﻿1
00:00:00,166 --> 00:00:01,800
Hello my friends, and welcome

2
00:00:01,800 --> 00:00:06,366
to this new practical activity
for Decision Tree Classification.

3
00:00:06,766 --> 00:00:10,166
And now please follow me into part
three classification.

4
00:00:10,200 --> 00:00:14,933
And to implement this time
the decision tree classification model.

5
00:00:15,366 --> 00:00:17,733
And as usual
we're going to start with Python.

6
00:00:17,733 --> 00:00:20,033
This Python folder contains two files.

7
00:00:20,033 --> 00:00:24,733
The implementation of the decision tree
classification model in ipynb format,

8
00:00:25,000 --> 00:00:28,566
and the same data
set social network at the CSV,

9
00:00:28,800 --> 00:00:33,000
which contains information
of 400 customers of a car dealership

10
00:00:33,233 --> 00:00:36,233
which just released
a brand new luxury SUV.

11
00:00:36,300 --> 00:00:39,233
And the strategic team
gathered this data to understand

12
00:00:39,233 --> 00:00:42,300
which customers buy the most SUV in order

13
00:00:42,300 --> 00:00:46,200
to target them with ads
that will be posted on social networks.

14
00:00:46,400 --> 00:00:49,566
So each of these rows correspond
to different customers

15
00:00:49,766 --> 00:00:51,166
and for each of these customers,

16
00:00:51,166 --> 00:00:54,466
we have two features
the age and the estimated salary with

17
00:00:54,466 --> 00:00:58,666
which we're going to predict the dependent
variable purchased, which takes

18
00:00:58,666 --> 00:01:02,766
binary values zero, meaning
that the customer didn't buy the SUV,

19
00:01:03,066 --> 00:01:06,766
and one meaning that the customer
but the SUV.

20
00:01:07,300 --> 00:01:08,700
All right, let's do this.

21
00:01:08,700 --> 00:01:12,000
Let's see if decision tree
classification can beat

22
00:01:12,133 --> 00:01:15,566
the previous accuracy
we got so far, meaning 93%.

23
00:01:15,900 --> 00:01:19,533
So without further ado, let's open
that with Google Colab.

24
00:01:19,700 --> 00:01:20,966
Choose your favorite.

25
00:01:20,966 --> 00:01:25,100
And I'm going to put it here because
this is our next classification model.

26
00:01:25,600 --> 00:01:26,633
All right. Perfect.

27
00:01:26,633 --> 00:01:31,133
So once again this implementation results
from that classification template

28
00:01:31,133 --> 00:01:33,233
we made in the first section of this part
three

29
00:01:33,233 --> 00:01:36,233
when we implemented
the logistic regression model.

30
00:01:36,300 --> 00:01:40,366
And therefore the only cell
that changes here in this implementation

31
00:01:40,366 --> 00:01:46,400
is, as usual, that cell where we build
and train the classification model.

32
00:01:46,600 --> 00:01:47,266
Right.

33
00:01:47,266 --> 00:01:52,133
This cell here and all the rest, you know,
all the other cells are exactly the same

34
00:01:52,133 --> 00:01:55,366
as in the logistic regression
code template

35
00:01:55,366 --> 00:01:57,433
or you know, the classification
code template,

36
00:01:57,433 --> 00:02:00,166
because indeed
we use the same variable names classifier.

37
00:02:00,166 --> 00:02:03,800
And then x train x test
everything is the same except the cell.

38
00:02:04,066 --> 00:02:07,033
So once again
we're going to re-implement this cell.

39
00:02:07,033 --> 00:02:11,700
But to do this we need to create a copy
because this is in read only mode.

40
00:02:11,966 --> 00:02:14,966
So right now
we're going to create that copy

41
00:02:15,000 --> 00:02:18,100
on which we will be able to re-implement
that cell.

42
00:02:18,300 --> 00:02:18,966
All right.

43
00:02:18,966 --> 00:02:20,966
So let's scroll down again

44
00:02:20,966 --> 00:02:25,000
to find that cell where we build and train
the decision tree classification model.

45
00:02:25,166 --> 00:02:29,700
Now let's put it in the trash immediately
because I don't want you to see,

46
00:02:29,700 --> 00:02:33,533
you know, the name of the class, because
indeed I want you to find it on your own.

47
00:02:33,733 --> 00:02:35,166
So there you go.

48
00:02:35,166 --> 00:02:40,300
Now is the time you press pause
on this video to find the right class

49
00:02:40,300 --> 00:02:44,333
name that allows us to build
the decision tree classification model.

50
00:02:44,633 --> 00:02:48,200
You will have to either look for it
directly through an online search

51
00:02:48,200 --> 00:02:52,900
or that other option is to look for it
inside the scikit learn API.

52
00:02:52,900 --> 00:02:55,900
And this is exactly
what we're going to do together.

53
00:02:56,066 --> 00:02:58,066
And all right let's do this.

54
00:02:58,066 --> 00:03:02,100
So let's go to the scikit
learn API to find the class

55
00:03:02,100 --> 00:03:05,766
that allows us to build a decision tree
classification model.

56
00:03:06,066 --> 00:03:06,433
All right.

57
00:03:06,433 --> 00:03:08,000
So API

58
00:03:08,000 --> 00:03:13,466
and by memory you know let's suppose
actually that I have no idea of the module

59
00:03:13,466 --> 00:03:17,033
or even the class that built this decision
tree classification model.

