1
00:00:00,366 --> 00:00:01,300
Hello my friends.

2
00:00:01,300 --> 00:00:02,766
I hope you're feeling fantastic.

3
00:00:02,766 --> 00:00:03,933
Now, after this,

4
00:00:03,933 --> 00:00:07,800
first data preprocessing phase
that we tackled in a flashlight together.

5
00:00:08,100 --> 00:00:12,033
Now we're ready for the exciting step,
which is to train the multiple linear

6
00:00:12,033 --> 00:00:12,766
regression model

7
00:00:12,766 --> 00:00:16,600
on the training set that we now collected
thanks to the previous step.

8
00:00:16,966 --> 00:00:21,433
But first, before we do this, I would like
to answer two important questions.

9
00:00:21,800 --> 00:00:25,466
The first question is do we have to avoid,
you know,

10
00:00:25,466 --> 00:00:29,066
to do something
to avoid the dummy variable trap?

11
00:00:29,666 --> 00:00:34,233
And the answer is no,
because indeed the multiple linear

12
00:00:34,233 --> 00:00:37,366
regression class
that we're about to import

13
00:00:37,366 --> 00:00:40,966
and which will build the multiple linear
regression model itself

14
00:00:40,966 --> 00:00:44,366
and actually also train it,
because remember that a class actually

15
00:00:44,366 --> 00:00:48,266
can complete several actions,
including building it and training it.

16
00:00:48,600 --> 00:00:52,533
And well, this class
will automatically avoid this trap.

17
00:00:52,800 --> 00:00:57,000
Which means that indeed, you know
one of the three first columns here.

18
00:00:57,066 --> 00:01:00,066
Because remember, Kirill explained
that one is redundant.

19
00:01:00,133 --> 00:01:04,366
Well, will automatically be outcast,
so you have nothing to worry about

20
00:01:04,500 --> 00:01:06,066
regarding the dummy variable trap.

21
00:01:06,066 --> 00:01:08,766
You don't have to remove
one of the columns here.

22
00:01:08,766 --> 00:01:10,300
And that's the beauty of classes.

23
00:01:10,300 --> 00:01:14,000
You know, these are advanced
implementations that allow you to build

24
00:01:14,000 --> 00:01:17,066
a model and machine learning model
in just a few lines of code,

25
00:01:17,100 --> 00:01:20,100
and this will take care of the dummy
variable trap for you.

26
00:01:20,600 --> 00:01:25,600
Now second question
do we have to work on the features

27
00:01:25,600 --> 00:01:28,666
to select the best ones
with, you know, the techniques

28
00:01:28,666 --> 00:01:32,500
that Kirill introduced to you,
like for example, backward elimination.

29
00:01:32,766 --> 00:01:35,766
Do we have to deploy the backward
elimination technique

30
00:01:35,933 --> 00:01:39,066
in order to select the features
that have the highest p values?

31
00:01:39,066 --> 00:01:42,066
And and there are
the most artistically significant.

32
00:01:42,533 --> 00:01:45,300
And the answer is once again no.

33
00:01:45,300 --> 00:01:46,100
Why is that?

34
00:01:46,100 --> 00:01:49,900
Well, for the exact same reason
as the dummy variable trap,

35
00:01:50,166 --> 00:01:54,466
the class that we're about to call
to build our multiple linear regression

36
00:01:54,466 --> 00:01:59,233
model will automatically identify
the best features.

37
00:01:59,233 --> 00:02:00,000
You know, the features

38
00:02:00,000 --> 00:02:03,900
that have the highest p values, or
that are the most artistically significant

39
00:02:04,166 --> 00:02:07,066
to figure out
how to predict the dependent variable.

40
00:02:07,066 --> 00:02:09,966
You know, the profit
with the highest accuracy.

41
00:02:09,966 --> 00:02:12,866
So once again,
you don't have to worry about this.

42
00:02:12,866 --> 00:02:14,666
The class of scikit learn.

43
00:02:14,666 --> 00:02:16,900
You know, this amazing
data science library

44
00:02:16,900 --> 00:02:19,800
will take care of everything for you.
Okay.

45
00:02:19,800 --> 00:02:22,300
So I'm glad to answer these two questions.

46
00:02:22,300 --> 00:02:25,833
You know, the purpose of building machine
learning models today is to be efficient

47
00:02:26,100 --> 00:02:29,766
because in your career, you know,
I can't tell you how many times I had

48
00:02:29,766 --> 00:02:33,600
to test several machine learning models
on my data sets and select the best one.

49
00:02:33,600 --> 00:02:36,633
Well, you know, if I had to deploy
the backward elimination

50
00:02:36,633 --> 00:02:39,800
technique on my data set, well,
I would have lost a lot of time.

51
00:02:39,900 --> 00:02:42,500
And here we have a class
that takes care of everything.

52
00:02:42,500 --> 00:02:43,800
You just have to deploy

53
00:02:43,800 --> 00:02:46,866
your class on your data set,
and then you will get an accuracy.

54
00:02:46,866 --> 00:02:50,966
And you will compare that accuracy
with the accuracies of other models.

55
00:02:51,000 --> 00:02:53,533
And you know, that process is called model
selection.

56
00:02:53,533 --> 00:02:56,533
That's what we actually cover
in the portion of this course.

