1
00:00:00,333 --> 00:00:00,900
All right.

2
00:00:00,900 --> 00:00:03,233
And now time for the little extra bonus.

3
00:00:03,233 --> 00:00:05,500
Let's plot the decision tree.

4
00:00:05,500 --> 00:00:08,800
So in order to have the best
interpretation, we are going to remove

5
00:00:08,900 --> 00:00:10,133
the feature scaling.

6
00:00:10,133 --> 00:00:15,233
And to do this we're going to clear
everything and re-execute the code.

7
00:00:15,533 --> 00:00:18,166
But without features scaling.
So let's do this.

8
00:00:18,166 --> 00:00:22,600
We're going to clear the data here
by clicking on this button here.

9
00:00:22,600 --> 00:00:26,966
And yes then clear the plots here. Yes.

10
00:00:27,633 --> 00:00:30,966
And then clear the console as well
by typing Ctrl L.

11
00:00:31,266 --> 00:00:32,833
Here we go. All right.

12
00:00:32,833 --> 00:00:34,000
Now everything is clear.

13
00:00:34,000 --> 00:00:36,933
We can re-execute the whole script.

14
00:00:36,933 --> 00:00:38,600
So let's do it step by step.

15
00:00:38,600 --> 00:00:42,366
We're first
going to take all the pre-processing step

16
00:00:42,366 --> 00:00:44,866
but without the feature scaling here.

17
00:00:44,866 --> 00:00:47,033
As you can see I'm not selecting this.

18
00:00:47,033 --> 00:00:47,400
All right.

19
00:00:47,400 --> 00:00:50,100
So now command and control
plus enter to execute.

20
00:00:51,066 --> 00:00:52,033
All good.

21
00:00:52,033 --> 00:00:55,966
And now you can see that
if we go to our data set training

22
00:00:55,966 --> 00:01:01,000
set and test set you can see that
the features are no longer scaled.

23
00:01:01,600 --> 00:01:04,633
All right so for example
in the training set we have the real age.

24
00:01:04,666 --> 00:01:10,500
I mean with their real values and the real
estimated salaries the real values okay.

25
00:01:10,500 --> 00:01:11,433
So nothing scaled.

26
00:01:11,433 --> 00:01:17,333
Now let's select this to fit the decision
tree classifier to the training set.

27
00:01:17,400 --> 00:01:20,400
So command and control
plus enter to execute.

28
00:01:20,533 --> 00:01:22,700
Here we go. Now it's fitted.

29
00:01:22,700 --> 00:01:26,033
By the way
I'm just going to replace this by

30
00:01:27,566 --> 00:01:30,500
decision tree

31
00:01:30,500 --> 00:01:33,100
classification.

32
00:01:33,100 --> 00:01:34,633
All right.

33
00:01:34,633 --> 00:01:38,733
Which we are really interested in
right now is to plot the decision tree.

34
00:01:38,766 --> 00:01:41,766
So we're going to add a new section here
below.

35
00:01:42,066 --> 00:01:44,566
And I'm going
to call this section plotting

36
00:01:45,933 --> 00:01:48,933
the Decision Tree.

37
00:01:49,600 --> 00:01:50,333
And as I

38
00:01:50,333 --> 00:01:54,533
told you in the Python tutorial
this is going to take two lines.

39
00:01:54,833 --> 00:01:55,933
This is very simple.

40
00:01:55,933 --> 00:01:57,100
So let's just do it.

41
00:01:57,100 --> 00:01:59,800
It's going to be very quick okay.

42
00:01:59,800 --> 00:02:03,566
And besides the two lines that we need to
write could not be more simple.

43
00:02:03,833 --> 00:02:06,900
Because indeed what we need to type
here is plot.

44
00:02:07,366 --> 00:02:10,366
And then in parenthesis classifier

45
00:02:10,700 --> 00:02:16,000
because you know the classifier here
is the classifier that we created here

46
00:02:16,000 --> 00:02:19,966
in this part using the r part
function of the R library.

47
00:02:20,533 --> 00:02:23,800
And that's all we just need
to type plot classifier.

48
00:02:23,800 --> 00:02:25,400
And that will plot the tree.

49
00:02:25,400 --> 00:02:29,100
But without the labels
without the conditions written explicitly.

50
00:02:29,733 --> 00:02:33,200
So in order to add these conditions
written explicitly,

51
00:02:33,533 --> 00:02:36,200
we just need to add below text

52
00:02:37,266 --> 00:02:38,700
classifier.

53
00:02:38,700 --> 00:02:40,533
Here we go. And now it's ready.

54
00:02:40,533 --> 00:02:43,100
Isn't that simple? There's only two lines.

55
00:02:43,100 --> 00:02:46,066
We'll plot an interpretable decision tree.

