wrong operand or parameter types, or wrong number of parameters for functions
ValueError The given value has correct type but the operation is not supported for the given
value

The following code illustrates each of these errors (except for the user-generated ones):

b = 0
a = a / b # ZeroDivisionError

x = [1,2,3]
print(x.length) # AttributeError: lists does not have a length attribute (use␣
,→len(x))

print(x[4]) # IndexError: last valid index of list x is 2

person = { 'name' : 'Han', 'surname': 'Solo'}

print(person['Name']) # KeyError: person does not have 'Name' key but 'name'

fp = open("example.txt") # FileNotFoundError: file "example.txt" does not exist

print([1,2,3] / 12) # TypeError: Division is not defined for lists

def f(x, y):

return x*x+y

print(f(5)) # TypeError: Only one element is supplied instead of 2.

print(int('thirtytwo')) # ValueError: string value does not represent an integer

a,b,c = [1,2,3,4] # ValueError: too many values on the right hand side

9.1. Types of Errors 177

9.1.4 Logical Errors

This is the worst among all errors because Python does not raise an error. It single-mindedly keeps running
your program; however, your program has a mistake about what it intends to compute. In other words, it
simply does not do what it is supposed to do because of an error.
Such an error can be due to a small typo, improper nesting of blocks, bad initialization of variables, and
many other reasons. However, the code segment containing the error has correct syntax and hence, it does
not cause a run-time error. As a result, you will not know such an error exists until you see some problems
in your output.
The following are a few examples for logical errors:

y = x / x+1 # you meant y = x / (x+1), forgetting about precedence

lastresult = 0
def missglobal(x):
result = x*x+1 # you intend to update the global variable
if lastresult != result: # but you assign a local variable instead
lastresult = result # you should have used "global lastresult"

def returnsnothing(x, y):

y = x*x+y*y
if x <= y:
return x # if x > y, the function returns nothing
print(returnsnothing(0.1, 0.1)) # does not have any value. prints "None"

s = 1
while i < n: # you forgot incrementing i as i+=1
s += s*x/i # loop will run forever. "infinite loop"

9.2 How to Work with Errors

Errors are unavoidable in programming. As you get experience, they will decrease but they will still be there.
Fortunately, we have methods to have less errors but completely getting rid of them is not possible.
The following is a list of strategies for eliminating errors:
1. Program with care
2. Place controls in your code
3. Handle exceptions
4. Write verification code and raise exceptions
5. Debug your program
6. Write test cases
Those strategies can be applied by programmers with different experience levels. In the extreme case, assur-
ing quality of programs and software testing are important professions of the software engineering discipline
and handled by dedicated engineers.

178 Chapter 9. Error Handling and Debugging

9.2.1 Program with Care

The best way to write error-free programs is not to cause one in the first place. Instead of cleaning your
living room every day, you may just choose not to throw garbage around. Programming is similar, the great
percentage of errors are due to the careless acts of a programmer.
Programming is not an evolutionary process where you start from an ugly code and make it better and better
as you correct errors. This way of programming is possible but not efficient. You better spend some time
before starting to write code to to design your solution and develop a strategy: which functions you need,
which data structures you use, which algorithms you use, and which order you will write the code. It is a
good practice to divide your problem into pieces (functions for example). Write one function at a time and
test it before going into the next step.
This strategy will get better as you become a more experienced programmer but nothing stops you from
paying more attention on your first day of coding.

9.2.2 Place Controls in Your Code

The values that a variable can have or a function may return are not known in advance. Especially, what
input user may provide is completely untrustable. Also, there are other environmental issues like existence
of a file.
If you have such an untrustable value, the next operation may fail as a result of it like:

a = [1,2,3]
age = {'Han': 30, 'Leia': 20, 'Luke': 20}

# CASE 1
n = int(input())
print(a[n]) # will fail for n > 2 or n < -2

# CASE 2
name = input()
print(age[name]) # will fail names other than 'Han', 'Leia', 'Luke'

# CASE 3
x = float(input())
y = math.sqrt(x) # will fail for x < 0
y = 1 / x # will fail for x == 0

In order to deal with such errors, you can check all values, especially user supplied ones. Checking all input
values before starting a computation is called Input Sanitization.

# CASE 1 with sanitization

n = int(input())
if -len(a) <= n < len(a):
print(a[n])
else:
print("n is not valid:", n)

# CASE 2 with sanitization

name = input()
if name in age: # membership test for dictionaries
print(age[name])
(continues on next page)

9.2. How to Work with Errors 179

(continued from previous page)
else:
print("dictionary does not have member:", name)

# CASE 3 with sanitization

x = float(input())
if x >= 0:
y = math.sqrt(x)
else:
print("invalid for sqrt operation: ", x)

if x != 0:
y = 1 / x
else:
print("divisor cannot be 0")

Writing such conditions look like a tedious job at first, but dealing with them while your program is running
is much worse for users. If you are writing a program that will be really useful, you have to do these types
of input checks.

9.2.3 Handle Exceptions

This is an alternative to putting check conditions in your code. Sometimes there are too many conditions,
one for each step of your computation, so that each line of code opens a new if block and nests like the
branches of a tree. Instead of writing whole nested-sequence of if ... else statements, exceptions give
you a chance to handle all errros in the same place. Especially, they allow you to handle the errors in the
caller function rather than the original place where error occurred. Before we can see how this works, we
have to learn a new syntax first:

try:
...... # a block with possible errors
...... # if there are function calls here
...... # and error occurs in the function, we can handle error here
except exceptionname: # exceptionname is optional
..... # this is error handling block.
..... # when there is an error, execution jumps here

This try-except block has two parts: A group of code (after try:) that possibly generates run-time errors
and a second or more blocks to handle the errors. When the first error occurs in the try part, the execution
jumps to the except part. If except part matches the corresponding block, it is executed. Multiple except
blocks are allowed for different kinds of exceptions. except :, without an exception name, matches all
errors and can be used to handle all of them one block.
The following is the same example in the previous section, but errors are handled in the same place.

import math

a = [1,2,3]
age = {'Han': 30, 'Leia': 20, 'Luke': 20}

try:
n = int(input())
(continues on next page)

180 Chapter 9. Error Handling and Debugging

(continued from previous page)
print(a[n]) # will fail for n > 2 or n < -2

name = input()
print(age[name]) # will fail names other than 'Han', 'Leia', 'Luke'

x = float(input())
y = math.sqrt(x) # will fail for x < 0
y = 1 / x # will fail for x == 0
except IndexError:
print('List index is not valid')
except KeyError:
print('Dictionary does not have such key')
except ValueError:
print('Invalid value for square root operation')
except ZeroDivisionError:
print('Division by zero does not have value')
except:
print('None of the known errors. Something happened even if nothing happened')

None of the known errors. Something happened even if nothing happened

You can try the above example with different inputs:

1. -3
2. 2 'Obi'
3. 2 'Han' -2
4. 2 'Han' 0
5. 2 'Han' 1
The first will have a[n] raise an IndexError. The second will have age[name] raise a KeyError. The third
will have sqrt(x) raise a ValueError. The fourth will have 1 / x raise a ZeroDivisionError. The last
one will have no error and finish the try block and continue with the next instruction jumping over except
blocks. You can also create exceptions with raise statement. raise ValueError will create the error.
Exceptions save you from nested if .. else statements for data validation, e.g.:

if cond1:
..1..
if cond2:
..2..
if cond3:
..3..
..4.. # success at last
else:
# report error
else:
# report error
else:
# report error

is harder to read compared to the following:

9.2. How to Work with Errors 181

try:
if !cond1:
raise Error

..1..

if !cond2:
raise Error

..2..

if !cond3:
raise Error

..3..
..4.. # success
except :
... Error handling

which is more flat and allows you to focus on the actual code.

9.2.4 Write verification code and raise exceptions

Even if you sanitize user input and check all arguments for valid values, your program can still have logical
errors and it might calculate incorrect intermediate/final values. If your program consists of multiple steps,
a logical error in step one will cause step two to calculate an incorrect value and this will create a snowball
effect, potentially causing all steps and the last to fail.
For example, you wrote a function for solving second-order equations ax2 + bx + c = 0, named it
solvesecond(a,b,c) which returns (x1, x2) as the roots. You are (almost) sure that you always send
correct a,b,c values to this function. However, it won’t hurt if you add a check like the following:

def solvesecond(a,b,c):
det = b*b - 4*a*c
# the following is the verification code
if det < 0:
print("Equation has no real roots for", a, b, c)
raise ValueError
....
...

The math.sqrt function would have raised the exception anyway. However, you add extra error message
about what caused the problem. Also, in logical errors, there is no run-time error and this error may be as
serious as finding imaginary roots.

182 Chapter 9. Error Handling and Debugging

9.2.5 Debug Your Code

Finding the position and the cause of a programming error and fixing it is called debugging. It involves
pinpointing the position of the error, reasons causing the error and updating the code so that it does not cause
the error any more.
Debugging methods are explained in detail in its own section below.

9.2.6 Write Test Cases

In order to make sure your program is working properly, in other words, it contains no logical errors, you
need to test it. Testing is an important yet non-trivial part of all engineering disciplines.
In order to test a program, you should create a set of inputs, run your program for each input case and collect
the outputs. If there is a way of verifying the outputs, you can verify them. For example, if you solve an
equation, you substitute the found variable in the equation to test if it holds:

(x1, x2) = findrootsecond(a,b,c)

if ax1x1 + bx1 + c != 0 or ax2X2 + bx2 +c != 0:

print('test failed for', a, b, c, 'roots', x1, x2)

You can generate millions of such numbers and automatically verify them.
If the problem is not verifiable, you can get a set of known solutions and compare your solutions against the
known solutions.

9.3 Debugging

Debugging is an act of looking for errors in a code, finding what causes the errors and correcting them. As
even the modest programs can go as large as hundreds of lines of code, debugging is not an easy task. Even
if a run-time error gives you the exact location of the error, the variable value causing the error can be owing
to a different place in your code and you will not get the variable state of the program at the moment.
Debugging is an iterative activity. You divide your program into parts and eliminate each part one by one,
ruling out the error: Narrow down all possibilities step by step to find the exact step where you made the
mistake. Run-time errors help you speed-up this process.
There are several methods for debugging. A programmer may use one or more of them for debugging.
The following is a sample program with an error (you may check the Colab link and run this interactive
example):

def startswith(srcstr, tarstr):

'''check if tarstr starts with srcstr
like srcstr="abra" tarstr="abracadabra" '''
for i in range(len(srcstr)): # check all characters of srcstr
if srcstr[i] != tarstr[i]: # if does not match return False
return False
return True # if False is not returned yet, it matches

(continues on next page)

9.3. Debugging 183

(continued from previous page)
def findstr(srcstr, tarstr):
'''Find position of srcstr in tarstr'''
for i in range(len(tarstr)):
# if scrstr is same as tarstr from i to rest
# return i
if startswith(srcstr, tarstr[i:]):
return i
return -1

print(findstr("ada", "abracadabra"))
print(findstr("aba", "abracadabra"))

which outputs the following when run:

5
---------------------------------------------------------------------------
IndexError Traceback (most recent call last)
<ipython-input-3-b00405b7708c> in <module>()
18
19 print(findstr("ada", "abracadabra"))
---> 20 print(findstr("aba", "abracadabra"))

1 frames
<ipython-input-3-b00405b7708c> in startswith(srcstr, tarstr)
3 like srcstr="abra" tarstr="abracadabra" '''
4 for i in range(len(srcstr)): # check all characters of srcstr
----> 5 if srcstr[i] != tarstr[i]: # if does not match return False
6 return False
7 return True # if False is not returned yet, it matches

IndexError: string index out of range

The first call returns 5, which is printed on screen, telling us that "ada" is at position 5 of "abracadabra",
matches after "abrac". However, the second call should return -1 since the "aba" substring does not exist
in "abracadabra" but it generates an error.

9.3.1 Debugging Using Debugging Outputs

This is one of the simplest, oldest but still an effective method for debugging a program. In this method,
you add extra output lines that shed light on the behaviour of your program. This way, you can trace where
your program diverted from the expected behaviour. For example (you may check the Colab link and run
this interactive example):

def startswith(srcstr, tarstr):

'''check if tarstr starts with srcstr
like srcstr="abra" tarstr="abracadabra" '''
for i in range(len(srcstr)): # check all characters of srcstr
print("check if ", srcstr, '!=', tarstr, 'for i=', i)
if srcstr[i] != tarstr[i]: # if does not match return False
return False
return True # if False is not returned yet, it matches

(continues on next page)

184 Chapter 9. Error Handling and Debugging

(continued from previous page)

def findstr(srcstr, tarstr):

'''Find position of srcstr in tarstr'''
for i in range(len(tarstr)):
# if scrstr is same as tarstr from i to rest
# return i
print("calling startswith", srcstr, tarstr[i:])
if startswith(srcstr, tarstr[i:]):
return i
return -1

print(findstr("aba", "abracadabra"))

which outputs the following when run:

calling startswith aba abracadabra

check if aba != abracadabra for i= 0
check if aba != abracadabra for i= 1
check if aba != abracadabra for i= 2
calling startswith aba bracadabra
check if aba != bracadabra for i= 0
calling startswith aba racadabra
check if aba != racadabra for i= 0
calling startswith aba acadabra
check if aba != acadabra for i= 0
check if aba != acadabra for i= 1
calling startswith aba cadabra
check if aba != cadabra for i= 0
calling startswith aba adabra
check if aba != adabra for i= 0
check if aba != adabra for i= 1
calling startswith aba dabra
check if aba != dabra for i= 0
calling startswith aba abra
check if aba != abra for i= 0
check if aba != abra for i= 1
check if aba != abra for i= 2
calling startswith aba bra
check if aba != bra for i= 0
calling startswith aba ra
check if aba != ra for i= 0
calling startswith aba a
check if aba != a for i= 0
check if aba != a for i= 1
---------------------------------------------------------------------------
IndexError Traceback (most recent call last)
<ipython-input-6-b415ffe46a64> in <module>()
19 return -1
20
---> 21 print(findstr("aba", "abracadabra"))

1 frames
<ipython-input-6-b415ffe46a64> in findstr(srcstr, tarstr)
15 # return i
16 print("calling startswith", srcstr, tarstr[i:])
(continues on next page)

9.3. Debugging 185

(continued from previous page)
---> 17 if startswith(srcstr, tarstr[i:]):
18 return i
19 return -1

<ipython-input-6-b415ffe46a64> in startswith(srcstr, tarstr)

4 for i in range(len(srcstr)): # check all characters of srcstr
5 print("check if ", srcstr, '!=', tarstr, 'for i=', i)
----> 6 if srcstr[i] != tarstr[i]: # if does not match return False
7 return False
8 return True # if False is not returned yet, it matches

