CS229

Python & Numpy

J ingbo Yang
How is python related to with others?

Python 2.0 released in 2000

(Python 2.7 “end-of-life” in 2020)

Python 3.0 released in 2008

(Python 3.6+ for CS 229)

Can run interpreted, like MATLAB

https://www.researchgate.net/figure/Genealogy-of-Programming-Languages-doi101371-
journalpone0088941g001_fig1_260447599
Before you start

Use Anaconda
Create a new environment (full Conda)

conda create -n cs229

Create an environment (Miniconda)

conda env create -f environment.yml

Activate an environment after creation

conda activate cs229

Notepad is not your friend …

Get a text editor/IDE

• PyCharm (IDE)

• Visual Studio Code (IDE??)

• Sublime Text (IDE??)

• Notepad ++/gedit

• Vim (for Linux)

To make you more prepared

PyCharm
• Great debugger
• Proper project
management

FYI, professional version free for students: https://www.jetbrains.com/student/

To make you more prepared

Visual Studio Code

• Light weight
• Wide variety of plugins
to enable support for
all languages
• Better UI
Basic Python
String manipulation
Formatting print('I love CS229. (upper)'.upper())
print('I love CS229. (rjust 20)'.rjust(20))
print('we love CS229. (capitalize)'.capitalize())
print(' I love CS229. (strip) '.strip())

Concatenation print('I like ' + str(cs_class_code) + ' a lot!')

print(f'{print} (print a function)')
print(f'{type(229)} (print a type)')

Formatting print('Old school formatting: {.2F}'.format(1.358))

List
list_1 = ['one', 'two', 'three’]
List creation
list_1.append(4)
list_1.insert(0, 'ZERO’)
list_2 = [1, 2, 3]
Insertion/extension list_1.extend(list_2)

long_list = [i for i in range(9)]

long_long_list = [(i, j) for i in range(3)
List comprehension for j in range(5)]
long_list_list = [[i for i in range(3)]
for _ in range(5)]

sorted(random_list)

Sorting random_list_2 = [(3, 'z'), (12, 'r'), (6, 'e’),

(8, 'c'), (2, 'g')]
sorted(random_list_2, key=lambda x: x[1])
Dictionary and Set
my_set = {i ** 2 for i in range(10)}
Set {0, 1, 64, 4, 36, 9, 16, 49, 81, 25}
(unordered, unique)

my_dict = {(5 - i): i ** 2 for i in range(10)}

Dictionary
{5: 0, 4: 1, 3: 4, 2: 9, 1: 16, 0: 25, -1: 36,
(mapping) -2: 49, -3: 64, -4: 81}
dict_keys([5, 4, 3, 2, 1, 0, -1, -2, -3, -4])

second_dict = {'a': 10, 'b': 11}

Dictionary update my_dict.update(second_dict)

Iterate through items for k, it in my_dict.items():

print(k, it)
Numpy & Scipy
What is Numpy?

Numpy – package for vector and matrix manipulation

FYI, if you are interested in how/why vectorization is faster,

checkout topics such as (completely optional)
AVX instruction set (SIMD) and structure of x86 and RISC
OpenMP and CUDA for multiprocessing
Assembly-level optimization, memory stride, caching, etc.
Or even about memory management, virtualization
More bare metal FPGA, TPU
Convenient math functions, read before use!
Python Command Description

np.linalg.inv Inverse of matrix (numpy as equivalent)

np.linalg.eig Get eigen value (Read documentation on eigh and numpy equivalent)

np.matmul Matrix multiply

np.zeros Create a matrix filled with zeros (Read on np.ones)

np.arange Start, stop, step size (Read on np.linspace)

np.identity Create an identity matrix

np.vstack Vertically stack 2 arrays (Read on np.hstack)

Your friend for debugging

Python Command Description

array.shape Get shape of numpy array

array.dtype Check data type of array (for precision, for weird behavior)

type(stuff) Get type of a variable

import pdb; pdb.set_trace() Set a breakpoint (https://docs.python.org/3/library/pdb.html)

print(f’My name is {name}’) Easy way to construct a message

Basic Numpy usage
array_1d = np.array([1, 2, 3, 4])
Initialization from Python lists array_1by4 = np.array([[1, 2, 3, 4]])

large_array = np.array([i for i in range(400)])

large_array = large_array.reshape((20, 20))