56
00:02:46,066 --> 00:02:47,533
Let's check it out.

57
00:02:47,533 --> 00:02:51,600
I'm going to select all this and press

58
00:02:51,600 --> 00:02:54,600
Command and Control plus enter to execute.

59
00:02:55,333 --> 00:02:56,800
And here is the tree.

60
00:02:56,800 --> 00:02:59,633
So as you can see we have at each

61
00:02:59,633 --> 00:03:02,633
split the condition
that is generating the splits.

62
00:03:02,700 --> 00:03:06,300
So for example the first split is made

63
00:03:06,433 --> 00:03:11,366
based upon the condition age
lower than 44.5 years old.

64
00:03:11,866 --> 00:03:17,100
So that means that if the user is below
44.5 years old,

65
00:03:17,133 --> 00:03:20,266
he will go to this subcategory
after the split.

66
00:03:20,666 --> 00:03:23,766
And if the user is older than 44.5
years old,

67
00:03:24,100 --> 00:03:27,100
he will end up
in this subcategory of this split.

68
00:03:27,866 --> 00:03:29,533
And then we have some new conditions.

69
00:03:29,533 --> 00:03:33,233
So here and you condition on the other
independent variable.

70
00:03:33,266 --> 00:03:36,366
The estimated salary in this condition
is estimated

71
00:03:36,366 --> 00:03:39,866
salary below $90,000.

72
00:03:40,200 --> 00:03:45,300
So what that means is that
if the user is younger than 44.5 years old

73
00:03:45,600 --> 00:03:51,266
and has an estimated salary below $90,000,
then according to our decision tree

74
00:03:51,266 --> 00:03:56,333
classifier, this user won't buy the SUV
because the result here is zero.

75
00:03:57,033 --> 00:04:00,300
And if the user is younger than 44.5

76
00:04:00,300 --> 00:04:03,300
years old and has an estimated salary over

77
00:04:03,300 --> 00:04:07,266
$90,000, then
according to our decision tree classifier,

78
00:04:07,366 --> 00:04:11,000
this user will buy the SUV
because the result here is one.

79
00:04:11,700 --> 00:04:12,033
Okay.

80
00:04:12,033 --> 00:04:14,566
And if we go to the other side
of the tree,

81
00:04:14,566 --> 00:04:18,500
well, this other side of the tree
first contains all the users.

82
00:04:18,700 --> 00:04:21,500
There are older than 44.5 years old.

83
00:04:21,500 --> 00:04:25,033
And then we have some new conditions
generating new splits.

84
00:04:25,066 --> 00:04:27,233
So another condition on the age here,

85
00:04:27,233 --> 00:04:29,833
then another condition
on the estimated salary.

86
00:04:29,833 --> 00:04:33,666
And then again another condition on the
estimated salary and following yes or no.

87
00:04:33,666 --> 00:04:36,900
This conditions
we end up in some final categories.

88
00:04:36,900 --> 00:04:41,100
The final nodes of the decision tree,
where the user is predicted

89
00:04:41,100 --> 00:04:45,733
not to buy the SUV for this node
and predicted to buy the SUV

90
00:04:45,900 --> 00:04:48,833
for this node and this node and this node.

91
00:04:48,833 --> 00:04:51,666
Okay, so that was worth looking at this.

92
00:04:51,666 --> 00:04:56,000
We sort of explored this decision tree
classifier behind the scenes.

93
00:04:56,233 --> 00:04:59,700
And what's important to understand
now when you are using decision trees

94
00:04:59,700 --> 00:05:02,933
is that a big advantage of this classifier
is that we can have

95
00:05:02,933 --> 00:05:06,066
these very interpretable results.

96
00:05:06,066 --> 00:05:10,300
Because here on this decision tree plot,
we can see everything that's happening.

97
00:05:10,300 --> 00:05:13,466
We can see how the decision tree decides

98
00:05:13,533 --> 00:05:18,266
whether a user will be predicted
to not buy the SUV or to buy the SUV.

99
00:05:18,300 --> 00:05:23,100
We clearly see the whole thinking process
of the decision tree, and in some way,

100
00:05:23,100 --> 00:05:26,333
we saw how it learned from the data

101
00:05:26,500 --> 00:05:29,500
how to classify
each of our user of the social network.

102
00:05:29,900 --> 00:05:30,266
All right.

103
00:05:30,266 --> 00:05:33,600
So I look forward to seeing you
in the next section about Random Forest.

104
00:05:33,900 --> 00:05:36,700
We will implement
Random Forest on Python in R,

105
00:05:36,700 --> 00:05:39,700
and I can't wait to show you
the final graphic results.

106
00:05:39,866 --> 00:05:41,766
Until next time, enjoy machine learning.