IndexError: string index out of range

In the output, you can see that execution failed after output “check if aba != a for i= 1”. Therefore,
we can reason that an error occurred at the next step. The following test is:
if srcstr[i] != tarstr[i] for "aba" and "a" for i = 1.
Apparently, getting tarstr[1] for tarstr == "a" fails since the length of tarstr is 1 and the only valid
index is 0. There are different ways to get rid of this error. One quick solution is to test if lengths of srcstr
and tarstr are compatible as len(tarstr) >= len(srcstr) should hold in startswith. The programmer
should have considered this and handled this case. Now, we can correct it as a result of our debugging session.
Of course, debugging output should be added in the correct places with sufficient descriptive information.
If you add too much debugging output, you may be lost in output lines. If there is not sufficient output, you
may not find the error.
For generic tracing of programs with multiple functions, you can use the following Python magic called
decorator, which reports all function calls with parameters when a function is decorated as (you may check
the Colab link and run this interactive example):

def tracedec(f):
def traced(*p, **kw):
print(' ' * tracedec.level + "->", f.__name__,'(',p, kw,')')
tracedec.level += 1
val = f(*p, **kw)
tracedec.level -= 1
print(' ' * tracedec.level + "<-", f.__name__, 'returns ', val)
return val
return traced
tracedec.level = 0

@tracedec
def startswith(srcstr, tarstr):
'''check if tarstr starts with srcstr
like srcstr="abra" tarstr="abracadabra" '''
for i in range(len(srcstr)): # check all characters of srcstr
if srcstr[i] != tarstr[i]: # if does not match, return False
return False
return True # if False is not returned yet, it matches

@tracedec
def findstr(srcstr, tarstr):
'''Find position of srcstr in tarstr'''
for i in range(len(tarstr)):
(continues on next page)

186 Chapter 9. Error Handling and Debugging

(continued from previous page)
# if scrstr is same as tarstr from i to rest
# return i
if startswith(srcstr, tarstr[i:]):
return i
return -1

print(findstr("aba", "abracadabra"))

which outputs the following when run:

-> findstr ( ('aba', 'abracadabra') {} )

-> startswith ( ('aba', 'abracadabra') {} )
<- startswith returns False
-> startswith ( ('aba', 'bracadabra') {} )
<- startswith returns False
-> startswith ( ('aba', 'racadabra') {} )
<- startswith returns False
-> startswith ( ('aba', 'acadabra') {} )
<- startswith returns False
-> startswith ( ('aba', 'cadabra') {} )
<- startswith returns False
-> startswith ( ('aba', 'adabra') {} )
<- startswith returns False
-> startswith ( ('aba', 'dabra') {} )
<- startswith returns False
-> startswith ( ('aba', 'abra') {} )
<- startswith returns False
-> startswith ( ('aba', 'bra') {} )
<- startswith returns False
-> startswith ( ('aba', 'ra') {} )
<- startswith returns False
-> startswith ( ('aba', 'a') {} )
---------------------------------------------------------------------------
IndexError Traceback (most recent call last)
<ipython-input-2-f5cc3f720290> in <module>()
30
31
---> 32 print(findstr("aba", "abracadabra"))

3 frames
<ipython-input-2-f5cc3f720290> in traced(*p, **kw)
3 print(' ' * tracedec.level + "->", f.__name__,'(',p, kw,')')
4 tracedec.level += 1
----> 5 val = f(*p,**kw)
6 tracedec.level -= 1
7 print(' ' * tracedec.level + "<-", f.__name__, 'returns ', val)

<ipython-input-2-f5cc3f720290> in findstr(srcstr, tarstr)

25 # if scrstr is same as tarstr from i to rest
26 # return i
---> 27 if startswith(srcstr, tarstr[i:]):
28 return i
29 return -1

(continues on next page)

9.3. Debugging 187

(continued from previous page)
<ipython-input-2-f5cc3f720290> in traced(*p, **kw)
3 print(' ' * tracedec.level + "->", f.__name__,'(',p, kw,')')
4 tracedec.level += 1
----> 5 val = f(*p,**kw)
6 tracedec.level -= 1
7 print(' ' * tracedec.level + "<-", f.__name__, 'returns ', val)

<ipython-input-2-f5cc3f720290> in startswith(srcstr, tarstr)

15 like srcstr="abra" tarstr="abracadabra" '''
16 for i in range(len(srcstr)): # check all characters of srcstr
---> 17 if srcstr[i] != tarstr[i]: # if does not match return False
18 return False
19 return True # if False is not returned yet, it matches

IndexError: string index out of range

Putting ‘@tracedec’ before function definitions will give you the entry and return states of these functions.
How this decorator works and the ‘@’ syntax are far beyond the scope of this book. You may adapt and use
this example if it suits you.

9.3.2 Handle the Exception to Get More Information

One problem with the run-time errors is that they do not give information about the current state of the
program, basically the set of variables. They give the traceback which includes the line causing the exception
and how program came to that state (which functions are active). In order to get the values of the variables,
you can handle the exception and add log output to the handler as follows (you may check the Colab link
and run this interactive example):

def startswith(srcstr, tarstr):

'''check if tarstr starts with srcstr
like srcstr="abra" tarstr="abracadabra" '''
try:
for i in range(len(srcstr)): # check all characters of srcstr
if srcstr[i] != tarstr[i]: # if does not match return False
return False
except IndexError:
print('Error: srcstr: ', srcstr, ', tarstr:', tarstr, ', i:',i)
raise IndexError # if you like, generate an additional exception
return True # if False is not returned yet, it matches

def findstr(srcstr, tarstr):

'''Find position of srcstr in tarstr'''
for i in range(len(tarstr)):
# if scrstr is same as tarstr from i to rest
# return i
if startswith(srcstr, tarstr[i:]):
return i
return -1

findstr("aba", "abracadabra")

which outputs the following when run:

188 Chapter 9. Error Handling and Debugging

Error: srcstr: aba , tarstr: a , i: 1
---------------------------------------------------------------------------
IndexError Traceback (most recent call last)
<ipython-input-11-b4556283ff77> in startswith(srcstr, tarstr)
5 for i in range(len(srcstr)): # check all characters of srcstr
----> 6 if srcstr[i] != tarstr[i]: # if does not match return False
7 return False

IndexError: string index out of range

During handling of the above exception, another exception occurred:

IndexError Traceback (most recent call last)

2 frames
<ipython-input-11-b4556283ff77> in startswith(srcstr, tarstr)
8 except IndexError:
9 print('Error: srcstr: ',srcstr, ', tarstr:', tarstr, ', i:',i)
---> 10 raise IndexError # if you like generate error again
11 return True # if False is not returned yet, it matches
12

IndexError:

This way, we get the output:

Error: srcstr: aba , tarstr: a , i: 1
which gives us the state of the variables at the moment of the error.

9.3.3 Use Python Debugger

All programming environments come with a debugger software which helps a programmer to execute a
program step by step, observe the current state, add break-points (stops) to inspect, and provide a controlled
execution environment.
For compiled languages, debuggers are external programs getting user’s program as input. For Python, it is
a module called pdb, which stands for Python DeBugger. If you use an integrated development environ-
ment (IDE), a debugger is embedded in the tool so you can use it via the graphical user interface controls.
Otherwise, you can use the command line interface.
In the command line, the only thing you need to do is to put:

import pdb

at the beginning of your program, and type:

pdb.set_trace()

at any place you like to stop execution and go into the debugging mode*. When you run your program
or call a function, your program will start executing and when the execution hits one of the trace points,
execution will stop and a debugger prompt ((Pdb)) will be displayed:

9.3. Debugging 189

> <ipython-input-14-110393975fb5>(7)startswith()
-> for i in range(len(srcstr)): # check all characters of srcstr
(Pdb) h

Documented commands (type help <topic>):

========================================
EOF c d h list q rv undisplay
a cl debug help ll quit s unt
alias clear disable ignore longlist r source until
args commands display interact n restart step up
b condition down j next return tbreak w
break cont enable jump p retval u whatis
bt continue exit l pp run unalias where

Miscellaneous help topics:

==========================
exec pdb

The first line tells which function and line execution have been stopped. When h or help is typed, the
following help lines are displayed. Single-two letter commands are abbreviated forms of the longer ones
(e.g. a for args, b for break, c for cont, cl for clear, …).
After this point, you can use next (n) to execute the program line by line. If the current statement has a
function call, you may choose to go into the function call using the step (s) command. Otherwise, next
will call and return it in single step. During debugging, print (p) command can be used to print content of
a variable. The following is a summary of the useful commands of the debugger:

Command Description
next Execute current line and stop at the next statement, jump over functions
step Execute current line, if there ise function go into it
args show arguments of the current function
break Add a new breakpoint. Execution will stop at that line too
clear Remove the breakpoint(s)
cont Continue execution until hitting a break point
print print current value of the variable
display display variable value whenever it changes in current function
list list program in current line
ll list current function
return continue execution until current function returns
where show currently active functions, which function calls brought code here

The following is an example run (you may check the Colab link and run this interactive example):

import pdb

def startswith(srcstr, tarstr):

'''check if tarstr starts with srcstr
like srcstr="abra" tarstr="abracadabra" '''
pdb.set_trace()
for i in range(len(srcstr)): # check all characters of srcstr
if srcstr[i] != tarstr[i]: # if does not match return False
return False
(continues on next page)

190 Chapter 9. Error Handling and Debugging

(continued from previous page)
return True # if False is not returned yet, it matches

def findstr(srcstr, tarstr):

'''Find position of srcstr in tarstr'''
for i in range(len(tarstr)):
# if scrstr is same as tarstr from i to rest
# return i
if startswith(srcstr, tarstr[i:]):
return i
return -1

print(findstr("aba", "abra"))

which outputs the following when run:

> <ipython-input-18-bf80973a0bf6>(7)startswith()
-> for i in range(len(srcstr)): # check all characters of srcstr
(Pdb) cont
> <ipython-input-18-bf80973a0bf6>(7)startswith()
-> for i in range(len(srcstr)): # check all characters of srcstr
(Pdb) help

Documented commands (type help <topic>):

Miscellaneous help topics:

==========================
exec pdb

(Pdb) cont
> <ipython-input-18-bf80973a0bf6>(7)startswith()
-> for i in range(len(srcstr)): # check all characters of srcstr
(Pdb) cont
> <ipython-input-18-bf80973a0bf6>(7)startswith()
-> for i in range(len(srcstr)): # check all characters of srcstr
(Pdb) cont
---------------------------------------------------------------------------
IndexError Traceback (most recent call last)
<ipython-input-18-bf80973a0bf6> in <module>()
20 return -1
21
---> 22 print(findstr("aba", "abra"))

1 frames
<ipython-input-18-bf80973a0bf6> in findstr(srcstr, tarstr)
16 # if scrstr is same as tarstr from i to rest
17 # return i
(continues on next page)

9.3. Debugging 191

(continued from previous page)
---> 18 if startswith(srcstr, tarstr[i:]):
19 return i
20 return -1

<ipython-input-18-bf80973a0bf6> in startswith(srcstr, tarstr)

5 like srcstr="abra" tarstr="abracadabra" '''
6 pdb.set_trace()
----> 7 for i in range(len(srcstr)): # check all characters of srcstr
8 if srcstr[i] != tarstr[i]: # if does not match return False
9 return False

IndexError: string index out of range

Debuggers are powerful tools especially when you have complex code interacting with many modules and
functions. However, you need time to master and control it. The good news is if you use an integrated
development interface, they are easier to control.

9.4 Important Concepts

We would like our readers to have grasped the following crucial concepts and keywords from this chapter:
• Different types of errors: Syntax, type, run-time and logical errors.
• How to deal with errors.
• Exceptions and exception handling.
• Debugging by “printing” values, exception handling and a debugger.

9.5 Further Reading

• Python’s built-in exceptions: https://docs.python.org/3/library/exceptions.html

• Python’s document on errors and exceptions: https://docs.python.org/3/tutorial/errors.html

192 Chapter 9. Error Handling and Debugging

10 | Scientific and Engineering Libraries

10.1 Numerical Computing with NumPy

NumPy can be considered as a library for working with vectors and matrices. NumPy calls vectors and
matrices as arrays.
Installation Notes
To be able to use the NumPy library, you will need to download it from numpy.org27 and install it on your
computer. If you are using a Python package manager (e.g. pip), you can install it directly using: $ pip
install numpy. If you are using a Windows/Mac machine, you should install anaconda28 first. If you are
using Colab or another Jupyter Notebook viewer, the platform may already have numpy installed.

10.1.1 Arrays and Their Basic Properties

Let us consider a simple vector and a matrix:

array1 = 1 2 3 (10.1.1)

and

1 2 3
array2 = (10.1.2)
4 5 6

Let us see how we can represent and work with these two arrays in NumPy:

>>> import numpy as np # Import the NumPy library

>>> array1 = np.array([1, 2, 3])
>>> array2 = np.array([[1, 2, 3], [4, 5, 6]])
>>> type(array1)
(continues on next page)
27
http://www.numpy.org
28
https://docs.anaconda.com/anaconda/install/

193
(continued from previous page)
<class 'numpy.ndarray'>
>>> type(array2)
<class 'numpy.ndarray'>
>>> array1
array([1, 2, 3])
>>> array2
array([[1, 2, 3],
[4, 5, 6]])
>>> print(array2)
[[1 2 3]
[4 5 6]]

We see from this example that we can pass lists of number as arguments to np.array function which creates
a NumPy array for us. If the argument is a nested list, each element of the list is used as a row of a 2D array.
Arrays can contain any data type as elements; however, we will limit ourselves to numbers (integers and real
numbers) in this chapter.
Shapes, Dimensions and Number of Elements of Arrays. The first thing we can do with a NumPy array
is check its shape. For our example array1 and array2, we can do so as follows:

>>> array1.shape
(3,)
>>> array2.shape
(2,3)

where we can see that array1 is a one-dimensional array with 3 elements and array2 is a 2 × 3 array (a 2D
matrix).
For a 2D array, a shape value (R, C) denotes the number of rows first (R), which is sometimes also called
the first dimension, and then the number of columns (C), which is the second dimension. For nD arrays with
n > 2, the meaning of the shape values is the same except that there are n values in the shape.
In NumPy, we can easily change the shape of an array without losing content:

>>> array1.reshape((3,1))
array([[1],
[2],
[3]])
>>> array2.reshape((1,6))
array([[1, 2, 3, 4, 5, 6]])

For many applications, we will need to access the number of dimensions of an array. For this purpose, we
can use .ndim value:

>>> array1.ndim
1
>>> array2.ndim
2

The number of elements in an array is another important value that we are frequently interested in. To access
that, we can use the .size value:

194 Chapter 10. Scientific and Engineering Libraries

>>> array1.size
3
>>> array2.size
6

Accessing elements in arrays. NumPy allows the same indexing mechanisms that you can use with Python’s
native container data types. For NumPy, let us look at some examples:

>>> array1[-1]
3
>>> array2[1][2]
6
>>> array2[-1]
array([4, 5, 6])

Creating arrays. We have already seen that we can create arrays using the np.array() function. However,
there are other ways for creating arrays conforming to predefined specifications. We can for example create
arrays filled with zeros or ones (note that the argument is a tuple describing the shape of the matrix):

>>> np.zeros((3, 4))

array([[0., 0., 0., 0.],
[0., 0., 0., 0.],
[0., 0., 0., 0.]])
>>> np.ones((2,6))
array([[1., 1., 1., 1., 1., 1.],
[1., 1., 1., 1., 1., 1.]])

