﻿1
00:00:00,033 --> 00:00:01,533
Okay, my friends, let's do this.

2
00:00:01,533 --> 00:00:05,233
Are you ready to implement the code
to visualize your results?

3
00:00:05,400 --> 00:00:07,266
Here we go. Let's do it together.

4
00:00:07,266 --> 00:00:08,833
So we're going to do it efficiently.

5
00:00:08,833 --> 00:00:09,633
We're going to start from

6
00:00:09,633 --> 00:00:13,566
our polynomial regression implementation
the visualization code at the end.

7
00:00:13,800 --> 00:00:15,300
So we're going to copy it.

8
00:00:15,300 --> 00:00:19,700
Then we're going to create a new code
cell here and paste it.

9
00:00:19,966 --> 00:00:20,633
Okay.

10
00:00:20,633 --> 00:00:24,300
And now the exercise is to figure out
what we have

11
00:00:24,300 --> 00:00:27,800
to change here
in order to adapt to our SVR model.

12
00:00:28,066 --> 00:00:30,633
So let's do this. First
let's make the obvious change.

13
00:00:30,633 --> 00:00:35,333
Let's replace here
polynomial regression by SVR.

14
00:00:35,566 --> 00:00:38,800
And then let's see row by row
what we have to change.

15
00:00:39,066 --> 00:00:39,400
Okay.

16
00:00:39,400 --> 00:00:42,633
So for the first row remember that x and y

17
00:00:42,633 --> 00:00:45,966
are the scaled values of the inputs
and the outputs.

18
00:00:46,100 --> 00:00:49,500
So if we want to have a nice plot
with the original values

19
00:00:49,500 --> 00:00:52,666
of the position levels and the salaries,
well the first thing we have to do

20
00:00:52,666 --> 00:00:57,800
is to reverse the scale of the inputs x
and y to put them in their original scale.

21
00:00:57,966 --> 00:00:58,600
Okay.

22
00:00:58,600 --> 00:01:01,500
And to do so well,
we're going to take our scaler objects

23
00:01:01,500 --> 00:01:05,833
X, NSC, y and applied inverse transform
method on each of them.

24
00:01:06,066 --> 00:01:08,466
Okay.
So let's do this. Let's start with x.

25
00:01:08,466 --> 00:01:13,966
We take our c x scaler object
from which we apply

26
00:01:13,966 --> 00:01:18,666
the inverse underscore transform
method onto x.

27
00:01:18,833 --> 00:01:20,166
Okay. Perfect.

28
00:01:20,166 --> 00:01:21,400
Now these are the same for y.

29
00:01:21,400 --> 00:01:24,500
Let's take our c y scaler

30
00:01:24,500 --> 00:01:27,500
object from which we call again to inverse

31
00:01:27,900 --> 00:01:31,166
transform form method onto y.

32
00:01:31,733 --> 00:01:32,433
Excellent.

33
00:01:32,433 --> 00:01:34,466
And we're done with the first row.

34
00:01:34,466 --> 00:01:35,966
Now second row.

35
00:01:35,966 --> 00:01:39,433
Let's take care separately of the input x
and the prediction.

36
00:01:39,600 --> 00:01:43,466
So for x once again we have to apply
the reverse transformation

37
00:01:43,633 --> 00:01:44,833
to get the original square.

38
00:01:44,833 --> 00:01:49,766
So there we go again
we call our c underscore x square object

39
00:01:50,066 --> 00:01:54,833
from which we apply again
the inverse transform method.

40
00:01:54,833 --> 00:01:57,833
There we go. Apply to x. Okay.

41
00:01:58,233 --> 00:02:00,600
And now this is where
things get interesting.

42
00:02:00,600 --> 00:02:04,200
We're going to take care of
the predictions now of the inputs and x.

43
00:02:04,466 --> 00:02:07,433
So of course we have to remove
this whole line rect to predict fully

44
00:02:07,433 --> 00:02:10,433
rect fit transform x
and replace it by something.

45
00:02:10,466 --> 00:02:13,400
And now the question is
what do we have to replace it with.

46
00:02:13,400 --> 00:02:18,166
Well it is going to be something similar
to the prediction we made here.

47
00:02:18,300 --> 00:02:21,300
Only here
we made it for a single observation.

48
00:02:21,300 --> 00:02:24,133
And so we need to adapt this to X. Right.

49
00:02:24,133 --> 00:02:27,133
Because now we're making the predictions
of the inputs in the x.

50
00:02:27,266 --> 00:02:27,833
Okay.

51
00:02:27,833 --> 00:02:31,166
So what we're going to do
is take this whole prediction here.

52
00:02:31,433 --> 00:02:32,233
We're going to replace

53
00:02:32,233 --> 00:02:36,900
that whole line rect to predict rect
fit transform x by the prediction.

54
00:02:36,900 --> 00:02:37,966
We just copied.

55
00:02:37,966 --> 00:02:39,900
And now let's see
how we have to adapt this.

56
00:02:39,900 --> 00:02:42,300
So let's start
from inside the predict method.

57
00:02:42,300 --> 00:02:44,766
Of course we want to replace 6.5 by x.

58
00:02:44,766 --> 00:02:48,600
But here
we had to apply x transform on 6.5.

59
00:02:48,600 --> 00:02:51,266
Because 6.5 is the original square.

60
00:02:51,266 --> 00:02:54,033
And the predict method
has to be applied on the scaled values.

61
00:02:54,033 --> 00:02:59,133
But now since we apply the break
method on x and since x is already scaled

62
00:02:59,300 --> 00:03:02,566
well, we must not reapply the transform
method on x,

63
00:03:02,566 --> 00:03:07,800
and therefore what we need to replace here
is this whole six transform 6.5

64
00:03:08,000 --> 00:03:11,566
by indeed x because x is already scaled.

65
00:03:11,633 --> 00:03:12,933
So there we go.

66
00:03:12,933 --> 00:03:15,000
And now I think that's it.
I think we're done.

67
00:03:15,000 --> 00:03:16,033
That was pretty efficient.

68
00:03:16,033 --> 00:03:18,500
But you know
we're never too safe from human error.

69
00:03:18,500 --> 00:03:22,633
So we're going to double
check this by executing right away.

70
00:03:22,833 --> 00:03:23,866
Could see up here.

71
00:03:23,866 --> 00:03:24,666
So there we go.

72
00:03:24,666 --> 00:03:25,933
Let's click run cell here.

73
00:03:25,933 --> 00:03:28,933
And let's see what happens.
And there we go.

74
00:03:28,966 --> 00:03:30,466
We did it all correctly.

75
00:03:30,466 --> 00:03:32,966
This is indeed the curve of the SVM model.

76
00:03:32,966 --> 00:03:38,000
And as we can see the predictions in blue
go pretty close to real outcomes in

77
00:03:38,000 --> 00:03:41,066
red except for that last one
which would make it an outlier.

78
00:03:41,066 --> 00:03:42,500
But that's how it is.
