﻿1
00:00:00,100 --> 00:00:00,900
Hello my friends.

2
00:00:00,900 --> 00:00:01,300
All right.

3
00:00:01,300 --> 00:00:02,933
So let's implement quickly

4
00:00:02,933 --> 00:00:06,733
and efficiently the linear regression
model on the whole data set.

5
00:00:06,966 --> 00:00:10,566
First a quick reminder on
what is the linear regression model

6
00:00:10,566 --> 00:00:11,766
that we're about to build.

7
00:00:11,766 --> 00:00:14,033
We're simply
about to build this one right.

8
00:00:14,033 --> 00:00:17,400
Because we have only one feature
which is the position levels

9
00:00:17,533 --> 00:00:21,033
and the dependent variable,
which is of course the salary to predict.

10
00:00:21,066 --> 00:00:21,866
All right.

11
00:00:21,866 --> 00:00:23,400
So that's what we'll do first.

12
00:00:23,400 --> 00:00:27,000
And then we'll of course implement
the polynomial regression model with

13
00:00:27,133 --> 00:00:31,733
I will tell you how many powers
of that same position level feature.

14
00:00:32,000 --> 00:00:32,366
Okay.

15
00:00:32,366 --> 00:00:36,100
So let's do this
starting by creating a new code cell.

16
00:00:36,733 --> 00:00:37,166
All right.

17
00:00:37,166 --> 00:00:41,066
So I hope you did it yourself first
to make sure you refresh your skills.

18
00:00:41,333 --> 00:00:46,766
And so if that's the case of course
you stole it from the scikit learn library

19
00:00:46,766 --> 00:00:50,100
because that's the library
that contains this linear

20
00:00:50,100 --> 00:00:53,100
regression class which allows to build
a linear regression model.

21
00:00:53,166 --> 00:00:55,400
So scikit learn from which we're going

22
00:00:55,400 --> 00:00:59,300
to get access to the linear model
which contains this class.

23
00:00:59,500 --> 00:01:02,400
And from this linear model
we're going to import

24
00:01:02,400 --> 00:01:06,766
the linear regression
class as Google Collab.

25
00:01:06,766 --> 00:01:09,433
Guess it's perfectly perfect.
So that's the first step.

26
00:01:09,433 --> 00:01:11,866
Then we're going to create
an object of this class.

27
00:01:11,866 --> 00:01:15,400
So I know we used to call it regressor,
but this time

28
00:01:15,400 --> 00:01:18,000
since we're going to have two regressors,
you know we're going to have

29
00:01:18,000 --> 00:01:21,666
the linear regression regressor
and the polynomial regression regressor.

30
00:01:21,666 --> 00:01:27,100
Therefore this time I'm going to call it
Lin rake for linear regressor okay.

31
00:01:27,266 --> 00:01:32,600
So Lin rake will be created as an object
of this linear regression class.

32
00:01:32,600 --> 00:01:36,066
So I'm copying and pasting that here
adding some parenthesis.

33
00:01:36,233 --> 00:01:39,666
And remember we don't have to input
anything in these parentheses

34
00:01:39,666 --> 00:01:42,866
because there is not much to tune
in the linear regression model.

35
00:01:43,633 --> 00:01:44,166
Good.

36
00:01:44,166 --> 00:01:47,466
And finally now we have,
you know, the linear regression model.

37
00:01:47,466 --> 00:01:49,100
But it is not smart yet.

38
00:01:49,100 --> 00:01:52,400
It is not trained yet on this data set.

39
00:01:52,633 --> 00:01:55,433
You know, to understand
and learn the correlations

40
00:01:55,433 --> 00:01:57,833
between the position levels
and the salaries.

41
00:01:57,833 --> 00:02:01,500
So that's what we're going to do right now
by using this fit method,

42
00:02:01,600 --> 00:02:06,666
which is exactly a training method
that will train the model on these data.

43
00:02:06,866 --> 00:02:07,700
All right.

44
00:02:07,700 --> 00:02:11,133
So to do this
we're going to call our object first Lin

45
00:02:11,466 --> 00:02:16,133
Rag from which we're going to call the
fit method

46
00:02:16,333 --> 00:02:20,933
which remember has to take as input
the couple of matrix

47
00:02:20,933 --> 00:02:24,700
of features and dependent variable
vector of the training set.

48
00:02:24,766 --> 00:02:26,000
Right. That's what we did before.

49
00:02:26,000 --> 00:02:30,200
But remember here that we didn't split
the data set into a training set.

50
00:02:30,200 --> 00:02:31,733
And it just said
because we want to leverage

51
00:02:31,733 --> 00:02:34,733
the maximum data 
in order to train our model.

52
00:02:34,866 --> 00:02:38,200
And therefore this time we're going
to take the whole matrix of features X

53
00:02:38,200 --> 00:02:40,433
and the whole dependent variable
vector y'all. Right.

54
00:02:40,433 --> 00:02:45,066
So here we just input x and y
and there we go.

55
00:02:45,100 --> 00:02:48,300
We built the linear
regression model in a flashlight.

56
00:02:48,300 --> 00:02:50,700
So we're going to execute this.

57
00:02:50,700 --> 00:02:51,600
Here we go.