Alternatively, we can create an array filled with a range of values using np.arange() function:

>>> np.arange(1,10) # 1: starting value. 10: ending value (excluded)

array([1, 2, 3, 4, 5, 6, 7, 8, 9])
>>> np.arange(1,10,2)
array([1, 3, 5, 7, 9])
>>> np.arange(1,10).reshape((3,3))
array([[1, 2, 3],
[4, 5, 6],
[7, 8, 9]])

10.1.2 Working with Arrays

The previous section covered how we can access elements in an array and properties of an array. Now let us
see how the different types of operations we can do with arrays.
Arithmetic, Relational and Membership Operations with Arrays
The arithmetic operations (+, -, *, /, **), the relational operations (==, <, <=, >, >=) and the membership
operations (in, not in) that we can apply on numbers and other data types in Python can be applied on arrays
with NumPy. These operations are performed elementwise. This means that the arrays that are provided as
operands to a binary arithmetic operator need to have the same shape.
Let us see some examples for arithmetic operations:

10.1. Numerical Computing with NumPy 195

>>> A = np.arange(4).reshape((2,2))
>>> A
array([[0, 1],
[2, 3]])
>>> B = np.arange(4, 8).reshape((2,2))
>>> B
array([[4, 5],
[6, 7]])
>>> print(B-A)
[[4 4]
[4 4]]
>>> print(B+A)
[[ 4 6]
[ 8 10]]
>>> A
array([[0, 1],
[2, 3]])
>>> B
array([[4, 5],
[6, 7]])

Note here that these operations create a new array whose elements are the results of applying the operation.
Therefore, the original arrays are not modified. If you are interested in in-place operations that modify an
existing array during the operation, you can use combined statements such as +=, -=, *=.
Relational and membership operations are also applied elementwise and we can easily anticipate the out-
comes of such operations, e.g. as follows:

>>> A < B
array([[ True, True],
[ True, True]])
>>> B > A
array([[ True, True],
[ True, True]])
>>> A > B
array([[False, False],
[False, False]])
>>> 4 in B
True
>>> 10 in B
False

Useful Functions
NumPy arrays provide several useful functions already provided. These include: - Standard mathematical
function such as exponent, sin, cos, square-root: np.exp(<array>), np.sin(<array>), np.cos(<array>),
np.sqrt(<array>). - Minimum and maximum: <array object>.min() and <array object>.max(). -
Summation, mean and standard deviation: <array object>.sum(), <array object>.mean() and <array
object>.std().
Note that minimum, maximum, summation, mean and standard deviation can be applied on the whole array
as well as along a pre-specified dimension (specified with an axis parameter). Let us see some examples to
clarify this important aspect:

196 Chapter 10. Scientific and Engineering Libraries

>>> A
array([[0, 1],
[2, 3]])
>>> A.sum()
6
>>> A.sum(axis=0)
array([2, 4])
>>> A.sum(axis=1)
array([1, 5])
>>> A.sum(axis=2)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/usr/local/lib/python3.7/site-packages/numpy/core/_methods.py", line 47, in _sum
return umr_sum(a, axis, dtype, out, keepdims, initial, where)
numpy.AxisError: axis 2 is out of bounds for array of dimension 2

where we see that axes start being numbered from zero.

Splitting and Combining Arrays
For many problems, we will need to split an array into multiple arrays or combine multiple arrays into one.
For splitting arrays, we can use functions such as np.hsplit (for horizontal split), np.vsplit (for vertical
split) and np.array_split (for more general split operations). Below is an example for hsplit and vsplit:

>>> L = np.arange(16).reshape(4,4)
>>> L
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11],
[12, 13, 14, 15]])
>>> np.hsplit(L,2) # Divide L into 2 arrays along the horizontal axis
[array([[ 0, 1],
[ 4, 5],
[ 8, 9],
[12, 13]]), array([[ 2, 3],
[ 6, 7],
[10, 11],
[14, 15]])]
>>> np.vsplit(L,2)
[array([[0, 1, 2, 3],
[4, 5, 6, 7]]), array([[ 8, 9, 10, 11],
[12, 13, 14, 15]])]

Note that the resultant arrays are provided as a Python list. Note also that hsplit and vsplit functions
work on even sizes (i.e. they can split into equally sized arrays) – for general split operations, you can use
array_split.
For combining multiple arrays, we can use np.hstack (for horizontal stacking), np.vstack (for vertical
stacking) and np.stack (for more general stacking operations). Below is an example (for the A and B arrays
that we have created before):

>>> A
array([[0, 1],
[2, 3]])
>>> B
(continues on next page)

10.1. Numerical Computing with NumPy 197

(continued from previous page)
array([[4, 5],
[6, 7]])
>>> np.hstack((A, B))
array([[0, 1, 4, 5],
[2, 3, 6, 7]])
>>> np.vstack((A, B))
array([[0, 1],
[2, 3],
[4, 5],
[6, 7]])

Iterations with Arrays.

NumPy library already provides for us many functionalities that we might need while working with arrays.
However, in many circumstances those will not be necessary and we will need to be able to iterate over the
elements of arrays and perform custom algorithmic steps.
Luckily, iteration with arrays is very similar to how we would perform iterations with other container data
types in Python. Let us see some examples:

>>> L
array([[ 0, 1, 2, 3],
[ 4, 5, 6, 7],
[ 8, 9, 10, 11],
[12, 13, 14, 15]])
>>> for r in L:
... print("row: ", r)
...
row: [0 1 2 3]
row: [4 5 6 7]
row: [ 8 9 10 11]
row: [12 13 14 15]

where we see that iteration over a multi-dimensional array iterates over the first dimension. To iterate over
each element, we have at least two options:

>>> for element in L.flat:

... print(element)
...
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
(continues on next page)

198 Chapter 10. Scientific and Engineering Libraries

(continued from previous page)
15
>>> for r in L:
... for element in r:
... print(element)
...
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

Now let us implement one of the operations we have seen above in Python from scratch as an exercise:

import numpy as np

def horizontal_stack(A,B):
"""A function that combines two 2D arrays A and B.
A and B need to have the same height.
"""
# Let us get dimensions first and check whether A and B
# are compatible for stacking.
(H_A, W_A) = A.shape
(H_B, W_B) = B.shape
if H_A != H_B:
print("Arguments A and B have incompatible heights!")
return None
# Let us create an empty `result` array:
H_result = H_A #or H_B
W_result = W_A + W_B
result = np.zeros((H_result, W_result))

# Now let us iterate over each position in A and B and place

# their elements into the corresponding positions
for i in range(H_A):
for j in range(W_A):
result[i][j] = A[i][j]
for j in range(W_B):
result[i][j+W_A] = B[i][j]

return result

# Let us test our code:

M = np.random.randn(2,2) # Create a random 3x4 array
N = np.random.randn(2,3) # Create a random 3x6 array
(continues on next page)

10.1. Numerical Computing with NumPy 199

(continued from previous page)
print("Random array M is:\n", M)
print("Random array N is:\n", N)
print("Horizontal stacking of M and N yields:\n", horizontal_stack(M, N))

Random array M is:

[[ 0.39521366 1.41939618]
[-0.13770429 1.19319949]]
Random array N is:
[[-1.61066212e-01 8.15354855e-01 7.19234023e-01]
[ 2.65985989e-04 -2.89111500e-01 4.52781848e-02]]
Horizontal stacking of M and N yields:
[[ 3.95213656e-01 1.41939618e+00 -1.61066212e-01 8.15354855e-01
7.19234023e-01]
[-1.37704293e-01 1.19319949e+00 2.65985989e-04 -2.89111500e-01
4.52781848e-02]]

10.1.3 Linear Algebra with NumPy

Now let us give a flavour of the Linear Algebra operations provided by NumPy. Note that some of these
operations is provided in the module numpy.linalg.
Transpose.
The transpose operation flips a matrix along the diagonal. For an example matrix A,

1 2
A= (10.1.3)
3 4

its transpose is the following:

T 1 3
A = (10.1.4)
2 4

In NumPy, the transpose of a matrix can be easily accessed by accessing its .T member variable or by calling
its .transpose() member function:

>>> A
array([[1, 2],
[3, 4]])
>>> A.T
array([[1, 3],
[2, 4]])
>>> A
array([[1, 2],
[3, 4]])

Note that .T is simply a member variable of the array object and is not defined as an operation. Try finding the
transpose of A with A.transpose() and see that you obtain the expected result. Note also that the transpose
operation does not change the original array.
Inverse.

200 Chapter 10. Scientific and Engineering Libraries

The inverse of an n × n matrix A is an n × n matrix A−1 that yields an n × n identity matrix I when
multiplied:

A × A−1 = I. (10.1.5)

In NumPy, we can use np.linalg.inv(<array>) to find the inverse of a square array. Here is an example:

>>> A
array([[1, 2],
[3, 4]])
>>> A_inv = np.linalg.inv(A)
>>> A_inv
array([[-2. , 1. ],
[ 1.5, -0.5]])

Let us check whether the inverse was correctly calculated:

>>> np.matmul(A, A_inv)

array([[1.0000000e+00, 0.0000000e+00],
[8.8817842e-16, 1.0000000e+00]])

which is the identity matrix (I) except for some rounding error at index (1, 0) owing to floating point approx-
imation. However, as we have seen in Chapter 2, while writing our programs, we should compare numbers
such that 8.8817842e-16 can be considered to be equal to zero.
Determinant, norm, rank, condition number, trace
While working with matrices, we often require the following properties which can be easily calculated. For
the sake of brevity and to keep the focus, we will omit their explanations:

Matrix Property How to Calculate with NumPy

Determinant (|A| or det(A)) np.linalg.det(A)
Norm (||A||) np.linalg.norm(A)
Rank (rank(A)) np.linalg.matrix_rank(A)
Condition number (κ(A)) np.linalg.cond(A)
Trace (tr(A)) np.trace(A)

Dot Product, Inner Product, Outer Product, Matrix Multiplication

The product of two matrices can be calculated in different ways:
P
• np.dot(a, b): For vectors (1D arrays), this is equivalent to the dot product (i.e. i ai × bi for two
vectors a and b). For 2D arrays and arrays with more dimensions, the result is matrix multiplication.
• np.inner(a,b): For vectors, this is dot product (like np.dot(a,b)). For higher-dimensional arrays,
the result is a sum product over the last axes. Consider the following example for clarification on this:

>>> A
array([[0, 1],
[2, 3]])
>>> B
array([[4, 5],
[6, 7]])
>>> np.inner(A,B)
(continues on next page)

10.1. Numerical Computing with NumPy 201

(continued from previous page)
array([[ 5, 7],
[23, 33]])

The last axes for both A and B are the horizontal axes. Therefore, each first-axis element of A (i.e. [0, 1] and
[2,3]) is multiplied with each first-axis element of B ([4, 5] and [6, 7]).
• np.outer(a,b): Outer product is defined on vectors such that result[i,j] = a[i] * b[j].
• np.matmul(a,b)
P : Matrix multiplication of matrices a and b. The result is simply calculated as
resultij = k aik bkj , as also illustrated in Fig. 10.1.1.

Fig. 10.1.1: Illustration of matrix multiplication of two matrices A and B. (Figure source: Wikipedia)

Eigenvectors and Eigenvalues

Explaining eigenvectors and eigenvalues is beyond the scope of the book. However, we can briefly remind
that eigenvectors and eigenvalues are the variables that satisfy the following for a matrix A which can be
considered as a transformation in a vector space:

Av = λv. (10.1.6)

In other words, v is such a vector that A just changes its scale by a factor λ.
In NumPy, eigenvectors and eigenvalues can be obtained by np.linalg.eigh(<array>) which returns a
tuple with the following two elements:
1. Eigenvalues (an array of λ) in decreasing order.

202 Chapter 10. Scientific and Engineering Libraries

2. Eigenvectors (v) as a column matrix. I.e. v[:, i] is the eigenvector corresponding to the eigenvalue
λ[i].
Matrix Decompositions
Numpy also provides many frequently used matrix decompositions, as listed below:

Matrix Decomposition How to Calculate with NumPy

Cholesky decomposition linalg.cholesky(A)
QR factorization np.linalg.qr(A)
Singular Value Decomposition linalg.svd(a)

Solve a linear system of equations

Given a linear system of equations as follows:

a11 x1 + a12 x2 + ... + a1n xn = b1 ,

a21 x1 + a22 x2 + ... + a2n xn = b2 ,
(10.1.7)
...
an1 x1 + a22 x2 + ... + ann xn = bn ,

which can be represented in compact form as:

ax = b, (10.1.8)

can be solved in NumPy using np.linalg.solve(a, b).

Below is an easy example:

>>> A
array([[0, 1],
[2, 3]])
>>> B
array([3, 4])
>>> np.linalg.solve(A,B)
array([-2.5, 3. ])
>>> X = np.linalg.solve(A,B)
>>> X
array([-2.5, 3. ])

which can be easily verified using multiplication:

>>> np.inner(A, X)
array([3., 4.])

which is equal to B.

10.1. Numerical Computing with NumPy 203

10.1.4 Why Use NumPy? Efficiency Benefits

The previous sections have illustrated how capable the NumPy library is and how easily we can do many
calculations with the pre-defined functionalities e.g. in its linear algebra module. Since we can access each
element of an array using Python’s indexing mechanisms, we can be tempted to implementing those func-
tionalities ourselves, from scratch.
However, those functionalities in NumPy are not implemented in Python but in C, another high-level pro-
gramming language where programs are directly compiled into machine executable binary code. Therefore,
they would execute much faster in comparison to what we would be implementing in Python.
The following example illustrates this. When executed, you should see an order of $~$1000 difference in
running time.
In other words, you are strongly advised, whenever possible, to use NumPy’s existing functions and routines,
and to write every operation in vector or matrix form as much as possible if you are working with matrices.

import numpy as np
from time import time

def matmul_2D(M, N):

"""Custom defined matrix multiplication for two 2D matrices M and N"""
(H_M, W_M) = M.shape
(H_N, W_N) = N.shape
if W_M != H_N:
print("Dimensions of M and N mismatch!")
return None

result = np.zeros((H_M, W_N))

for i in range(H_M):
for j in range(W_N):
for k in range(W_M):
result[i][j] += M[i][k] * N[k][j]

return result

# First let us check that our code works as expected

M = np.random.randn(2,3)
N = np.random.randn(3,4)

print("Our matmul_2D result: \n", matmul_2D(M, N))

print("Correct result: \n", np.matmul(M, N))

# Now let us measure the running-time performances

# Create two 2D large matrices
M = np.random.randn(100, 100)
N = np.random.randn(100, 100)

# Option 1: Use NumPy's matrix multiplication

t1 = time()
result = np.matmul(M, N)
t2 = time()
print("NumPy's matmul took ", t2-t1, "ms.")

# Option 2: Use our matmul_2D function

t1 = time()
(continues on next page)

204 Chapter 10. Scientific and Engineering Libraries

(continued from previous page)
result = matmul_2D(M, N)
t2 = time()
print("Our matmul_2D function took ", t2-t1, "ms.")

Our matmul_2D result:

[[-0.24208963 1.18790239 -1.5397112 -1.14521578]
[-2.00697411 4.23675778 -3.74909068 -0.23854542]]
Correct result:
[[-0.24208963 1.18790239 -1.5397112 -1.14521578]
[-2.00697411 4.23675778 -3.74909068 -0.23854542]]
NumPy's matmul took 0.00040411949157714844 ms.
Our matmul_2D function took 1.0593342781066895 ms.