57
00:02:56,833 --> 00:02:58,800
So really I want you to be efficient with,

58
00:02:58,800 --> 00:03:01,800
you know, your implementations
and your machine learning toolkit

59
00:03:02,033 --> 00:03:05,033
so that you can optimize your model
selection process.

60
00:03:05,500 --> 00:03:05,966
All right.

61
00:03:05,966 --> 00:03:07,266
So now that we said this

62
00:03:07,266 --> 00:03:10,800
well let's build together
the multiple linear regression model.

63
00:03:11,366 --> 00:03:14,066
And now actually I have some good news

64
00:03:14,066 --> 00:03:17,466
because you know
we're still doing linear regression.

65
00:03:17,800 --> 00:03:20,966
And in the previous section
we actually did simple linear regression

66
00:03:21,300 --> 00:03:24,166
with which we did
for had one feature in our data set.

67
00:03:24,166 --> 00:03:27,266
And now we're doing multiple linear
regression, which is exactly the same

68
00:03:27,266 --> 00:03:30,266
as simple linear regression
except that we have several features.

69
00:03:30,600 --> 00:03:33,600
So the good news is actually that
the class

70
00:03:33,600 --> 00:03:37,700
that we're about to use to build and train
this multiple linear regression model

71
00:03:37,900 --> 00:03:41,466
is actually the exact same class
as for the simple linear regression model.

72
00:03:41,466 --> 00:03:43,800
It will just recognize
that there are several features,

73
00:03:43,800 --> 00:03:46,600
and therefore that we are doing
multiple linear regression,

74
00:03:46,600 --> 00:03:48,500
but the rest will be exactly the same.

75
00:03:48,500 --> 00:03:52,433
It will be trained to understand the
correlations between all your features.

76
00:03:52,433 --> 00:03:53,833
Actually I can open it here.

77
00:03:53,833 --> 00:03:58,533
All your features and the profit
which is your dependent variable.

78
00:03:58,800 --> 00:04:00,900
And then it will take care
of the dummy variable.

79
00:04:00,900 --> 00:04:04,066
And it will also take care of selecting
the best features

80
00:04:04,066 --> 00:04:07,066
that are
the most statistically significant.

81
00:04:07,133 --> 00:04:08,000
So that's the good news.

82
00:04:08,000 --> 00:04:11,166
But still I just want
to re-implement this again from scratch

83
00:04:11,166 --> 00:04:14,400
to make sure it is
best integrated in your head.

84
00:04:14,700 --> 00:04:15,000
All right.

85
00:04:15,000 --> 00:04:18,866
So let's close this and let's implement
our multiple linear regression model.

86
00:04:19,433 --> 00:04:21,200
I actually tried to do it faster than me.

87
00:04:21,200 --> 00:04:22,200
Try to do it before me.

88
00:04:22,200 --> 00:04:24,966
You know you can press pause on this video
and do it.

89
00:04:24,966 --> 00:04:27,766
And me, I'm going to do it
pretty efficiently.

90
00:04:27,766 --> 00:04:31,800
So remember
we have to start from the sklearn,

91
00:04:31,800 --> 00:04:34,933
the scikit learn library
from which we're going to get access

92
00:04:34,933 --> 00:04:38,633
to this specific module
which is linear model.

93
00:04:38,666 --> 00:04:41,600
There
we go. Google Colab guesses it perfectly.

94
00:04:41,600 --> 00:04:44,600
And from this module
we're going to import it.

95
00:04:44,633 --> 00:04:48,166
Well the linear regression

96
00:04:48,166 --> 00:04:51,533
class Google Collab
guess is again perfectly all right.

97
00:04:51,533 --> 00:04:52,300
So that's the exact

98
00:04:52,300 --> 00:04:55,466
same class as in the previous section
on simple linear regression.

99
00:04:55,833 --> 00:04:56,466
And there you go.

100
00:04:56,466 --> 00:04:58,133
We're going to do just exactly the same.

101
00:04:58,133 --> 00:05:01,500
You know, this is exactly the same code
as what we did in the previous section,

102
00:05:01,766 --> 00:05:06,066
because now we're going to create a new
variable which will be our multiple linear

103
00:05:06,066 --> 00:05:07,100
regression model,

104
00:05:07,100 --> 00:05:10,866
which will be created as an object
of this linear regression class.

105
00:05:11,100 --> 00:05:15,700
So let's introduce
this new variable regressor equals.

106
00:05:15,900 --> 00:05:19,800
Well since this will be an object
of the linear regression class, I'm

107
00:05:19,800 --> 00:05:25,266
just copying this class and pasting it
here and adding some parenthesis.

108
00:05:25,266 --> 00:05:25,633
All right.

109
00:05:25,633 --> 00:05:30,100
So that regressor is created as an
instance of this linear regression class.

110
00:05:30,666 --> 00:05:33,666
Now the question is
do we have to enter any parameters here.

111
00:05:33,933 --> 00:05:37,633
Well just like simple linear regression
the answer is no.

112
00:05:37,733 --> 00:05:38,800
We are just going to keep

113
00:05:38,800 --> 00:05:42,666
the default values of the parameters
inside this linear regression class.

114
00:05:43,066 --> 00:05:45,866
I will explain in a whole section
parameter tuning.

115
00:05:45,866 --> 00:05:47,633
You know when you can improve your model.

116
00:05:47,633 --> 00:05:49,500
But for linear regression
it's pretty simple.

117
00:05:49,500 --> 00:05:51,900
So usually we don't
have anything to input here.