1
00:00:00,066 --> 00:00:01,733
And now let's check it out.

2
00:00:01,733 --> 00:00:02,866
The model is actually ready.

3
00:00:02,866 --> 00:00:05,900
We can actually replace that
by decision tree

4
00:00:05,900 --> 00:00:08,900
regression.

5
00:00:09,866 --> 00:00:12,266
And now let's check it out
and you will see

6
00:00:12,266 --> 00:00:16,233
what the real decision tree regression
model looks like in 1D.

7
00:00:16,500 --> 00:00:18,000
So let's do it.

8
00:00:18,000 --> 00:00:21,000
Let's select all this and execute.

9
00:00:21,533 --> 00:00:23,266
And here it is.

10
00:00:23,266 --> 00:00:24,500
That's what it looks like.

11
00:00:24,500 --> 00:00:25,200
And actually

12
00:00:25,200 --> 00:00:28,733
since it's a non continuous model
we should even have some strict

13
00:00:28,800 --> 00:00:29,933
vertical line here.

14
00:00:29,933 --> 00:00:32,933
You know to represent
better than non continuity.

15
00:00:32,933 --> 00:00:35,866
And to do this
we just need to increase the resolution.

16
00:00:35,866 --> 00:00:39,000
So I'm going to put 0.01
and you'll see that

17
00:00:39,000 --> 00:00:42,800
we'll get the real representation
of the decision tree regression results.

18
00:00:43,366 --> 00:00:45,766
So let's do it.

19
00:00:45,766 --> 00:00:47,700
Here it is now almost vertical.

20
00:00:47,700 --> 00:00:51,166
And that's a clear representation
of the decision tree regression model.

21
00:00:51,466 --> 00:00:53,033
So now let's zoom on it.

22
00:00:53,033 --> 00:00:57,700
And now it makes much more sense
because as Carol explained

23
00:00:57,700 --> 00:01:00,800
based on the entropy
and the information gain, it splits

24
00:01:00,800 --> 00:01:04,333
the whole range of your independent
variable into different intervals.

25
00:01:04,500 --> 00:01:07,166
So here we can clearly see
where the intervals are.

26
00:01:07,166 --> 00:01:10,400
The first interval is from 1 to 6.5.

27
00:01:10,666 --> 00:01:14,533
The second interval is from 6.5 to 8.5.

28
00:01:14,566 --> 00:01:18,866
Then the third interval
is from 8.5 to 9.5,

29
00:01:19,033 --> 00:01:22,533
and finally the last interval
from 9.5 to 10.

30
00:01:22,833 --> 00:01:23,700
So there we go.

31
00:01:23,700 --> 00:01:25,633
We can clearly see the intervals now.

32
00:01:25,633 --> 00:01:30,133
And as Carol explained in the intuition
tutorial, the decision tree regression

33
00:01:30,133 --> 00:01:33,900
model is considering the average
of the dependent variable values

34
00:01:34,000 --> 00:01:35,766
in each of the intervals.

35
00:01:35,766 --> 00:01:38,833
This one, 
this one, this one, and this one.

36
00:01:39,000 --> 00:01:39,733
So for example,

37
00:01:39,733 --> 00:01:44,100
if we consider this interval here, the
average of the salaries in this interval.

38
00:01:44,100 --> 00:01:45,100
Well that's very simple.

39
00:01:45,100 --> 00:01:47,966
It's actually 250,000.

40
00:01:47,966 --> 00:01:52,500
And so for each of the level
between 6.5 and 8.5,

41
00:01:52,600 --> 00:01:56,766
the salaries will be predicted
to be $250,000.

42
00:01:56,833 --> 00:02:02,400
So we already know what our model
will predict for our 6.5 level here.

43
00:02:02,400 --> 00:02:05,233
It's going to predict 250 K.

44
00:02:05,233 --> 00:02:06,666
So speaking of this prediction.

45
00:02:06,666 --> 00:02:10,433
And now that we get the decision tree
regression graphic results very well,

46
00:02:10,600 --> 00:02:14,666
let's actually check
that the previous salary of this employee

47
00:02:14,666 --> 00:02:19,933
that had a 6.5 level
in its previous company is actually 250 K.

48
00:02:20,366 --> 00:02:21,600
Let's check it out.

49
00:02:21,600 --> 00:02:24,200
Let's select this line and execute.

50
00:02:24,200 --> 00:02:29,066
And here it is
250,000 exactly like we predicted

51
00:02:29,100 --> 00:02:32,000
because we can clearly see
that on this plot.

52
00:02:32,000 --> 00:02:34,333
So now we just want to say two things
to conclude.

53
00:02:34,333 --> 00:02:38,100
The decision tree regression
model is not an interesting model in 1D,

54
00:02:38,333 --> 00:02:42,233
but it can be a very interesting and
very powerful model in more dimensions.

55
00:02:42,566 --> 00:02:45,600
So that's why you can use this code here
for your data set.

56
00:02:45,700 --> 00:02:47,633
You have the code here
that builds the model.

57
00:02:47,633 --> 00:02:49,666
And this one here then makes a prediction.

58
00:02:49,666 --> 00:02:51,433
But then you won't be able
to use this code

59
00:02:51,433 --> 00:02:52,800
because you will have probably

60
00:02:52,800 --> 00:02:55,733
a lot of independent variables
and therefore a lot of dimensions.

61
00:02:55,733 --> 00:02:58,066
But you know, care
will give you the explanation in 2D.

62
00:02:58,066 --> 00:03:01,200
I'm giving you the explanation of 1D
so that now in your mind,

63
00:03:01,200 --> 00:03:03,400
you can perfectly represent in your head

64
00:03:03,400 --> 00:03:06,000
the decision tree regression
and how it works.

65
00:03:06,000 --> 00:03:08,800
And now I would like to end this tutorial
by an enigma.

66
00:03:08,800 --> 00:03:12,633
In the next section
you will see random forests and a random

67
00:03:12,633 --> 00:03:14,133
forest is actually really simple.

68
00:03:14,133 --> 00:03:17,100
It's just a team of several decision
trees.

69
00:03:17,100 --> 00:03:20,100
So knowing
that this is the result of one tree,

70
00:03:20,300 --> 00:03:23,500
what do you think
will get with a team of ten trees

71
00:03:23,833 --> 00:03:26,766
or even 100 trees or 500 trees?

72
00:03:26,766 --> 00:03:29,200
The first question is,
do you think we will get this shape

73
00:03:29,200 --> 00:03:30,600
of some stairs here?

74
00:03:30,600 --> 00:03:35,000
And the second question is, do you think
we'll get a much more accurate prediction,

75
00:03:35,000 --> 00:03:39,333
like a prediction
that is getting very close to the 160 K

76
00:03:39,333 --> 00:03:41,633
that is supposed to be the real salary.

77
00:03:41,633 --> 00:03:44,633
So these are the two questions
I will let you think about that.

78
00:03:44,733 --> 00:03:47,733
And I look forward to giving you
the solution in the next section.

79
00:03:48,200 --> 00:03:49,866
Until then, enjoy machine learning.