Lists with different types from_list = np.array([1, 2, 3])

from_list_2d = np.array([[1, 2, 3.0], [4, 5, 6]])
(Numpy auto-casts to higher from_list_bad_type = np.array([1, 2, 3, 'a'])
precision, but it should be
print(f'Data type of integer is {from_list.dtype}')
reasonably consistent) print(f'Data type of float is {from_list_2d.dtype}')

array_1 + 5
array_1 * 5
Numpy supports many types np.sqrt(array_1)
np.power(array_1, 2)
of algebra on an entire array np.exp(array_1)
np.log(array_1)
Dot product and matrix multiplication
array_1 @ array_2
A few ways to write dot array_1.dot(array_2)
np.dot(array_1, array_2)
product
weight_matrix = np.array([1, 2, 3, 4]).reshape(2, 2)
sample = np.array([[50, 60]]).T
Matrix multiplication like Ax np.matmul(weight_matrix, sample)

mat1 = np.array([[1, 2], [3, 4]])

2D matrix multiplication mat2 = np.array([[5, 6], [7, 8]])
np.matmul(mat1, mat2)

a = np.array([i for i in range(10)]).reshape

Element-wise multiplication (2, 5)

a * a
np.multiply(a, a)
np.multiply(a, 10)
Broadcasting
Numpy compares dimensions of operands, then infers missing/mismatched
dimensions so the operation is still valid. Be careful with DIMENSIONS
op1 = np.array([i for i in range(9)]).reshape(3, 3)
op2 = np.array([[1, 2, 3]])
op3 = np.array([1, 2, 3])
array([[ 1, 3, 5],
[ 4, 6, 8],
[ 7, 9, 11]])
# Notice that the results here are DIFFERENT!
pp.pprint(op1 + op2) array([[ 1, 2, 3],
pp.pprint(op1 + op2.T) [ 5, 6, 7],
[ 9, 10, 11]])

array([[ 1, 3, 5],
# Notice that the results here are THE SAME! [ 4, 6, 8],
pp.pprint(op1 + op3) [ 7, 9, 11]])
pp.pprint(op1 + op3.T)
array([[ 1, 3, 5],
[ 4, 6, 8],
[ 7, 9, 11]])
 Broadcasting for pairwise distance
samples = np.random.random((15, 5))

# Without broadcasting
expanded1 = np.expand_dims(samples, axis=1)
tile1 = np.tile(expanded1, (1, samples.shape[0], 1))
expanded2 = np.expand_dims(samples, axis=0)
tile2 = np.tile(expanded2, (samples.shape[0], 1 ,1))
diff = tile2 - tile1
Both achieve the effect of
distances = np.linalg.norm(diff, axis=-1)

# With broadcasting
diff = samples[: ,np.newaxis, :]
- samples[np.newaxis, :, :]
distances = np.linalg.norm(diff, axis=-1)

# With scipy (another math toolbox)

import scipy
distances = scipy.spatial.distance.cdist(samples, samples)
Why should I vectorize my code?
Shorter code, faster execution
a = np.random.random(500000)
b = np.random.random(500000)

With loop Numpy dot product

dot = 0.0 print(np.array(a).dot(np.array(b)))

for i in range(len(a)):
dot += a[i] * b[i]

print(dot)

Wall time: 345ms Wall time: 2.9ms

An example with pairwise distance
Speed up depends on setup and nature of computation

samples = np.random.random((100, 5))

With loop Numpy with broadcasting

total_dist = [] diff = samples[: ,np.newaxis, :] -

for s1 in samples: samples[np.newaxis, :, :]
for s2 in samples: distances = np.linalg.norm(diff, axis=-1)
d = np.linalg.norm(s1 - s2) avg_dist = np.mean(distances)
total_dist.append(d)

avg_dist = np.mean(total_dist)

Wall time: 162ms Wall time: 3.5ms

(imagine without Numpy norm)
Plotting
Other Python packages/tools
Jupyter Notebook
• Interactive, re-execution, result storage

Pandas (https://pandas.pydata.org/)
• Dataframe (database/Excel-like)
• Easy filtering, aggregation (also plotting, but few
people uses Pandas for plotting)