60
00:03:17,166 --> 00:03:21,000
So here I'm scrolling down, you know,
to observe the different modules.

61
00:03:21,000 --> 00:03:24,100
Maybe it is an ensemble method
because you know, I know that

62
00:03:24,100 --> 00:03:27,833
random forest we built, you know,
the Random Forest Regressor in part two.

63
00:03:28,000 --> 00:03:30,233
Random forest is an in simple method.

64
00:03:30,233 --> 00:03:32,333
But no, here
we don't see the decision tree.

65
00:03:32,333 --> 00:03:35,333
So let's scroll back down again.

66
00:03:35,400 --> 00:03:36,333
Which makes sense, right?

67
00:03:36,333 --> 00:03:38,333
Because the decision tree
is just a single model.

68
00:03:38,333 --> 00:03:40,800
It is not an ensemble of models.

69
00:03:40,800 --> 00:03:43,800
All right.
So many learning metrics. No, no.

70
00:03:43,966 --> 00:03:44,866
You know it's really important

71
00:03:44,866 --> 00:03:47,866
that you get familiar with this API
because the more you get familiar,

72
00:03:48,000 --> 00:03:50,933
the more expert you'll become
and the better you'll juggle

73
00:03:50,933 --> 00:03:51,833
with all the machine

74
00:03:51,833 --> 00:03:55,100
learning tools you know, besides the one
that I'm giving you in this course.

75
00:03:55,600 --> 00:03:57,666
All right,
so here I'm scrolling down more.

76
00:03:57,666 --> 00:03:57,966
All right.

77
00:03:57,966 --> 00:04:01,200
Preprocessing,
random projection, semi-supervised

78
00:04:01,200 --> 00:04:04,266
learning,
support vector machines, decision trees.

79
00:04:04,266 --> 00:04:05,433
There we go.

80
00:04:05,433 --> 00:04:08,500
You know, usually the module
you're looking for in the circular

81
00:04:08,533 --> 00:04:11,833
API will have the same name here
as the model you want to build.

82
00:04:12,033 --> 00:04:12,500
Right.

83
00:04:12,500 --> 00:04:15,266
And you know,
this is organized in alphabetical order.

84
00:04:15,266 --> 00:04:18,000
So that's why I wanted you to do this
exercise.

85
00:04:18,000 --> 00:04:20,966
It's actually very easy
to find what you want.

86
00:04:20,966 --> 00:04:22,100
All right decision trees.

87
00:04:22,100 --> 00:04:24,700
Now which one
do we want to get among these.

88
00:04:24,700 --> 00:04:26,266
Well that's of course the first one.

89
00:04:26,266 --> 00:04:28,866
You know, that was the one we used in part
two regression.

90
00:04:28,866 --> 00:04:30,466
But now we're doing classification.

91
00:04:30,466 --> 00:04:33,633
So we want the decision tree
classifier class.

92
00:04:33,700 --> 00:04:34,233
All right.

93
00:04:34,233 --> 00:04:36,366
So now what do we have to do first.

94
00:04:36,366 --> 00:04:38,700
Well let's do you know
what do we have to do

95
00:04:38,700 --> 00:04:41,433
anyway which is to import that class.

96
00:04:41,433 --> 00:04:42,966
And so I'm copying this.

97
00:04:42,966 --> 00:04:46,733
And then back in my copy
we're going to create a new code cell.

98
00:04:46,966 --> 00:04:48,233
We're going to paste that.

99
00:04:48,233 --> 00:04:51,766
And once again
we're going to start from scikit learn.

100
00:04:51,766 --> 00:04:54,500
And then the tree module by scikit learn.

101
00:04:54,500 --> 00:04:57,500
And from that tree module
we're going to import

102
00:04:57,600 --> 00:05:00,233
the decision tree classifier.

103
00:05:00,233 --> 00:05:02,066
So that's the first step right.

104
00:05:02,066 --> 00:05:03,533
You know it by heart. Now.

105
00:05:03,533 --> 00:05:06,166
And now the second step
I don't even have to tell you.

106
00:05:06,166 --> 00:05:09,533
It is
of course to create our classifier object,

107
00:05:09,900 --> 00:05:13,266
which will be created
as an instance of this decision

108
00:05:13,266 --> 00:05:16,566
tree classifier class
and which will represent nothing else.

109
00:05:16,566 --> 00:05:19,566
Then the decision tree classifier model.

110
00:05:19,933 --> 00:05:22,366
All right.
So here we add some parenthesis.

111
00:05:22,366 --> 00:05:27,600
And now the second question is what
do we have to input here as parameters.

112
00:05:27,600 --> 00:05:29,600
You know if this decision tree
classifier class.

113
00:05:29,600 --> 00:05:30,600
So let's see.

114
00:05:30,600 --> 00:05:33,666
Let's go back to the documentation
to scikit learn API.

115
00:05:34,000 --> 00:05:37,566
And well we will just change one
you know default

116
00:05:37,566 --> 00:05:41,066
value of the parameters
which is this one criterion.

117
00:05:41,333 --> 00:05:44,600
So the default value of criterion is Gini.

118
00:05:45,200 --> 00:05:48,466
But with respect to what you learned
in the theory with key rules

119
00:05:48,466 --> 00:05:52,300
intuition lecture, we're going to choose
an entropy criterion.

120
00:05:52,300 --> 00:05:56,100
So the criterion is of course the function
to measure the quality of a splits.

121
00:05:56,366 --> 00:05:58,966
And that quality is measures by entropy.