10.2 Scientific Computing with SciPy

SciPy is a library that includes many methods and facilities for Scientific Computing. It is closely linked
with NumPy so much that NumPy needs to be imported first to be able to use SciPy.
Installation Notes
To be able to use the SciPy library, you will need to download it from scipy.org29 and install it on your
computer. If you are using a Python package manager (e.g. pip), you can install it directly using: $ pip
install scipy. If you are using a Windows/Mac machine, you should install anaconda30 first. If you are
using Colab or another Jupyter Notebook viewer, the platform may already have scipy installed.
Below is a list of some modules provided by SciPy. Even a brief coverage of these modules is not feasible
in a chapter. However, we will see an example in Chapter 12 using SciPy’s stats and optimize modules.

Module Description
cluster Clustering algorithms
constants Physical and mathematical constants
fftpack Fast Fourier Transform routines
integrate Integration and ordinary differential equation solvers
interpolate Interpolation and smoothing splines
io Input and Output
linalg Linear algebra
ndimage N-dimensional image processing
odr Orthogonal distance regression
optimize Optimization and root-finding routines
signal Signal processing
sparse Sparse matrices and associated routines
spatial Spatial data structures and algorithms
special Special functions
stats Statistical distributions and functions
29
http://www.scipy.org
30
https://docs.anaconda.com/anaconda/install/

10.2. Scientific Computing with SciPy 205

10.3 Data handling & analysis with Pandas

Pandas is a very handy library for working with files with different formats and analyzing data in different
format. To keep things simple, in this section we will just look at CSV files. Note that the facilities for other
formats are very similar.
Installation Notes
To be able to use the Pandas library, you will need to download it from pandas.pydata.org31 and install it on
your computer. If you are using a Python package manager (e.g. pip), you can install it directly using: $ pip
install pandas. If you are using a Windows/Mac machine, you should install anaconda32 first. If you are
using Colab or another Jupyter Notebook viewer, the platform may already have pandas installed.

10.3.1 Supported File Formats

As we have seen before, Python already provides facilities for reading and writing files. However, if you
need to work with “structured” files such as CSV, XML, XLSX, JSON, HTML, the native facilities of Python
would need to extended significantly.
That’s where Pandas comes in. It complements Python’s native facilities to be able to work with structured
files for both reading data from them and creating new files. Below is a list of the different file formats that
Pandas supports and the corresponding functions for reading or writing them.
Table 10.1. The file formats supported by the Pandas library. The table is adapted from and the reader is
encouraged to check the following page to see a more up-to-date list: Pandas IO Reference.
31
https://pandas.pydata.org/
32
https://docs.anaconda.com/anaconda/install/

206 Chapter 10. Scientific and Engineering Libraries

Format Type Data Description Reader Writer
text CSV read_csv to_csv
text Fixed-Width Text File read_fwf •

text JSON read_json to_json

text HTML read_html to_html
text Local clipboard read_clipboard to_clipboard
• MS Excel read_excel to_excel

binary OpenDocument read_excel •

binary HDF5 Format read_hdf to_hdf

binary Feather Format read_feather to_feather
binary Parquet Format read_parquet to_parquet
binary ORC Format read_orc •

binary Msgpack read_msgpack to_msgpack

binary Stata read_stata to_stata
binary SAS read_sas •

binary SPSS read_spss •

binary Python Pickle Format read_pickle to_pickle

SQL SQL read_sql to_sql
SQL Google BigQuery read_gbq to_gbq

10.3.2 Data Frames

Pandas library is built on top of a data type called DataFrame which is used to store all types of data while
working with Pandas. In the following, we will illustrate how you can get your data into a DataFrame object:
1. Loading data from a file. If you have your data in a file that is supported by Pandas, you can use its
reader for directly obtaining a DataFrame object. Let us look at an example CSV file:

# Download an example CSV file:

!wget -nc https://raw.githubusercontent.com/sinankalkan/CENG240/master/figures/ch10_example.
,→csv

# Import the necessary libraries

import pandas as pd

# Read the file named 'ch10_example.csv'

df = pd.read_csv('ch10_example.csv')

# Print the CSV file's contents:

print("The CSV file contains the following:\n", df, "\n")

(continues on next page)

10.3. Data handling & analysis with Pandas 207

(continued from previous page)
# Check the types of each column
df.dtypes

--2021-06-21 20:22:24-- https://raw.githubusercontent.com/sinankalkan/CENG240/master/

,→figures/ch10_example.csv
Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.109.133, 185.199.
,→110.133, 185.199.108.133, ...
Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.109.133|:443...␣
,→connected.

HTTP request sent, awaiting response... 200 OK

Length: 162 [text/plain]
Saving to: ‘ch10_example.csv’

ch10_example.csv 100%[===================>] 162 --.-KB/s in 0s

2021-06-21 20:22:25 (2.53 MB/s) - ‘ch10_example.csv’ saved [162/162]

The CSV file contains the following:

Name Grade Age
0 Jack 40.2 20
1 Amanda 30.0 25
2 Mary 60.2 19
3 John 85.0 30
4 Susan 70.0 28
5 Bill 58.0 28
6 Jill 90.0 27
7 Tom 90.0 24
8 Jerry 72.0 26
9 George 79.0 22
10 Elaine 82.0 23

Name object
Grade float64
Age int64
dtype: object

Note the following critical details:

• read_csv file automatically understood that our CSV file had a header (“Name”, “Grade” and “Age”).
If your file does not have a header, you can call read_csv with the header parameter set to None as
follows: pd.read_csv(filename, header=None).
• read_csv read all columns in the CSV file. If you wish to load only some of the columns
(e.g. ‘Name’, ‘Age’ in our example), you can relay this using the usecols parameter as follows:
pd.read_csv(filename, usecols=[ 'Name', 'Age']).
2. Convert Python data into a ``DataFrame``. Alternatively, you can have already a Python data object
which you can provide as argument to a DataFrame constructor as illustrated with the following example:

lst = [('Jack', 40.2, 20), ('Amanda', 30, 25), ('Mary', 60.2, 19)]
df = pd.DataFrame(data = lst, columns=['Name', 'Grade', 'Age'])
print(df)

208 Chapter 10. Scientific and Engineering Libraries

Name Grade Age
0 Jack 40.2 20
1 Amanda 30.0 25
2 Mary 60.2 19

In many cases, we will require the rows to be associated with names, or sometimes called as keys. For
example, instead of referring to a row as “the row at index 1”, we might require accessing a row with a
non-integer value. This can be achieved as follows (note how the printed DataFrame looks different):

names = ['Jack', 'Amanda', 'Mary']

lst = [(40.2, 20), (30, 25), (60.2, 19)]
df = pd.DataFrame(data = lst, index=names, columns=['Grade', 'Age'])
print(df)

Grade Age
Jack 40.2 20
Amanda 30.0 25
Mary 60.2 19

Alternatively, we can obtain a DataFrame from a dictionary, which might be easier for us to create data
column-wise – note that the column names are obtained from the keys of the dictionary:

d = {'Grade': [40.2, 30, 60.2],

'Age': [20, 25, 19]}
names = ['Jack', 'Amanda', 'Mary']
df = pd.DataFrame(data = d, index=names)
print(df)

Grade Age
Jack 40.2 20
Amanda 30.0 25
Mary 60.2 19

10.3.3 Accessing Data with DataFrames

We can access data columnwise or row-wise.

1. Columnwise access.
For columnwise access, you can use df["Grade"], which returns a sequence of values. To access a certain
element in that sequence, you can either use an integer index (as in df["Grade"][1]) or named index if it
has been defined (as in df["Grade"]["Amanda"]). The following example illustrates this for the example
DataFrame we have created above:

>>> print(df)
Grade Age
Jack 40.2 20
Amanda 30.0 25
Mary 60.2 19
>>> print(df['Grade'][1])
30.0
(continues on next page)

10.3. Data handling & analysis with Pandas 209

(continued from previous page)
>>> print(df['Grade']['Amanda'])
30.0

2. Rowwise access.
To access a particular row, you can either use integer indexes with df.iloc[<row index>] or df.loc[<row
name>] if a named index is defined. This is illustrated below:

>>> print(df)
Grade Age
Jack 40.2 20
Amanda 30.0 25
Mary 60.2 19
>>> print(df.iloc[1])
Grade 30.0
Age 25.0
Name: Amanda, dtype: float64
>>> print(df.loc['Amanda'])
Grade 30.0
Age 25.0
Name: Amanda, dtype: float64

You can also provide both column index and name index in a single operation, i.e. [<row index>, <column
index>] or [<row name>, <column name>] as illustrated below:

>>> df.loc['Amanda','Grade']
30.0
>>> df.iloc[1, 1]
25

While accessing a DataFrame with integer indexes, you can also use Python’s slicing;
i.e. [start:end:step] indexing.

10.3.4 Modifying Data with DataFrames

Although modifying data in a DataFrame is simple, you need to be careful about one crucial concept: Let
us say you use column and row names to change the contents of a cell and you access first the column and
then the corresponding row as follows:

>>> print(df)
Grade Age
Jack 40.2 20
Amanda 30.0 25
Mary 60.2 19
>>> df['Grade']['Amanda'] = 45
/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:6: SettingWithCopyWarning:
A value is trying to be set on a copy of a slice from a DataFrame
See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_
,→guide/indexing.html#returning-a-view-versus-a-copy

This is called chained indexing and Pandas will give you a warning that you are modifying a DataFrame
that was returned while accessing columns or rows in your original DataFrame: df['Grade'] returns a

210 Chapter 10. Scientific and Engineering Libraries

direct access for the original Grade column or a copy to it and depending on how your data is structured and
stored in the memory, the end result may be different and your change might not be reflected on the original
DataFrame.
Instead, you should use a single access operation as illustrated below (you may do this with loc and iloc
as well):

>>> print(df)
Grade Age
Jack 40.2 20
Amanda 30.0 25
Mary 60.2 19
>>> df.loc['Amanda','Grade'] = 45
>>> df.iloc[1,1] = 30
>>> print(df)
Grade Age
Jack 40.2 20
Amanda 45.0 30
Mary 60.2 19

With these facilities, you can now access every item in a DataFrame and modify them.

10.3.5 Analyzing Data with DataFrames

Once we have our data in a DataFrame, we can use Pandas’s built-in facilities for analyzing our data. One
very simple way to analyze data is via descriptive statistics, which you can access with df.describe() and
df[<column>].value_counts():

print(df)
df.describe()

Grade Age
Jack 40.2 20
Amanda 30.0 25
Mary 60.2 19

Apart from these descriptive functions, Pandas provides functions for sorting (using .sort_values() func-
tion), finding the maximum or the minimum (using .max() or .min() functions) or finding the largest or
smallest n values (using .nsmallest() or .nlargest() functions:

print("Maximum grade is: ", df['Grade'].max())

print("\nRecords sorted according to age:\n", df.sort_values(by="Age"))
print("\n\nTop two grades are:\n", df['Grade'].nlargest(2))

Maximum grade is: 60.2

Records sorted according to age:

Grade Age
Mary 60.2 19
Jack 40.2 20
Amanda 30.0 25

(continues on next page)

10.3. Data handling & analysis with Pandas 211

(continued from previous page)

Top two grades are:

Mary 60.2
Jack 40.2
Name: Grade, dtype: float64

Note that, the displayed output includes the names because we selected names as the indices for accessing
the rows.

10.3.6 Presenting Data in DataFrames

Pandas provides a very easy mechanism for plotting your data via .plot() function. Examples are provided
below to illustrate how you can plot your own data. This plotting facility is provided by the Matplotlib
which we will see next. If you are not using Colab, you need to use the following method to show the plot:
df.plot().figure.show()

#Plots all columns in different colours

df.plot()

<AxesSubplot:>

# Plots a single column

df['Age'].plot()

<AxesSubplot:>

212 Chapter 10. Scientific and Engineering Libraries

10.4 Plotting data with Matplotlib

Matplotlib is a very capable library for drawing different types of plots in Python. It is very well integrated
with Numpy, Scipy and Pandas, and therefore, all these libraries are very frequently used together seamlessly.
Installation Notes
To be able to use the matplotlib library, you will need to download it from matplotlib.org33 and install it on
your computer. If you are using a Python package manager (e.g. pip), you can install it directly using: $ pip
install matplotlib. If you are using a Windows/Mac machine, you should install anaconda34 first. If you
are using Colab or another Jupyter Notebook viewer, the platform may already have matplotlib installed.

33
https://matplotlib.org/
34
https://docs.anaconda.com/anaconda/install/

10.4. Plotting data with Matplotlib 213

10.4.1 Parts of a Figure

A figure consists of the following elements (see also Fig. 10.4.1): - Title of the figure. - Axes, together with
their ticks, tick labels, and axis labels. - The canvas of plot, which consists of a drawing of your data in
the form of a dots (scatter plot), lines (line plot), bars (bar plot), surfaces etc. - Legend, which informs the
perceiver about the different plots in the canvas.

Fig. 10.4.1: A figure consists of several components all of which you can change in matplotlib easily. Figure
source: Matplotlib Usage Guides.

10.4.2 Preparing your Data for Plotting

Matplotlib expects numpy arrays as input and therefore, if you have your data in a numpy array, you can
directly plot it without any data type conversion. With pandas DataFrame, the behavior is not guaranteed
and therefore, it is recommended that the values in a DataFrame are first converted to a numpy array, using
e.g.:

print(df)
age_array = df['Age'].values
print("The `Age` values in an array form are:", age_array)
print("The type of our new data is: ", type(age_array))

214 Chapter 10. Scientific and Engineering Libraries

Grade Age
Jack 40.2 20
Amanda 30.0 25
Mary 60.2 19
The Age values in an array form are: [20 25 19]
The type of our new data is: <class 'numpy.ndarray'>

10.4.3 Drawing Single Plots

There are two ways to plot with matplotlib: In an object-oriented style or the so-called Pyplot style:
1. Drawing in an Object-Oriented Style.
In this style, we create a figure object and an axes object and work with those to create our plots. This is
illustrated in the following example:

import matplotlib.pyplot as plt

import numpy as np

# Uniformly sample 50 x values between -2 and 2:

x = np.linspace(-2, 2, 50)

# Create an empty figure

fig, ax = plt.subplots()

# Plot y = x
ax.plot(x, x, label='$y=x$')

# Plot y = x^2
ax.plot(x, x**2, label='$y=x^2$')

# Plot y = x^3
ax.plot(x, x**3, label='$y=x^3$')

# Set the labels for x and y axes:

ax.set_xlabel('x')
ax.set_ylabel('y')

# Set the title of the figure

ax.set_title("Our First Plot -- Object-Oriented Style")

# Create a legend
ax.legend()

# Show the plot

# fig.show() # Uncomment if not using Colab

<matplotlib.legend.Legend at 0x125532250>

10.4. Plotting data with Matplotlib 215

2. Drawing in a Pyplot Style. In the Pyplot style, we directly call functions in the pyplot module
(matplotlib.pyplot) to create a figure and draw our plots. This style does not work with an explicit
figure or axes objects. This is illustrated in the following example:

# Uniformly sample 50 x values between -2 and 2:

x = np.linspace(-2, 2, 50)

# Plot y = x
plt.plot(x, x, label='$y=x$')

# Plot y = x^2
plt.plot(x, x**2, label='$y=x^2$')

# Plot y = x^3
plt.plot(x, x**3, label='$y=x^3$')

# Set the labels for x and y axes:

plt.xlabel('x')
plt.ylabel('y')