Matplotlib
• Visualization (line, scatter, bar, images and even
interactive 3D)
 Example plots
https://matplotlib.org/3.1.1/gallery/index.html

import matplotlib
Import import matplotlib.pyplot as plt
import numpy as np

# Data for plotting

Create data t = np.arange(0.0, 2.0, 0.01)
s = 1 + np.sin(2 * np.pi * t)

fig, ax = plt.subplots()
Plotting ax.plot(t, s)

ax.set(xlabel='time (s)', ylabel='voltage (mV)',

Format plot title='About as simple as it gets, folks')
ax.grid()

fig.savefig("test.png")
Save/show plt.show()
Plot with dash lines and legend
import numpy as np
import matplotlib.pyplot as plt

x = np.linspace(0, 10, 500)

y = np.sin(x)

fig, ax = plt.subplots()

line1, = ax.plot(x, y, label='Using set_dashes()')

# 2pt line, 2pt break, 10pt line, 2pt break
line1.set_dashes([2, 2, 10, 2])

line2, = ax.plot(x, y - 0.2, dashes=[6, 2],

label='Using the dashes parameter')

ax.legend()
plt.show()
Using subplot

x = np.arange(0, 3 * np.pi, 0.1)

y_sin = np.sin(x)
y_cos = np.cos(x)

# Setup grid with height 2 and col 1.

# Plot the 1st subplot
plt.subplot(2, 1, 1)

plt.grid()
plt.plot(x, y_sin)
plt.title('Sine Wave')

# Now plot on the 2nd subplot

plt.subplot(2, 1, 2)
plt.plot(x, y_cos)
plt.title('Cosine Wave')

plt.grid()
plt.tight_layout()
Plot area under curve
 Confusion matrix
https://scikit-learn.org/stable/auto_examples/model_selection/plot_confusion_matrix.html
fig, ax = plt.subplots()
im = ax.imshow(cm, interpolation='nearest', cmap=cmap)
ax.figure.colorbar(im, ax=ax)
# We want to show all ticks...
ax.set(xticks=np.arange(cm.shape[1]),
yticks=np.arange(cm.shape[0]),
xticklabels=classes, yticklabels=classes,
ylabel='True label', xlabel='Predicted label’,
title=title)

# Rotate the tick labels and set their alignment.

plt.setp(ax.get_xticklabels(), rotation=45, ha='right',
rotation_mode='anchor')

# Loop over data dimensions and create text annotations.

fmt = '.2f' if normalize else 'd'
thresh = cm.max() / 2.
for i in range(cm.shape[0]):
for j in range(cm.shape[1]):
ax.text(j, i, format(cm[i, j], fmt),
ha='center', va='center',
color="white" if cm[i, j] > thresh else "black")
fig.tight_layout()
Good luck on your
HW/Project!

Questions?
Links

CS 231N Python Tutorial

Additional slides in
case of Q&A
Where does my program start?
It just works
A function

Properly
What is a class?

Initialize the class to

get an instance using
some parameters

Instance variable

Does something
with the instance
To use a class

Instantiate a class,

get an instance

Call an instance method

String manipulation
Formatting
stripped = ' I love CS229! '.strip()
upper_case = 'i love cs 229! '.upper()
capitalized = 'i love cs 229! '.capitalize()
Concatenation
joined = ‘string 1’ + ‘ ’ + ‘string 2'
Formatting
formatted = ‘Formatted number {.2F}’.format(1.2345)
Basic data structures
List
example_list = [1, 2, '3', 'four’]
Set (unordered, unique)
example_set = set([1, 2, '3', 'four’])
Dictionary (mapping)
example_dictionary =
{
'1': 'one',
'2': 'two',
'3': 'three'
}
More on List

2D list
list_of_list = [[1,2,3], [4,5,6], [7,8,9]]
List comprehension
initialize_a_list = [i for i in range(9)]
initialize_a_list = [i ** 2 for i in range(9)]
initialize_2d_list = [[i + j for i in range(5)] for j in range(9)]
Insert/Pop
my_list.insert(0, ‘stuff)
print(my_list.pop(0))
More on List

Sort a list
random_list = [3,12,5,6]
sorted_list = sorted(random_list)

random_list = [(3, ‘A’),(12, ’D’),(5, ‘M’),(6, ‘B’)]

sorted_list = sorted(random_list, key=lambda x: x[1])
More on Dict/Set

Comprehension
my_dict = {i: i ** 2 for i in range(10)}
my_set = {i ** 2 for i in range(10)}

Get dictionary keys

my_dict.keys()
Another way for legend
Scatter plot