numpy module

pandas module
visualization using matplotlib and seaborn
hello world of ml (Dev a perceptron)
kaggle,github, machine hack site for compettiv ml prog
scikit learn algo are available
regression -data set is nt fixed
classification-predicting on fixed set of data,discrete set f data for prediction
- logistic Regression
- Decision Tree
- Naive Bayes
- knn(k nearest neighbour)
- metrics (regression and classification)

cluster

project work

day 1
REPL(repeat execute print loop)
-compiled prog run faster,pyton is interpreted
-interpreted -interactive -object oriented prog
type(x)-gives the data type of the variable

we take log aor expo or scaling for managing the data set
1/(1+math.exp(x)) sigmoid fuction gives value between 0-1,may b used for
calculating probability
soft math,
for library with big names we can simply write
import math as m
now we can use m everytime now
if any function in the library is too big we can define it as
from math import sqrt as s
now we can use s as sqrt

day 2
list==[] replace array ,list doesnt have fixed size
set=={}
tuple==()
dictionary=={}==like hashmap in java
numpy have array
pandas have series and datagram
marks=[[1,2],[7,3]] 2 d array
access elements of 2d list

marks[1][2] it will print 3

len(list name) gives the size of list

list have negative index too, its follows reverse order
ex marks=[1,2,3,4,5,6,7,8,9]
marks[-1] will print 9

slicing marks[1:5] will print 2,3,4,5

marks[start:end+1]
marks[start:] will print till last
marks[:] print whole array
del listname(index) will delete the element of that index

marks=list() or marks=[] defines blank list

marks=list(range(-20,21)) range is the generator here list function
comprehension
sqrtmarks=[math.sqrt(x) for x in range(start,end)]
import math
marks=list(range(-21,22))
sigmoid=[1/(1+math.exp(-x)) for x in marks]
print(sigmoid)

list 1= elements...
list2=......
str="score of {} is {}"
fial=str.format(list1[0],list2[0]) complets the template string

ex
for i in range(len(names)):
print(str.format(names[i],score[i]))

or
print("score of {} is {}.format(list1[0],list2[0])

Tuple
t1=()
t1=(12,213,113)
t1=tuple()
tuple is read only list
t1[0]=40 ==> will not work
l1=list(t1)==> converting a tuple to a list
t2=tuple(l1)=== convert a list to tuple

day3
dictionary 2d array key value pair,values are return with key
if key is repeated then the last entry will be saved
d1={"key1":"value1","key2":"value2"}
get funtion
d1.get(1)
d1[1]

set
s1={}
s1=set()
s1={324,53646,57876,453,2342,23424,3564,46}
s2={123,343,44,34,344,44,4}
s3=s1.union(s2)
s4=s1.intersection(s2)
zip function
roll=[1,2,3,4,5]
name=[a,b,c,d,e]

students=list(zip(roll,names))
zip is generator it doesnt creat a list
for e in zip(roll,names):
print e

enumerate (roll) work on list

it will return (0,12) basically add sr no
(1,12)
function

def funnctionname(parameter):
'''description
'''
def circle(radius):

def area():
return 3.14*radius**2
def circum():
return 2*3.14*radius
def diameter():
return radius*2
return(area,circum,diameter)

radius=8

ar,pr,di=circle(radius)
print(ar(),pr(),di())

day4
string ,list
s="amit,sumit,prason"
l=s.split(":") #breaks the string in list
print(l)
class and object
no concept of visibility
method must have one parameter,by convention we write self it is similar to this
in c++

class student:
def setdata(self,r,n,m):#we need to pass just 3 parameter self is handeled by
python complier
self.roll=r
self.name=n
self.marks=m
def prd(self):
print("roll",self.roll)
print("name",self.name)
s=student()
s.setdata(32,"delta",99)
s.prd()

def __init__(self,r,n,m):#__ is used for system specific function,its the

constructoor of the class
self.roll=r
self.name=n
self.marks=m

there are two ways to return from to function

1: yeild
 2: return
yeild s executed via next() ,it act as generator function ,execution happen part
by part the no of time next is called
the next yeild the next value ,until yeild is defined next can be called t iterate

import re #regular expression package

#re.sub()
#re.find()
regstr=["a-zA-Z0-9"]
s="cats are good,cats are ugly.bats are flying"
#s="cat cat cat rat rat"
#regexpr="cat"
regexpr="[^cr]at"#finds both cat or rat
obj=re.search(regexpr,s)
#obj=re.find(regexpr,s)
#print(type(obj))
if obj is not None:
print("found")
p';l