# Set the title of the figure

plt.title("Our First Plot -- Pyplot Style")

# Create a legend
plt.legend()

# Show the plot

#plt.show() # Uncomment if not using Colab

<matplotlib.legend.Legend at 0x1256ab9a0>

216 Chapter 10. Scientific and Engineering Libraries

10.4.4 Drawing Multiple Plots in a Figure

In many situations, you will need to draw multiple plots side by side in a single figure. This can be performed
in the object-oriented style using the subplots() function to create a grid and then use the created subplots
and axes to draw the plots. This is illustrated with an example below:

# Create a 2x2 grid of plots

fig, axes = plt.subplots(2, 2)

# Plot (1,1)
axes[0,0].plot(x, x)
axes[0,0].set_title("$y=x$")

# Plot (1,2)
axes[0,1].plot(x, x**2)
axes[0,1].set_title("$y=x^2$")

# Plot (2,1)
axes[1,0].plot(x, x**3)
axes[1,0].set_title("$y=x^3$")

# Plot (2,2)
axes[1,1].plot(x, x**4)
axes[1,1].set_title("$y=x^4$")

# Adjust vertical space between rows

plt.subplots_adjust(hspace=0.5)

# Show the plot

#fig.show() # Uncomment if not using Colab

10.4. Plotting data with Matplotlib 217

It is possible to do this in the PyPlot style as well, as illustrated below:

# Plot (1,1)
plt.subplot(2, 2, 1)
plt.plot(x, x)
plt.title('$y=x$')

# Plot (1,2)
plt.subplot(2, 2, 2)
plt.plot(x, x**2)
plt.title('$y=x^2$')

# Plot (2,1)
plt.subplot(2, 2, 3)
plt.plot(x, x**3)
plt.title('$y=x^3$')

# Plot (2,2)
plt.subplot(2, 2, 4)
plt.plot(x, x**4)
plt.title('$y=x^4$')

# Adjust vertical space between rows

plt.subplots_adjust(hspace=0.5)

# Show the plot

#plt.show() # Uncomment if not using Colab

218 Chapter 10. Scientific and Engineering Libraries

10.4.5 Changing elements of a plot

All elements that are visualized in Figure 10.4.1 can be changed in matplotlib. We will skip these details
to keep our focus in the book to the practical uses of these libraries. The interested reader can look up the
extensive documentation at https://matplotlib.org/2.1.1/contents.html or look at the help page of a function
(e.g. help(plt.plot)) to see how to modify all elements of a figure.

10.5 Important Concepts

We would like our readers to have grasped the following crucial concepts and keywords from this chapter:
• NumPy arrays and their properties: array shape, dimensions, sizes, elements.
• Accessing and modifying elements of a NumPy array.
• Simple algebraic functions on NumPy arrays.
• SciPy and its basic capabilities.
• Pandas, DataFrame, loading files with Pandas.
• Accessing and modifying content in DataFrames.
• Analyzing and presenting data in DataFrames.
• Matplotlib and different ways to make plots.
• Drawing single and multiple plots. Changing elements of a plot.

10.5. Important Concepts 219

10.6 Further Reading

• NumPy documentation: https://numpy.org/doc/

• SciPy documenation: https://docs.scipy.org/doc/scipy/reference/
• Pandas documentation: https://pandas.pydata.org/docs/
• Matplotlib documentation: https://matplotlib.org/2.1.1/contents.html
• Introduction to Algebra with Python: https://pabloinsente.github.io/intro-linear-algebra

10.7 Exercises

• Define functions that work like the sum, mean, min and max operations provided by NumPy. These
functions should take a single 2D array and return the result as a number. You can assume that the
operation applies to the whole array and not to a single axis.
• Create a simple CSV file using your favorite spreadsheet editor (e.g. Microsoft Excel or Google
Spreadsheets) and create a file with your exams and their grades as two separate columns. Save the
file, upload it to the Colab notebook and do the following:
– Load the file using Pandas.
– Calculate the mean of your exam grades.
– Calculate the standard deviation of your grades.
• Using Matplotlib, generate the following plots with suitable names for the axes and the titles.
– Draw the following four functions in separate single plots: sin(x), cos(x), tan(x), cot(x).
– Draw these four functions in a single plot.
– Draw a multiple 2x2 plot where each subplot is one of the four functions.

220 Chapter 10. Scientific and Engineering Libraries

11 | An Application: Approximation and
Optimization

11.1 Approximating Functions with Taylor Series

In some engineering or scientific problems, we have limited access to a function: We might be provided
only the value of the function or its derivatives at certain input values and we might be required to make
estimations (approximations) about the values of the function at other input values. Taylor’s series is one
method that provides us a very elegant solution for approximating a function.
The Taylor series of a function f (·) is essentially an infinite sum of terms that converges to the value of the
function at x, i.e. f (x). In more formal terms:

f ′ (a) f ′′ (a) f ′′′ (a)

f (x) = f (a) + (x − a) + (x − a)2 + (x − a)3 ... (11.1.1)
1! 2! 3!
which can be re-written in a more compact form:
∞
X f (n) (a)
f (x) = (x − a)n . (11.1.2)
n!
n=0

In the limit (n → ∞), f (x) will be equal to its Taylor series. However, in realistic applications, it is
impractical to take n → ∞. In practice, we take the first m terms of the series and approximate the function
with this:
f ′ (a) f ′′ (a) f ′′′ (a) f (m) (a)
f (x) ≈ f (a) + (x − a) + (x − a)2 + (x − a)3 ... + (x − a)m , (11.1.3)
1! 2! 3! m!
ignoring the terms for n > m.
Note that the denominator (n!) gets large values quickly when n gets large. Therefore, higher-order terms in
the Taylor series are likely to be very small values. For this reason, taking the first m terms in the series do
provide sufficient approximations to f (x) in practice.

221
We can provide some intuition for the Taylor Series as follows: If we know the value of function f at x = a,
then we can calculate the value of the function at any other value of x as long as we know all (or the first
m) derivatives of the function at x = a, i.e. f (n) (a) for n = 1, ..., ∞ (or m). To see this, consider just the
′
Taylor Series approximation with the first two terms: f (x) ≈ f (a) + f 1!(a) (x − a), which is illustrated in
Fig. 11.1.1.

Fig. 11.1.1: Taylor Series approximates a function by using its derivatives. The drawing illustrates f (x0 )
being approximated (with the first term of the Taylor Series only) by the known value of the function f (a)
and the derivative f ′ (a) at that point. Green line denotes the tangent line at f (a) whose slope is f ′ (a).

11.1.1 Taylor Series Example in Python

Let us take a “simple” function, namely f (x) = x3 , and approximate it with a few terms of the Taylor Series
expansion:

# x^3 function
def xcube_a(a): return a*a*a

# Function for the derivatives of x^3

# Inputs: n (integer) => derivative degree
# a (real number) => constant at which derivative is calculated
def xcube_nth_deriv_at_a(n, a):
if n == 1: return 3*a*a # 1st derivative: 3x*x
if n == 2: return 6*a # 2nd derivative: 6x
if n == 3: return 6 # 3rd derivative: 6
if n > 3: return 0 # nth derivative (n>3): 0

# Factorial function
def fact(n): return 1 if n < 1 else n*fact(n-1)

# Function approximating x^3 with m terms of the Taylor series

def xcube_approx(x, m):
a = 1 # You can try different a values
result = xcube_a(a)
for n in range(1, m+1):
result += xcube_nth_deriv_at_a(n, a)/fact(n)*pow(x-a, n)
(continues on next page)

222 Chapter 11. An Application: Approximation and Optimization

(continued from previous page)

return result

# Now let us evaluate how close our approximation is for various values of m.
import numpy as np
import matplotlib.pyplot as plt

# Let us compare the approximations for a set of x values:

x_vector=np.arange(-5, 5, 0.1)

# Find the approximations for our x values:

y_m_1=[xcube_approx(x,1) for x in x_vector]
y_m_2=[xcube_approx(x,2) for x in x_vector]
y_m_3=[xcube_approx(x,3) for x in x_vector]

# We have our ground truth (correct) values:

y_correct = [x*x*x for x in x_vector]

# Let's plot the results and the correct function (x**3)

# We have a multi-plot plot, so we will use subplot() function
plt.rcParams['figure.figsize'] = 15, 4 # Sets the figure size

# First subplot: The correct (original) function that we are approximating

plt.subplot(1, 4, 1)
plt.plot(x_vector,y_correct, label='x**3')
plt.title('$x^3$')
plt.xlabel('$x$')

# Second subplot: Approximation with one term

plt.subplot(1, 4, 2)
plt.plot(x_vector,y_m_1)
plt.title("$x^3$ Approx. with $m$=1")
plt.xlabel('$x$')

# Third subplot: Approximation with two terms

plt.subplot(1, 4, 3)
plt.plot(x_vector,y_m_2)
plt.title("$x^3$ Approx. with $m$=2")
plt.xlabel('$x$')

# Fourth subplot: Approximation with three terms

plt.subplot(1, 4, 4)
plt.plot(x_vector,y_m_3)
plt.title("$x^3$ Approx. with $m$=3")
plt.xlabel('$x$')

Text(0.5, 0, '$x$')

11.1. Approximating Functions with Taylor Series 223

11.2 Finding the Roots of a Function

Let us assume that we have a non-linear, continuous function f (x) that intersects the horizontal axis as in
Fig. 11.2.1. We are interested in finding r0 , ..., rn such that f (ri ) = 0. We call these intersections (r0 , r1 , r2
in Fig. 11.2.1) the roots of function f ().
For many problems in Engineering and Mathematics, finding the values of these intersections is very useful.
For example, many problems requiring solutions to equations like g(x) = h(x) can be reformulated as a
root finding problem for g(x) − h(x) = 0.

Fig. 11.2.1: Finding the roots of a function f (x) means finding r0 , .., rn for which the function is zero,
f (ri ) = 0.

224 Chapter 11. An Application: Approximation and Optimization

11.2.1 Newton’s Method for Finding the Roots

Newton’s method is one of the many methods for finding the roots of a function. It is a very common method
that randomly starts at an x0 value and iteratively takes one step at a time towards a root. The method can
be described with the following iterative steps:
Step 1: Set an iteration variable, i, to 0. Initialize xi randomly; however, if you have a good guess, it is
always better to initialize xi with it. For our example function f (x) in Fig. 11.2.1, see the selected xi in Fig.
11.2.2.

Fig. 11.2.2: One initial step of Newton’s method. The tangent line at f (x0 ) is used to calculate the next
value of xi , getting us closer to a root.

Step 2: Find the intersection of the tangent line at xi with the horizontal axis. The tangent line to the function
at xi is illustrated as the green dashed line in Figure 11.3. This line can be easily derived using xi and f ′ (xi ).
Let us use xi+1 to denote the intersection of the tangent line with the horizontal axis. Using the definition of
the derivative and simple geometric rules, we can easily show that xi+1 satisfies the following:
xi − xi+1 1
= ′ , (11.2.1)
f (xi ) f (xi )

with which we can formulate how we can find xi+1 as follows:

f (xi )
xi+1 = xi − . (11.2.2)
f ′ (xi )

Step 3: Repeat Step 2 by replacing xi ← xi+1 until “convergence”.

The intuition behind the Newton’s method is simple: For a linear function such as y = ax + b, the root
is r = −b/a, which can be easily derived from the slope and a known y value. For a non-linear function
y = f (x), we make an approximation by assuming f () to be linear at point x0 and use the slope of this linear
approximation to find a new x0 that is closer to the root.

11.2. Finding the Roots of a Function 225

11.2.2 Misc Details on Newton’s Method for the Curious

It is not possible to cover all aspects of these elegant methods in detail. However, it would be a shame to
skip the following details and hide them from a curious mind.
The Taylor Series Perspective for Newton’s Method
Newton’s method for finding a root of a function f () works iteratively starting from an initial value x0 by
using the first-order expansion of f () around xi :
f ′ (xi )
f (xi+1 ) ≈ f (xi ) + (xi+1 − xi ), (11.2.3)
1!
which can be simplified by rewriting xi+1 = xi + δ:
f ′ (xi )
f (xi + δ) ≈ f (xi ) + δ. (11.2.4)
1!
We are interested in finding δ such that the approximation is equal to zero, in other words:
f ′ (xi )
0 = f (xi ) + δ, (11.2.5)
1!
which yields the delta value for which the approximation is zero:
f (xi )
δ=− . (11.2.6)
f ′ (xi )
Plugging this into xi+1 = xi + δ, we obtain the equation for calculating the next value in the sequence
towards a minimum of function f ():
f (xi )
xi+1 ← xi − . (11.2.7)
f ′ (xi )
Notes on Convergence
A careful reader should have identified two important issues with Newton’s method:
1- How can we be sure that this method converges to a solution, i.e. a root?
If function f () satisfies certain conditions, then we can guarantee that Newton’s method converges to a root.
The details of these conditions and the proofs are beyond the scope of the book and the interested reader can
check e.g. Wikipedia35 .
To be on the safe side, we can limit the maximum number of steps that can we take and use the calculated
value at the end.
2- If convergence can be guaranteed, how can we know that we have converged to a solution?
A common technique in such iterative algorithms is to check whether the difference between the consecutive
values is tiny. In our case, if |xi+1 − xi | < ϵ, where ϵ is a very small number, then we can assume that
consecutive steps do not lead to much change in xi and therefore, we can stop our algorithm. Of course, the
value of ϵ will be critical in determining how many iterations are needed to ‘converge’ to a result.
Notes on Multiple Roots
Newton’s method converges only to a single root based on the initial value x0 . In order to find other roots,
the algorithm needs to be executed with a different value of x0 , which will hopefully lead to a different root
than the ones that we have identified before.
Since our knowledge about f () or its roots is limited, running the algorithm for N times for different values
of x0 is the only option. If, on the other hand, we knew roughly the ranges of x where function f () might
have roots, then we could use this knowledge to choose better x0 values.
35
https://en.wikipedia.org/wiki/Newton%27s_method#Analysis

226 Chapter 11. An Application: Approximation and Optimization

11.2.3 Newton’s Method in Python

Let us implement Newton’s method in Python and try to find the roots of an example function.

def find_root(f, f_deriv, x_0, max_steps=10, delta=0.001, verbose=True):

"""
Use Newton's method to find a root of function f, given its derivative f_deriv.
Inputs: f => function
f_deriv => derivative of function f
x_0 => initial x value
max_steps => maximum number of iterative steps (default: 10)
delta => if |x_i+1 - x_i| < delta, then assume convergence
"""
x_old = x_0
for i in range(max_steps):
# Step 2
try:
x_new = x_old - f(x_old) / f_deriv(x_old)
except ZeroDivisionError:
# Encountered division-by-zero, return None
return None

if verbose: print("Iteration ", i, ". (x_old, x_new): ", (x_old, x_new), " |x_new-x_
,→ old|: ", abs(x_new-x_old))
if abs(x_new-x_old) < delta: break

# Step 3
x_old = x_new

return x_new

# Let us pick a function and test our implementation

# Assume that our function is: f(x) = x^2 - 4

def f(x_i): return x_i**2-4

# This is the derivative, f'(x_i)

def f_deriv(x_i): return 2*x_i

# Let us draw our function first

import matplotlib.pyplot as plt
import numpy as np

def draw_f():
# Uniformly sample 50 x values between -5 and 5:
x = np.linspace(-3, 3, 50)

plt.rcParams['figure.figsize'] = 5, 4

plt.plot(x, f(x))
plt.title('$f(x)=x^2 - 4$')

draw_f()