58
00:02:51,600 --> 00:02:57,100
And now we have not only our model
but also a trained model on this data set

59
00:02:57,566 --> 00:02:58,700
okay great.

60
00:02:58,700 --> 00:03:00,000
Now we're going to close this.

61
00:03:00,000 --> 00:03:02,200
And we're going to focus on

62
00:03:02,200 --> 00:03:06,633
the heart of the matter
which is the polynomial regression model.

63
00:03:06,633 --> 00:03:10,233
I'm going to teach you now how to build
the polynomial regression model.

64
00:03:10,933 --> 00:03:13,000
All right.
So we're going to create a new code cell.

65
00:03:13,000 --> 00:03:14,133
And now

66
00:03:14,133 --> 00:03:17,966
we're going to go back to this slide
to explain exactly what we're going to do.

67
00:03:18,566 --> 00:03:22,400
All right so as you understand what
we've just built so far is this model.

68
00:03:22,500 --> 00:03:23,666
We only have one feature

69
00:03:23,666 --> 00:03:25,200
the position levels and the dependent

70
00:03:25,200 --> 00:03:28,200
variable vector to predict
which are the salaries.

71
00:03:28,200 --> 00:03:31,666
Now what we're going to do
is we're going to create a multiple linear

72
00:03:31,666 --> 00:03:32,833
regression model.

73
00:03:32,833 --> 00:03:36,600
But instead of having different features
you know, X1, x2 and Zn,

74
00:03:36,700 --> 00:03:42,300
well these features will be x1,
x1, squared and x1 are the power of n.

75
00:03:42,300 --> 00:03:46,866
And we'll actually tune this parameter
a bit to try several powers.

76
00:03:47,200 --> 00:03:49,466
So not to confuse the polynomial.

77
00:03:49,466 --> 00:03:54,100
Linear regression is not a linear model,
because you will see that it can learn

78
00:03:54,100 --> 00:03:57,600
some nonlinear correlations, but
we call it polynomial linear regression

79
00:03:57,600 --> 00:04:02,466
because indeed there is this linear
combination of the squared and,

80
00:04:02,466 --> 00:04:06,633
you know, powered features
x1, x1, squared and x1 and the power of n.

81
00:04:07,033 --> 00:04:07,700
All right.

82
00:04:07,700 --> 00:04:11,333
And so the process of building
this model on Python

83
00:04:11,500 --> 00:04:15,500
will be first to create
a matrix of the powered features,

84
00:04:15,500 --> 00:04:17,833
you know a matrix of features
but not containing

85
00:04:17,833 --> 00:04:22,200
different features like X1, x2, an x n,
but a matrix of features containing x1

86
00:04:22,200 --> 00:04:25,433
as a first feature,
then x1 squared as a second feature,

87
00:04:25,433 --> 00:04:28,533
and then x1 to the power of
and as an end feature.

88
00:04:28,800 --> 00:04:32,800
So that will be our matrix of features,
and we will call it x poly.

89
00:04:33,300 --> 00:04:35,966
And then we will create a linear

90
00:04:35,966 --> 00:04:38,966
regressor object,
you know, from the linear regression class

91
00:04:39,233 --> 00:04:43,200
to integrate these powered features
of this matrix of features

92
00:04:43,366 --> 00:04:46,600
in this new linear regressor,
you see the idea.

93
00:04:46,800 --> 00:04:48,666
So it's a building process in two steps.

94
00:04:48,666 --> 00:04:50,600
We're going to first create
the matrix of features

95
00:04:50,600 --> 00:04:53,600
containing these features
at different powers.

96
00:04:53,666 --> 00:04:57,566
And then we'll integrate that
into a linear regression model.

97
00:04:57,600 --> 00:05:01,433
Because indeed this is a linear
combination of these powered features.

98
00:05:02,066 --> 00:05:03,000
All right perfect.

99
00:05:03,000 --> 00:05:05,466
So let's do this first step.

100
00:05:05,466 --> 00:05:08,533
The first step is to actually import
this tool

101
00:05:08,533 --> 00:05:12,366
that will allow us to create this
matrix of powered features.

102
00:05:12,366 --> 00:05:15,366
You know X1X1 squared
and x one are the power of n.

103
00:05:15,466 --> 00:05:18,000
And this class is called
polynomial features.

104
00:05:18,000 --> 00:05:19,466
So we're going to import it first

105
00:05:19,466 --> 00:05:24,000
which we can get from once again
of course the scikit learn library.

106
00:05:24,233 --> 00:05:25,166
There we go.

107
00:05:25,166 --> 00:05:30,133
But this time we're going to get access
to the pre-processing module

108
00:05:30,133 --> 00:05:32,766
which contains that polynomial features
class.

109
00:05:32,766 --> 00:05:37,666
This is a class because indeed
we're kind of preprocessing our X1 feature

110
00:05:37,666 --> 00:05:43,433
because we want to create out of x1 well
x1, x1 squared and x1 add the power of n.

111
00:05:43,433 --> 00:05:46,433
So we're kind of doing
preprocessing of the features.

112
00:05:46,700 --> 00:05:47,066
All right.

113
00:05:47,066 --> 00:05:49,900
So from this preprocessing module
we're going to import.

114
00:05:49,900 --> 00:05:54,533
Well that class put Li no more features.

115
00:05:54,533 --> 00:05:54,900
Perfect.