2,4,4,6,7,8, ,100000

pandas ===panel data

######################################################

DAY 6

#############################################################

for i in autodf.modelyear.unique():
print("modelyear:",i,autodf[autodf.modelyear==i]["cylinders"].value_counts())

###################################################################################
###########################################
ml

1-Formula
2-tree
3-probability
sgd classifier ,random forest ,neive bayes ,percepton etc

artifical general intelligence

ml
1-supervised-feed many examples on wich machine learns and predict
2-unsupervised
3-reinforcement
4-ensamble -random forest
unsupervised -predict som grouping in the giving data,data doest have any label,
(clustring works on unsupervised)
the learning process
-data gathering process
-data processing
-diensionality Reduction-(topic modeling-Probablititc topic modeling,)
-model learning-(classification,Regression,Clustering,Description)
-model testing
Training data ===>feed to model and testing data===>model=====>predict

perceptron ====>(single layer perceptron (slp))

google-ai winter
perseptron is a linear seperator
multilayer perseptron is nueral network
--Epochs
--Learning rate
--weights
--weight update
--loss function
--gradient(partial derivative)

loss function

#machine hack,machine veda

Classfication:
a>logistic Regession
b>Decision Tress
C>naive Bayes
D>KNN --- non parametric learning,lazy learner doesnt create model
e>random Forest
F>SVM
g>Gaussion kernal
H>Perceptron
i>Neural Network
h>SGD classifier

confusion mat
0 1
0 TN FP

1 FN TP

TP
--------- ===>pression
FP+ TP
 TP
---------- ======>recall/sensitivity (TPR)
FN +TP

a==>decide what metric is important ?

b==>drop bad records so that total no of dropped record < 70 -100
c==>take all columns and fit model and check train metrics and test metrics overfit
or under fit
d==>do some feature engg(reduction here)..apply select Kbest or select from model
to find feature ,build model and check metrics
e==> think about converting some continous feature into categorical feature and
create model

decision tree (regression as well as classification)

cart tree -classification regression tree

ID3 tree-
entropy -coming from information gain theory ,more the entropy more is the
uncertain is the infomation
parameter and hyper parameter

1>what u need to achive?(recall,precission,....) why?(In domain terms)

2>inspect entire dataset
a> which is dependent (y)/independent(X)
b>total records/total columns
c> must have indepth knowledge what values are there(any abnormal
values,values which are not matching with meaning of columns)
d> for categorical column must be clear about frequencies
e> for continous columns must clear about wheather there are outliers .if the
outliers what are the corresponding values
number statistics
f>any np.NAN are there,if there what to do with them
3>iT is possible that ,u may not known meaning of some columns .In that case
find out weather they are inuncing your dependent variable

knn,decission tree,-- response encoding

logistic, sgd ---one hot
There is a Getting Started Category in the Kaggle competition page.

Classification Problem: https://www.kaggle.com/c/titanic

Regression Problem: https://www.kaggle.com/c/house-prices-advanced-regression-

techniques

Computer Vision: https://www.kaggle.com/c/digit-recognizer

Image Processing: https://www.kaggle.com/c/facial-keypoints-detection

Natural Language Processing: https://www.kaggle.com/c/word2vec-nlp-tutorial

Data set
https://www.kaggle.com/tmdb/tmdb-movie-metadata

https://www.kaggle.com/russellyates88/suicide-rates-overview-1985-to-2016

https://www.kaggle.com/c/porto-seguro-safe-driver-prediction
https://www.kaggle.com/c/amazon-employee-access-challenge

Greetings from firstnaukri.com!!!

Firstnaukri.com is a group portal of Naukri.com and is exclusively focused on

enabling companies to hire from campuses. We facilitate companies to connect with
campuses through our campus hiring platform, which is designed to ensure privacy of
student details and is operated through a TPO invite mechanism. Companies can
connect with students only if the TPO accepts there invite and further student
details are visible to company only if a student applies to the job.
Since campus hiring is highly dependent on approaching the companies at the right
time, we are trying to complete the 2020 batch registrations as soon as possible.
This will allow us enough time to connect with more companies and deliver
meaningful invites to your campus.Kindly ask the pre-final year students i.e 2020
passing out batch to register on the below dedicated campus registration link

https://www.firstnaukri.com/campusreg/ncccyrx

This link is for engineering students of Kalyani Government Engineering College

only. Student details will be visible in your gec_kalyani_campus login immediately)

Complete their registration process, Firstnaukri conducts recruitment drive in our

college, so go throung their website and get yourself registerd on their portal.

Thank You
Training and Placement Cell