11.2. Finding the Roots of a Function 227

# Let us test our solution with 10 random initial values
import random

draw_f()
for i in range(10):
x_0 = random.randint(-5, 5) # A random integer in [-5, 5]
r = find_root(f, f_deriv, x_0, max_steps=50, delta=0.0001, verbose=False)
if (not r) or (abs(f(r)) > 0.001): continue # Diverged or divison-by-zero, ignore
print("trial", i, " ended with root: ", r, " value at root, f(r): ", f(r))
plt.scatter(r, f(r))

trial 0 ended with root: 2.000000000000002 value at root, f(r): 8.881784197001252e-15

trial 1 ended with root: -2.000000000000002 value at root, f(r): 8.881784197001252e-15
trial 2 ended with root: 2.0 value at root, f(r): 0.0
trial 3 ended with root: 2.000000000026214 value at root, f(r): 1.0485656787295738e-10
trial 4 ended with root: 2.000000000026214 value at root, f(r): 1.0485656787295738e-10
trial 5 ended with root: -2.000000000000002 value at root, f(r): 8.881784197001252e-15
trial 6 ended with root: -2.000000000000002 value at root, f(r): 8.881784197001252e-15
trial 7 ended with root: 2.000000000026214 value at root, f(r): 1.0485656787295738e-10
trial 8 ended with root: 2.000000000000002 value at root, f(r): 8.881784197001252e-15
trial 9 ended with root: 2.0 value at root, f(r): 0.0

228 Chapter 11. An Application: Approximation and Optimization

11.2.4 Newton’s Method in SciPy

Let us see how we can use Newton’s method from the SciPy library (from Chapter 10) to find the roots of
a function. Note that we do not need to provide the derivative of the function to SciPy. It used numerical
differentiation to calculate the derivative for us.

from scipy.optimize import newton

import random

draw_f() # Defined above

for i in range(10):
x_0 = random.randint(-5, 5) # A random integer in [-5, 5]
r = newton(f, x_0, fprime=None, args=(), tol=0.0001, maxiter=50)
if (not r) or (abs(f(r)) > 0.001): continue # Diverged or divison-by-zero, ignore
print("trial", i, " ended with root: ", r, " value at root, f(r): ", f(r))
plt.scatter(r, f(r))

trial 0 ended with root: 2.000000075022694 value at root, f(r): 3.000907806693931e-07

trial 1 ended with root: -2.0000000032851974 value at root, f(r): 1.3140789789645169e-08
trial 2 ended with root: 2.000000075022694 value at root, f(r): 3.000907806693931e-07
trial 3 ended with root: 2.0000000032851974 value at root, f(r): 1.3140789789645169e-08
trial 5 ended with root: -2.0 value at root, f(r): 0.0
trial 6 ended with root: -2.0000000003832255 value at root, f(r): 1.532901805489928e-09
trial 7 ended with root: 2.000000075022694 value at root, f(r): 3.000907806693931e-07
trial 8 ended with root: -2.0000000003832255 value at root, f(r): 1.532901805489928e-09
trial 9 ended with root: -2.0000000032851974 value at root, f(r): 1.3140789789645169e-08

11.2. Finding the Roots of a Function 229

11.3 Finding a Minimum of Functions

An important problem frequently encountered in many disciplines is optimization. In such problems, as

illustrated in Fig. 11.3.1, we have a function f (x) and we wish to find its minimum value (f ∗ ):

f ∗ ← min f (x), (11.3.1)

x∈R

or the value (x∗ ) that minimizes it:

x∗ ← arg min f (x). (11.3.2)

x∈R

In many practical settings, the functions that we work with may have multiple minima. In such a case, the
minimum point, x∗ , can be either (i) a global minimum such that there may not be any other point x̂ with
f (x̂) < f (x∗ ), (ii) a local minimum such that x∗ is the minimum of a local neighborhood.
The methods generally exploit two important cues: (i) At a minimum, the sign of the first-order derivative
changes (on the sides of the minimum). (ii) At a minimum, the first derivative is zero.

230 Chapter 11. An Application: Approximation and Optimization

Fig. 11.3.1: Finding a minimum of a function f (x) means finding x∗ for which f (x∗ ) is the lowest value (or
lower than all others in a local neighborhood) that function f () can take.

11.3.1 Newton’s Method for Finding the Minimum of a Function

Similar to find the root of a function explained in Section 11.2, Newton’s method for finding a local minimum
of a function f () works iteratively starting from an initial value x0 by using the second-order expansion of
f () around xi this time:

f ′ (xi ) f ′′ (xi )
f (xi+1 ) ≈ f (xi ) + (xi+1 − xi ) + (xi+1 − xi )2 , (11.3.3)
1! 2!
which can be simplified by rewriting xi+1 = xi + δ:

f ′ (xi ) f ′′ (xi ) 2
f (xi + δ) ≈ f (xi ) + δ+ δ . (11.3.4)
1! 2!
We are interested in finding δ that minimizes the approximation. This can be obtained by setting the first
derivative to zero:

d f ′ (xi ) f ′′ (xi ) 2
f (xi ) + δ+ δ = f ′ (xi ) + f ′′ (xi )δ = 0, (11.3.5)
dδ 1! 2!

which yields the delta value minimizing the approximation:

f ′ (xi )
δ=− . (11.3.6)
f ′′ (xi )

Plugging this into xi+1 = xi + δ, we obtain the equation for calculating the next value in the sequence
towards a minimum of function f ():

f ′ (xi )
xi+1 ← xi − . (11.3.7)
f ′′ (xi )

11.3. Finding a Minimum of Functions 231

11.3.2 Misc Details for the Curious

Notes on Convergence
Newton’s method can converge to a local minimum if (i) f is a strongly convex function with Lipschitz
Hessian and (ii) x0 is close enough to x∗ ← arg minx∈R f (x). More formally when:
1
∥xi+1 − x∗ ∥ ≤ ∥xi − x∗ ∥2 . ∀i ≥ 0. (11.3.8)
2
In plain English, (i) we have a “well-shaped” local neighborhood in which we can talk about the concept of
a minimum, (ii) our initial value, x0 , is inside such a neighborhood.
Notes on Multiple Minima
Similar to the case for root finding, Newton’s method converges only to a single minimum based on the
initial value x0 . In order to find other minima, the algorithm needs to be executed with a different value of
x0 , which will hopefully lead to a different minimum than the ones that we have identified before.

11.3.3 Newton’s Method in Python

Now let us implement Newton’s method for finding a minimum of a function from scratch, in Python.

def find_local_min(f_deriv, f_second_deriv, x_0, max_steps=10, delta=0.001, verbose=True):

"""
Use Newton's method to find a local minimum of function f, given its derivative f_deriv.
Inputs: f => function
f_deriv => derivative of function f
x_0 => initial x value
max_steps => maximum number of iterative steps (default: 10)
delta => if |x_i+1 - x_i| < delta, then assume convergence
"""
x_old = x_0
for i in range(max_steps):
try:
x_new = x_old - f_deriv(x_old) / f_second_deriv(x_old)
except ZeroDivisionError:
# Encountered division-by-zero, return None
return None

if verbose: print("Iteration ", i, ". (x_old, x_new): ", (x_old, x_new), " |x_new-x_
,→ old|: ", abs(x_new-x_old))
if abs(x_new-x_old) < delta: break

# Step 4
x_old = x_new

return x_new

# Assume that our function is: f(x) = x^2 - 4

def f_deriv(x_i): return 2*x_i
def f_second_deriv(x_i): return 2

# Let us test our solution with 10 random initial values

import random
(continues on next page)

232 Chapter 11. An Application: Approximation and Optimization

(continued from previous page)

draw_f() # defined in Section 11.2

for i in range(10):
x_0 = random.randint(-5, 5) # A random integer in [-5, 5]
x_min = find_local_min(f_deriv, f_second_deriv, x_0, max_steps=50, delta=0.0001,␣
,→verbose=False)
if (x_min == None) or (abs(f_deriv(x_min)) > 0.001): continue # Diverged or divison-by-
,→zero, ignore

print("trial", i, "led to x_min: ", x_min, ", f'(x_min): ", f_deriv(x_min))

plt.scatter(x_min, f(x_min))

trial 0 led to x_min: 0.0 , f'(x_min): 0.0

trial 1 led to x_min: 0.0 , f'(x_min): 0.0
trial 2 led to x_min: 0.0 , f'(x_min): 0.0
trial 3 led to x_min: 0.0 , f'(x_min): 0.0
trial 4 led to x_min: 0.0 , f'(x_min): 0.0
trial 5 led to x_min: 0.0 , f'(x_min): 0.0
trial 6 led to x_min: 0.0 , f'(x_min): 0.0
trial 7 led to x_min: 0.0 , f'(x_min): 0.0
trial 8 led to x_min: 0.0 , f'(x_min): 0.0
trial 9 led to x_min: 0.0 , f'(x_min): 0.0

11.3. Finding a Minimum of Functions 233

11.3.4 Newton’s Method for Finding Minima in SciPy

Unfortunately, Newton’s method as explained in Section 11.3.1 does not exist in SciPy. However, there are
many alternatives and we will just illustrate only one of them here.

from scipy.optimize import minimize_scalar

import random

draw_f()
for i in range(10):
x_0 = random.randint(-5, 5) # A random integer in [-5, 5]
result = minimize_scalar(f, method='golden', tol=0.001, options={'maxiter': 50})
x_min = result['x']
print("trial", i, "led to x_min: ", x_min, ", f'(x_min): ", f_deriv(x_min))
plt.scatter(x_min, f(x_min))

trial 0 led to x_min: 1.4872654968309704e-08 , f'(x_min): 2.974530993661941e-08

trial 1 led to x_min: 1.4872654968309704e-08 , f'(x_min): 2.974530993661941e-08
trial 2 led to x_min: 1.4872654968309704e-08 , f'(x_min): 2.974530993661941e-08
trial 3 led to x_min: 1.4872654968309704e-08 , f'(x_min): 2.974530993661941e-08
trial 4 led to x_min: 1.4872654968309704e-08 , f'(x_min): 2.974530993661941e-08
trial 5 led to x_min: 1.4872654968309704e-08 , f'(x_min): 2.974530993661941e-08
trial 6 led to x_min: 1.4872654968309704e-08 , f'(x_min): 2.974530993661941e-08
trial 7 led to x_min: 1.4872654968309704e-08 , f'(x_min): 2.974530993661941e-08
trial 8 led to x_min: 1.4872654968309704e-08 , f'(x_min): 2.974530993661941e-08
trial 9 led to x_min: 1.4872654968309704e-08 , f'(x_min): 2.974530993661941e-08

234 Chapter 11. An Application: Approximation and Optimization

11.4 Important Concepts

We would like our readers to have grasped the following crucial concepts and keywords from this chapter:
• Taylor Series for function approximation.
• Newton’s method for finding the roots and the minima of functions.
• Local minimum vs. global minimum.

11.5 Further Reading

• Taylor Series: https://openstax.org/books/calculus-volume-2/pages/6-3-taylor-and-maclaurin-series

• Newton’s Method:
– https://openstax.org/books/calculus-volume-1/pages/4-9-newtons-method
– http://www.opentextbookstore.com/calc/2_7.pdf

11.6 Exercises

• For the Taylor series approximation of x3 in Section 11.1.1:

1. Discuss how and why the approximation is changing when we add more terms.
2. Discuss why having 3 terms in the approximation yields a function that is very close to the
original function.
3. Write a Python code that checks whether an approximation is exactly the same as the function
being approximated. Use this code to check whether using 3 terms in Taylor series expansion
was sufficient for approximating x3 .
• Approximate the sin function using Taylor Series:
1. Approximate the sin function with the Taylor Series expansion by taking a = 0. You will need
dn
the nth derivative of sin, which turns out to be simply dx n sin(x) = sin(x + nπ/2).

2. Plot the approximation for different number of terms (m): 1, 2, 3 and 4. Observe what happens
when more terms are added to the approximation.
3. Compare your approximation results to those we obtained for x3 . Discuss the reasons for simi-
larities and differences.
• Find the roots of the following functions using Newton’s Method:
1. f (x) = x2 − 4 for x ∈ [−5, 5].
2. f (x) = sin(x) for x ∈ [0, π].
3. f (x) = log(x) for x ∈ [0.5, 2.5].
• Consider a second-order function, e.g. f (x) = x2 − 4, which has a minimum at x = 0.
1. Plot f () and its derivative f ′ () using matplotlib.
2. Do you see a link between the root of f ′ () and the minimum of f ()?

11.4. Important Concepts 235

3. Using scipy.optimize.newton() function, which is for finding the roots of a function, find
the minimum of function f (). Hint: A minimum is a root of the first order derivative.

236 Chapter 11. An Application: Approximation and Optimization

12 | An Application: Solving a Simple Re-
gression Problem

12.1 Introduction

Let us first formulate the problem and clearly outline the notation. In a regression problem, we have a
set of x, y values: {(x0 , y0 ), (x1 , y1 ), ..., (xn , yn )}, where xi and yi can be multi-dimensional variables
(i.e. vectors) but for the sake of simplicity, in this chapter we will work with one-dimensional xi and yi
values. In a regression problem, we are interested in finding the function f () that goes through those (xi , yi )
values. In other words, we wish to find f () that satisfies the following:

yi = f (xi ). (12.1.1)

At first, this might look difficult since we are trying to estimate a function from its input-output values
and there can be infinitely many functions that can go through a finite set of (xi , yi ) values. However, by
restructuring the problem a little bit and making some assumptions on the general form of f (), we can solve
such complicated regression problems very well.
Let us briefly describe how we can do so: We first re-write f () as a parametric function and try to formulate
the regression problem as the problem of finding the parameters of f ():

X
m
y = f (x) = f (x; θ) = gi (x; θi ), (12.1.2)
i=1

where θ = (θ1 , ..., θn ) is the set of parameters that we need to find to solve the regression problem; and
g1 (), ..., gm () are our “simpler” functions (compared to f ()) whose parameters can be identified more easily
compared to f ().
With this parameterized definition of f (), we can formally define regression as an optimization (minimiza-
tion) problem (see also Chapter 11):
X
θ∗ ← arg min (yi − f (xi ; θ))2 , (12.1.3)
θ∈Rd
i

237
where the summation runs over the (x0 , y0 ), (x1 , y1 ), ..., (xn , yn ) values that we are trying to regress;
(yi − f (xi ; θ))2 is the error between the estimated (regressed) value (i.e. f (xi ; θ)) and the correct yi value
with θ as the parameters.
We can describe this optimization formulation in words as: Find parameters θ that minimize the error between
yi and what the function is predicting given xi as input.
This is a minimization problem and in Chapter 11, we have seen a simple solution and how SciPy can be
used for such problems. In this chapter, we will spend more time on this with sample regression problems
and solutions. Namely,
• Linear regression, where f () is assumed to be a line, y = ax + b, and θ is (a, b).
• Non-linear regression, where f () has a non-linear nature, e.g. a quadratic, exponential or logarithmic
form.

12.1.1 Why is regression important?

In many disciplines, we observe an environment, an event, or an entity through sensors or some data collected
in a different manner, and obtain some information about what we are observing in the form of a variable x.
Each x generally has an associated outcome variable y. Or x can represent an action that we are exerting in
an environment and we are interested in finding how x affects the environment by observing a variable y.
Here are some examples for x and y for which finding a function f () can be really useful:
• x: The current that we are providing to a motor. y: The speed of the motor.
• x: The day of a year. y: The average rain fall on a year.
• x: The changes in stock value of a bond in the last day. y: The value of a bond in one hour.
• x: The number of COVID cases in a day. y: The number of deaths owing to COVID.

12.1.2 The form of the function

As discussed before, solving the regression problem without making any assumption about the underlying
function is very challenging. We can simplify the problem by making an assumption about the function.
However, an incorrect assumption can yield inaccurate regression model.
This is illustrated very well in a famous cartoon in Fig. 12.1.1. As illustrated, if the form of the function is
too simple (e.g. linear), the fitted function will not be able to capture the true nature of the data. However, if
the assumed function form is too complex (e.g. a higher-order polynomial), the fitted function will try to go
through every data point, which may not be ideal since the data can be noisy.
Choosing the right form of the function is therefore tricky and it requires some prior idea about the function
or the problem. If the relationship between x and y is known to be linear, then a linear function form should
be chosen. If the relation between x and y is known to be non-linear but the exact form is unknown, x and y
can be plotted first to get a feeling for the underlying form. This can give an idea for what kinds of function
forms can be suitable.

238 Chapter 12. An Application: Solving a Simple Regression Problem

Fig. 12.1.1: The challenging regression problem can be simplified by making an assumption about the un-
12.1. Introduction 239
derlying function; as illustrated in the drawings, the function can be assumed to be linear, quadratic, log-
arithmic etc. However, an incorrect assumption can yield inaccurate regression. (Figure source: https:
//xkcd.com/2048/)
12.2 Least-Squares Regression

A commonly-used method for regression is called Least-Squares Regression (LSE). In LSE, the minimization
problem in Section 12.1 is solved by setting the gradient of the objective to zero:

∂ X
(yi − f (xi ; θ))2 = 0, for each j. (12.2.1)
∂θj
i

Linear Least-Squares Regression. If our function is linear, i.e. f (xi ; θ) = θ1 xi + θ2 , setting the gradient
to zero provides us one set of equations for each θj . If the number of data points (xi , yi ) is bigger than the
number of variables, these equations can be easily solved to obtain θ that minimizes the objective in Section
12.1. Those interested in the derivation of the solution and its exact form can look up e.g. Wikipedia36 .
Non-linear Least-Squares Regression. If our function is not linear, a solution can be obtained iteratively
by making a linear approximation/step at each iteration using the so-called Gauss-Newton method37 .

12.3 Linear Regression with SciPy

SciPy has a function called scipy.stats.linregress(x, y) which we can use to regress a line passing
through a set of points. Let us go through an example to see how we can do that. Moreover, we will use our
regressed line to analyze some properties of our data.

12.3.1 Create Artificial Data

To keep things simple, let us generate some artificial data in Python and save it in a file called
ch12_linear_data_example.csv. We have already run this code and saved this CSV file on web – in
the following steps, we will continue with this file on the web. However, you can change the following code
the generate data in different forms and with different noise levels and observe how the different regression
methods perform.

import numpy as np
import matplotlib.pyplot as plt
import pandas as pd

def plot_func(f, params, xs, legend):

ys = f(xs, *params) # This is called parameter unpacking -- Google and learn this
plt.plot(xs, ys, label=legend)

def line_func(x, a, b):

"""Line function."""
#(a,b) = params
return a*x + b

xs = np.linspace(-10, 10, 100)

ys = line_func(xs, 3, 0.5) + np.random.normal(0, 4, xs.size) # add random noise

df = pd.DataFrame({'x': xs, 'y': ys})

df.to_csv("ch12_linear_data_example.csv")
(continues on next page)
36
https://en.wikipedia.org/wiki/Linear_least_squares
37
https://en.wikipedia.org/wiki/Gauss%E2%80%93Newton_algorithm

240 Chapter 12. An Application: Solving a Simple Regression Problem

(continued from previous page)

print("Created {0} many data points and saved them into CSV file.".format(xs.size))

Created 100 many data points and saved them into CSV file.

12.3.2 Download and Visualize Data

Let us download the ch12_linear_data.csv file from web so that we can provide you the experience of a
full data analysis pipeline.

# Let us download our data

!wget -nc https://raw.githubusercontent.com/sinankalkan/CENG240/master/figures/ch12_linear_
,→data.csv

# Import the necessary libraries

import pandas as pd

# Read the file named 'ch12_linear_data.csv'

df = pd.read_csv('ch12_linear_data.csv')

# Print the CSV file's contents:

print("The CSV file contains the following:\n", df, "\n")

# Check the types of each column

df.dtypes

--2021-06-21 20:22:34-- https://raw.githubusercontent.com/sinankalkan/CENG240/master/

,→figures/ch12_linear_data.csv
Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.110.133, 185.199.
,→108.133, 185.199.109.133, ...
Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.110.133|:443...␣
,→connected.

HTTP request sent, awaiting response... 200 OK

Length: 4049 (4.0K) [text/plain]
Saving to: ‘ch12_linear_data.csv’

ch12_linear_data.cs 100%[===================>] 3.95K --.-KB/s in 0.002s

2021-06-21 20:22:35 (2.56 MB/s) - ‘ch12_linear_data.csv’ saved [4049/4049]

The CSV file contains the following:

Unnamed: 0 x y
0 0 -10.000000 -29.198799
1 1 -9.797980 -33.379918
2 2 -9.595960 -25.661456
3 3 -9.393939 -24.037193
4 4 -9.191919 -25.264474
.. ... ... ...
95 95 9.191919 27.905021
96 96 9.393939 24.494575
97 97 9.595960 32.212389
(continues on next page)

12.3. Linear Regression with SciPy 241

(continued from previous page)
98 98 9.797980 28.511129
99 99 10.000000 33.593768

[100 rows x 3 columns]

Unnamed: 0 int64
x float64
y float64
dtype: object

# Let us visualize the data

import matplotlib.pyplot as plt

xs = df['x'][:].values
ys = df['y'][:].values

# Plot y vs. x
plt.scatter(xs, ys)

# Set the labels for x and y axes:

plt.xlabel('x')
plt.ylabel('y')

# Set the title of the figure

plt.title("y vs x")

Text(0.5, 1.0, 'y vs x')

242 Chapter 12. An Application: Solving a Simple Regression Problem

We see that our data is rather noisy, though it more or less follows a linear function. Let us see whether we
can capture that.

12.3.3 Fit a Linear Function with SciPy

Now, let us fit a line (linear function) to our data using scipy.stats.linregress. The specific method
being used by this function is Least-Squares Regression which we briefly explained in Section 12.2.

# Import stats module

from scipy import stats

# Fit a linear model:

solution = stats.linregress(xs, ys)

# Get the slope (a) and the intercept (b) in y = ax + b

a_estimated = solution.slope
b_estimated = solution.intercept

print("The estimated solution is: y = %4f x + %4f" % (a_estimated, b_estimated))

The estimated solution is: y = 2.963812 x + 0.442237

12.3.4 Analyze the Solution

We have obtained a solution. Now, let us analyze how good the solution is. We can do that qualitatively (by
visual inspection) or quantitatively (by looking at some numerical measures of how good the fit is).
1- Qualitative Analysis. For regression, a qualitative analysis can be performed by plotting the data, the
obtained solution and the correct solution. If the obtained solution goes through the data well and is close to
the original solution in the graph, then we can visually confirm whether the solution looks acceptable.
Let us do this for our example:

# Plot y vs. x
plt.scatter(xs, ys)

# Plot the estimated line

plot_func(line_func, (a_estimated, b_estimated), xs, "Estimated line")

# Plot the correct line

# The data was generated from y = 3x + 0.5
a_correct = 3
b_correct = 0.5
plot_func(line_func, (a_correct, b_correct), xs, "Correct line")

# Set the labels for x and y axes:

plt.xlabel('x')
plt.ylabel('y')

# Set the title of the figure

plt.title("y vs x")

(continues on next page)

12.3. Linear Regression with SciPy 243

(continued from previous page)
# Legend
plt.legend()

<matplotlib.legend.Legend at 0x12e6d8d30>

We can confirm that the obtained solution (blue line) goes through the original data points well and it is
almost identical to the correct line, i.e. the line from which we generated the original (x, y) pairs. So, our
regression solution appears to be accurate visually.
2- Quantitative Analysis. Qualitative analysis can be difficult sometimes and its judgment is subjective;
i.e. different observers might draw different conclusions especially in cases where the solution is different
from the correct solution. Therefore, in addition to the qualitative analysis like we did above, it is very
common to measure the goodness of our solution with a number.
Depending on the problem, different measures can be used. For regression, we frequently use the following
three:
1. Mean Squared Error (MSE). For N data points, assuming ŷi is the predicted value for xi , MSE can be
defined as follows:
1 X
N
M SE = (yi − ŷi )2 , (12.3.1)
N
i=1

which essentially accumulates the squared errors for each data point.
2. Root Mean Squared Error (RMSE). RMSE is simply the square-root of MSE:
v
u
u1 X N
RM SE = t (yi − ŷi )2 . (12.3.2)
N
i=1

244 Chapter 12. An Application: Solving a Simple Regression Problem

3. :math:`R^2` Error (also called the coefficient of determination). R2 is a normalized measure which
considers the error of the “correct” solution as well. Since data can be noisy, this can be a better way
to understand how good/bad our error value (MSE or RMSE) was. R2 Error is defined as:
P PN
MSE of estimates 1/N N i=1 (yi − ŷi )
2 (yi − ŷi )2
R =1−
2
=1− PN = 1 − Pi=1
N
, (12.3.3)
MSE wrt. the mean of y 1/N i=1 (yi − y i )2 i=1 (yi − y)
2
P
where y = 1/N i yi is the mean of y values. If R2 = 1, that means MSE of the estimates is zero; i.e. we
have obtained perfect regression. If R2 = 0, all yi = y; in other words, all estimates are the mean of the
y values, therefore the regression method did not find a good solution. If, on the other hand, R2 < 0, that
means the obtained solution is even worse than the mean yi values.
Now let us analyze our solution using these three metrics.

def MSE(estimated_y, correct_y):

"""
Mean Squared Error between the estimated_y and correct_y values.
Inputs are both numpy arrays.
"""
result = 0
N = estimated_y.size
for i in range(N):
result += (estimated_y[i] - correct_y[i]) ** 2
return result / N

def RMSE(estimated_y, correct_y):

"""
Root Mean Squared Error between the estimated_y and correct_y values.
Inputs are both numpy arrays.
"""
return MSE(estimated_y, correct_y) ** 0.5

def R_squared(estimated_y, correct_y):

"""
R^2 error between the estimated_y and correct_y values.
Inputs are both numpy arrays.
"""
MSE_of_estimates = MSE(estimated_y, correct_y)
mean_y = correct_y.mean()
mean_y_array = np.full(correct_y.size, mean_y)
MSE_wrt_mean_of_y = MSE(estimated_y, mean_y_array) # an array filled with mean-y
result = 1 - MSE_of_estimates / MSE_wrt_mean_of_y

return result

estimated_y = xs * a_estimated + b_estimated

correct_y = ys

print("MSE Error: ", MSE(estimated_y, correct_y))

print("RMSE Error: ", RMSE(estimated_y, correct_y))
print("R-squared Error: ", R_squared(estimated_y, correct_y))

MSE Error: 14.710406962785715

RMSE Error: 3.835414835814467
R-squared Error: 0.950755415581623

12.3. Linear Regression with SciPy 245

We see that we have obtained very low errors. Especially the R2 value is very close to one, which suggests
good approximation of the regressed line to the underlying solution. Note that since our data was noisy, it
was not possible to obtain R2 = 1.

12.4 Non-linear Regression with SciPy

Let us now see how we can perform non-linear regression.

12.4.1 Create Artificial Data

To keep things simple, let us generate some artificial data in Python and save it in a file called
ch12_nonlinear_data_example.csv. We have already run this code and saved this CSV file on web –
in the following steps, we will continue with this file on the web. However, you can change the follow-
ing code the generate data in different forms and with different noise levels and observe how the different
regression methods perform.

import numpy as np
import matplotlib.pyplot as plt
import pandas as pd

def nonlinear_f(xs, params):

(a,b) = params
return a*np.sin(xs)+b

xs = np.linspace(-10, 10, 100)

ys = nonlinear_f(xs, params=(3, 0.5)) + np.random.normal(0, 0.7, xs.size) # add random noise

df = pd.DataFrame({'x': xs, 'y': ys})

df.to_csv("ch12_nonlinear_data_example.csv")

print("Created {0} many data points and saved them into CSV file.".format(xs.size))

Created 100 many data points and saved them into CSV file.

12.4.2 Download and Visualize Data

Let us download the ch12_linear_data.csv file from web so that we can provide you the experience of a
full data analysis pipeline.

# Let us download our data

!wget -nc https://raw.githubusercontent.com/sinankalkan/CENG240/master/figures/ch12_
,→nonlinear_data.csv

# Import the necessary libraries

import pandas as pd

# Read the file named 'ch12_nonlinear_data.csv'

df = pd.read_csv('ch12_nonlinear_data.csv')

(continues on next page)

246 Chapter 12. An Application: Solving a Simple Regression Problem

(continued from previous page)
# Print the CSV file's contents:
print("The CSV file contains the following:\n", df, "\n")

# Check the types of each column

df.dtypes

--2021-06-21 20:22:36-- https://raw.githubusercontent.com/sinankalkan/CENG240/master/

,→figures/ch12_nonlinear_data.csv
Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.108.133, 185.199.
,→109.133, 185.199.110.133, ...
Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.108.133|:443...␣
,→connected.
HTTP request sent, awaiting response... 200 OK
Length: 4064 (4.0K) [text/plain]
Saving to: ‘ch12_nonlinear_data.csv’

ch12_nonlinear_data 100%[===================>] 3.97K --.-KB/s in 0.001s

2021-06-21 20:22:37 (3.44 MB/s) - ‘ch12_nonlinear_data.csv’ saved [4064/4064]

The CSV file contains the following:

Unnamed: 0 x y
0 0 -10.000000 3.911998
1 1 -9.797980 1.781296
2 2 -9.595960 1.125093
3 3 -9.393939 0.738523
4 4 -9.191919 -0.343199
.. ... ... ...
95 95 9.191919 0.667528
96 96 9.393939 -0.729794
97 97 9.595960 0.643281
98 98 9.797980 -1.533940
99 99 10.000000 -0.760156

[100 rows x 3 columns]

Unnamed: 0 int64
x float64
y float64
dtype: object

# Let us visualize the data

import matplotlib.pyplot as plt

xs = df['x'][:].values
ys = df['y'][:].values

# Plot y vs. x
plt.scatter(xs, ys)

# Set the labels for x and y axes:

plt.xlabel('x')
plt.ylabel('y')
(continues on next page)

12.4. Non-linear Regression with SciPy 247

(continued from previous page)

# Set the title of the figure

plt.title("y vs x")

Text(0.5, 1.0, 'y vs x')

12.4.3 Fitting a Non-linear Function with SciPy

For non-linear regression, we will use scipy.optimize.curve_fit which internally performs Non-linear
Least-squares Regression.

# Import optimize module

from scipy import optimize

def nonlinear_f(x, a, b):

return a * np.sin(x) + b

# Fit a linear model:

solution, _ = optimize.curve_fit(nonlinear_f, xs, ys, method='lm')

print("The estimated solution is: ", solution)

The estimated solution is: [3.11746172 0.48206522]

248 Chapter 12. An Application: Solving a Simple Regression Problem

12.4.4 Analyzing the Solution

Similar to our analysis in Section 12.3.4, let us analyze our solution qualitatively and quantitatively.
1- Qualitative Analysis. Let us first analyze the found solution qualitatively.

# Plot y vs. x
plt.scatter(xs, ys)

# Plot the estimated line

plot_func(nonlinear_f, solution, xs, "Estimated curve")

# Plot the correct curve

plot_func(nonlinear_f, (3, 0.5), xs, "Correct curve")

# Set the labels for x and y axes:

plt.xlabel('x')
plt.ylabel('y')

# Set the title of the figure

plt.title("y vs x")

# Legend
plt.legend()

<matplotlib.legend.Legend at 0x12e7b9b80>

Visually, it appears that our estimated solution is a very good fit for the data and it overlaps nicely with the
correct function with which we had generated the data.
2- Quantitative Analysis. Now, let us look at some the metrics we have defined in Section 12.3 to get a

12.4. Non-linear Regression with SciPy 249

better feeling for how good our fit was:

estimated_y = nonlinear_f(xs, *solution)

correct_y = ys

print("MSE Error: ", MSE(estimated_y, correct_y))

print("RMSE Error: ", RMSE(estimated_y, correct_y))
print("R-squared Error: ", R_squared(estimated_y, correct_y))

MSE Error: 0.6748838046926273

RMSE Error: 0.8215131190021419
R-squared Error: 0.8540114575179684

We see again that we have obtained low error values, despite the noise. However, we couldn’t obtain perfect
regression (R2 = 1) since our data was very noisy.

12.5 Important Concepts

We would like our readers to have grasped the following crucial concepts and keywords from this chapter:
• Regression problem and its uses.
• Least-squares regression.
• Linear vs. non-linear regression.
• Linear and non-linear regression with SciPy.
• Qualitatively and quantitatively analyzing a regression solution.

12.6 Further Reading

• The Elements of Statistical Learning, a free book that provides a good coverage of linear and non-linear
regression: https://web.stanford.edu/~hastie/ElemStatLearn/
• Practical Regression and Anova using R, another free book that provides a practical coverage: https:
//cran.r-project.org/doc/contrib/Faraway-PRA.pdf

12.7 Exercises

1. For clarity, we implemented the error metrics in Section 12.3 using for loops. Re-implement them
using NumPy functions without using explicit for or while loops.
2. Generate 100 random (x, y) values from a non-linear function, e.g. y = x2 , for x ∈ [−10, 10] and use
the linear regression steps in Section 12.3 to regress a linear solution. Qualitatively and quantitatively
analyze the obtained solution.
3. Choose your favorite city and download its temperature readings for two consecutive days from me-
teoblue38 . Download the readings as CSV files and load them using Pandas. Plot the data using
38
https://content.meteoblue.com/en/content/view/full/2879

250 Chapter 12. An Application: Solving a Simple Regression Problem

Matplotlib and see the general shape of the function. Choose a suitable non-linear function and follow
the steps in Section 12.4 to regress a function to the temperature data. Measure also how good the
function represents the temperature readings of the second day.

12.7. Exercises 251

(Ebook PDF) Biological Psychology 13th Edition by James W. Kalat 2024 Scribd Download
100% (5)
(Ebook PDF) Biological Psychology 13th Edition by James W. Kalat 2024 Scribd Download
41 pages
Effective Xgboost
No ratings yet
Effective Xgboost
221 pages
Instant Download Regression Analysis An Intuitive Guide For Using and Interpreting Linear Models 1st Edition Jim Frost PDF All Chapter
0% (1)
Instant Download Regression Analysis An Intuitive Guide For Using and Interpreting Linear Models 1st Edition Jim Frost PDF All Chapter
62 pages
A Practical Guide To Hemiplegia Treatment
100% (7)
A Practical Guide To Hemiplegia Treatment
528 pages
Cashs Textbook of Neurology For Physiotherapists
79% (52)
Cashs Textbook of Neurology For Physiotherapists
468 pages
Combinatorial Optimization: Alexander Schrijver
No ratings yet
Combinatorial Optimization: Alexander Schrijver
34 pages
Principles of Algorithmic Problem Solving PDF
100% (1)
Principles of Algorithmic Problem Solving PDF
351 pages
Python - Follow Dr. AngShu (@drangshu) For More
100% (1)
Python - Follow Dr. AngShu (@drangshu) For More
300 pages
Statistics and Machine Learning in Python
100% (1)
Statistics and Machine Learning in Python
166 pages
PythonScientific Simple PDF
100% (2)
PythonScientific Simple PDF
335 pages
Student Booklet For Sep 2015 v6
100% (1)
Student Booklet For Sep 2015 v6
50 pages
Python Tutorial: Release 2.0
100% (1)
Python Tutorial: Release 2.0
77 pages
StatisticsMachineLearningPythonDraft PDF
100% (1)
StatisticsMachineLearningPythonDraft PDF
323 pages
Scientific Python (By Stanford)
No ratings yet
Scientific Python (By Stanford)
58 pages
Flask Restplus
No ratings yet
Flask Restplus
86 pages
Learning IPython For Interactive Computing and Data Visualization - Second Edition - Sample Chapter
0% (1)
Learning IPython For Interactive Computing and Data Visualization - Second Edition - Sample Chapter
64 pages
Python Programming Notes
No ratings yet
Python Programming Notes
144 pages
Knapsack Algorithm
No ratings yet
Knapsack Algorithm
9 pages
Notes Computing
No ratings yet
Notes Computing
201 pages
API Reference - Scikit-Learn 0.19.2 Documentation
No ratings yet
API Reference - Scikit-Learn 0.19.2 Documentation
21 pages
Simple Libraries in Python
No ratings yet
Simple Libraries in Python
12 pages
(Series in Computational Physics) David J. Pine - Introduction To Python For Science and Engineering (2019, CRC Press) PDF
100% (4)
(Series in Computational Physics) David J. Pine - Introduction To Python For Science and Engineering (2019, CRC Press) PDF
389 pages
Let Us C Y Kanitkar 01 - First Few Pages
No ratings yet
Let Us C Y Kanitkar 01 - First Few Pages
7 pages
Applied Statistics with Python
No ratings yet
Applied Statistics with Python
320 pages
Textbook Python New
No ratings yet
Textbook Python New
318 pages
Role of Machine Learning in The Field of Fiber Reinforced Polymer
No ratings yet
Role of Machine Learning in The Field of Fiber Reinforced Polymer
6 pages
State Oriented Programming
No ratings yet
State Oriented Programming
32 pages
Machine Learning Guide Line
No ratings yet
Machine Learning Guide Line
10 pages
Differentiation Methods With Numerical Computing and Python Programming Language.12
No ratings yet
Differentiation Methods With Numerical Computing and Python Programming Language.12
13 pages
Object Oriented Programming Using C++
No ratings yet
Object Oriented Programming Using C++
2 pages
Umberto Michelucci - Fundamental Mathematical Concepts for Machine Learning in Science-Springer (2024)
No ratings yet
Umberto Michelucci - Fundamental Mathematical Concepts for Machine Learning in Science-Springer (2024)
259 pages
Simulation and 3D Visualization of Physical Phenomena On Mobile Devices
No ratings yet
Simulation and 3D Visualization of Physical Phenomena On Mobile Devices
80 pages
ScipyLectures Simple
100% (1)
ScipyLectures Simple
657 pages
Mathematical Tools For Data Science
No ratings yet
Mathematical Tools For Data Science
9 pages
Prealgebra Via Python Programming (En)
No ratings yet
Prealgebra Via Python Programming (En)
270 pages
Python in Hidrology Book
100% (1)
Python in Hidrology Book
153 pages
Python Tutorial
No ratings yet
Python Tutorial
32 pages
Python Parallel Programming Cookbook - Sample Chapter
67% (3)
Python Parallel Programming Cookbook - Sample Chapter
39 pages
Python For Software Development
100% (1)
Python For Software Development
196 pages
Tutorials Point DAA PDF
No ratings yet
Tutorials Point DAA PDF
102 pages
Peter R. Turner, Thomas Arildsen, Kathleen Kavanagh - Applied Scientific Computing With Python-Springer (2018)
No ratings yet
Peter R. Turner, Thomas Arildsen, Kathleen Kavanagh - Applied Scientific Computing With Python-Springer (2018)
277 pages
Introduction To Python Programming: Dr. R. Rajeswara Rao Professor & Head Dept. of CSE Jntuk-Ucev Vizianagaram
No ratings yet
Introduction To Python Programming: Dr. R. Rajeswara Rao Professor & Head Dept. of CSE Jntuk-Ucev Vizianagaram
27 pages
Notes
No ratings yet
Notes
422 pages
Calculus for Machine Learning
100% (1)
Calculus for Machine Learning
283 pages
bookAdomaitisAppliedMath4ChE 19utf81 PDF
No ratings yet
bookAdomaitisAppliedMath4ChE 19utf81 PDF
216 pages
Python - Programming
No ratings yet
Python - Programming
9 pages
Neuromorphic Computing
No ratings yet
Neuromorphic Computing
14 pages
Integer Factorization - Master Thesis
No ratings yet
Integer Factorization - Master Thesis
92 pages
Scikit Learn Docs
100% (1)
Scikit Learn Docs
2,201 pages
An Introduction To PyCUDA Using Prefix Sum Algorithm PDF
No ratings yet
An Introduction To PyCUDA Using Prefix Sum Algorithm PDF
6 pages
PyFEM PDF
No ratings yet
PyFEM PDF
22 pages
K Means Clustering
100% (1)
K Means Clustering
10 pages
Python GUI Programming With PySide PDF
100% (1)
Python GUI Programming With PySide PDF
25 pages
Deep Learning Kathi
No ratings yet
Deep Learning Kathi
18 pages
Informatics College Pokhara: Fundamentals of Computing CS4051NP
No ratings yet
Informatics College Pokhara: Fundamentals of Computing CS4051NP
37 pages
Scilab Manual
No ratings yet
Scilab Manual
700 pages
ECON2125/8013 Maths Notes: John Stachurski March 4, 2015
No ratings yet
ECON2125/8013 Maths Notes: John Stachurski March 4, 2015
162 pages
Dynamic programming The Ultimate Step-By-Step Guide
From Everand
Dynamic programming The Ultimate Step-By-Step Guide
Gerardus Blokdyk
No ratings yet
The Essential R Reference
From Everand
The Essential R Reference
Mark Gardener
No ratings yet
Python Unleashed: Mastering the Art of Efficient Coding
From Everand
Python Unleashed: Mastering the Art of Efficient Coding
James Livingston
No ratings yet
New Learning of Python by Practical Innovation and Technology
From Everand
New Learning of Python by Practical Innovation and Technology
Sudhir Pathania
No ratings yet
Backpropagation: Fundamentals and Applications for Preparing Data for Training in Deep Learning
From Everand
Backpropagation: Fundamentals and Applications for Preparing Data for Training in Deep Learning
Fouad Sabry
No ratings yet
Building Modern GUIs with tkinter and Python: Building user-friendly GUI applications with ease (English Edition)
From Everand
Building Modern GUIs with tkinter and Python: Building user-friendly GUI applications with ease (English Edition)
Saurabh Chandrakar
No ratings yet
Immediate download (Ebook) Atlas of Psychiatric Pharmacotherapy, Second Edition by Shiloh Roni, David J. Nutt, Abraham Weizman ISBN 9781841842813, 9781264332427, 1264332424, 1841842818 ebooks 2024
100% (4)
Immediate download (Ebook) Atlas of Psychiatric Pharmacotherapy, Second Edition by Shiloh Roni, David J. Nutt, Abraham Weizman ISBN 9781841842813, 9781264332427, 1264332424, 1841842818 ebooks 2024
71 pages
Psych Chapter 12 Learning Memory and Intellegence707
No ratings yet
Psych Chapter 12 Learning Memory and Intellegence707
13 pages
Nervous Coordination Dongo
No ratings yet
Nervous Coordination Dongo
21 pages
Unit Test Nervous System 14.1
No ratings yet
Unit Test Nervous System 14.1
4 pages
08 Intro Nervous System Coloring Book
No ratings yet
08 Intro Nervous System Coloring Book
4 pages
Risk Factors, Diagnosis, Prognosis and Treatment of Autism
No ratings yet
Risk Factors, Diagnosis, Prognosis and Treatment of Autism
36 pages
Nervous System Mcqs YT Insta Medico Slides
No ratings yet
Nervous System Mcqs YT Insta Medico Slides
16 pages
Biology Exam
No ratings yet
Biology Exam
12 pages
Summary and Activities - Unit 4 Interaction Function - 2024
No ratings yet
Summary and Activities - Unit 4 Interaction Function - 2024
8 pages
CV Monica Feliu-Mojer, Ph.D.
No ratings yet
CV Monica Feliu-Mojer, Ph.D.
13 pages
4P's and Causes of Abnormality
No ratings yet
4P's and Causes of Abnormality
5 pages
7828 Dpsy535 Biological Basis of Behaviour
No ratings yet
7828 Dpsy535 Biological Basis of Behaviour
110 pages
X Class Biology CBSE Phase - II Funworks Key 23-24
No ratings yet
X Class Biology CBSE Phase - II Funworks Key 23-24
5 pages
Memory Eric Kandel
No ratings yet
Memory Eric Kandel
6 pages
Prof. Dr. Hussein Abd EL Dayem
No ratings yet
Prof. Dr. Hussein Abd EL Dayem
64 pages
Keeper by Andrea Gillies - Excerpt
100% (3)
Keeper by Andrea Gillies - Excerpt
35 pages
12 Bio CH 17 MCQs
No ratings yet
12 Bio CH 17 MCQs
28 pages
10.1201 9780367823467 Previewpdf
No ratings yet
10.1201 9780367823467 Previewpdf
100 pages
Foundations of Speech and Hearing Anatomy and Physiology 2nd Edition Jeannette D. Hoit download
100% (1)
Foundations of Speech and Hearing Anatomy and Physiology 2nd Edition Jeannette D. Hoit download
50 pages
XII - Zoology Exercise Questions
No ratings yet
XII - Zoology Exercise Questions
34 pages
SC - M1 - Ktunotes - in
No ratings yet
SC - M1 - Ktunotes - in
190 pages
Drugs Revision Biology Olevel IGCSE
No ratings yet
Drugs Revision Biology Olevel IGCSE
39 pages
Canine and Feline Epilepsy Diagnosis and Management De Risio - Quickly access the ebook and start reading today
100% (1)
Canine and Feline Epilepsy Diagnosis and Management De Risio - Quickly access the ebook and start reading today
84 pages
Biolistrik: DR - Susy Olivia
No ratings yet
Biolistrik: DR - Susy Olivia
97 pages
Synapse
100% (1)
Synapse
7 pages
B22 Neural Mechanisms of Learning &amp Memory
No ratings yet
B22 Neural Mechanisms of Learning &amp Memory
21 pages