﻿1
00:00:00,366 --> 00:00:00,666
Okay.

2
00:00:00,666 --> 00:00:03,100
So let's go back to our classifier.

3
00:00:03,100 --> 00:00:05,000
So the pre-processing is done.

4
00:00:05,000 --> 00:00:07,933
And now we're going
to create our classifier.

5
00:00:07,933 --> 00:00:09,533
Everything should go well.

6
00:00:09,533 --> 00:00:12,733
I don't have to select this
because mine was already selected.

7
00:00:13,233 --> 00:00:16,500
So let's press Command and Control
plus enter to execute.

8
00:00:16,966 --> 00:00:20,166
And here it is classifier created.

9
00:00:20,166 --> 00:00:21,833
All good. Okay.

10
00:00:21,833 --> 00:00:26,466
And now we can make some predictions
of new observations which are here to test

11
00:00:26,466 --> 00:00:27,000
that observation.

12
00:00:27,000 --> 00:00:30,100
So let's do it. Why pred wipe. Right.

13
00:00:30,133 --> 00:00:30,633
All right.

14
00:00:30,633 --> 00:00:34,500
So let's have a look at why pred y pred.

15
00:00:36,033 --> 00:00:39,033
So these are all the predictions
of the test set observations.

16
00:00:39,033 --> 00:00:40,966
That is, for each user of the test

17
00:00:40,966 --> 00:00:44,300
set, our classifier predicts
if the user buys the SUV or not.

18
00:00:44,800 --> 00:00:45,733
So let's compare

19
00:00:45,733 --> 00:00:49,500
the predictions with the truth
which are contained in the test set.

20
00:00:50,000 --> 00:00:54,033
So this column tells the truth
about whether the users

21
00:00:54,033 --> 00:00:57,033
but yes or no, the SUV
and this are the predictions.

22
00:00:57,033 --> 00:01:00,800
That is our SVM classifier predicted
whether each user bought

23
00:01:00,800 --> 00:01:02,100
yes or no the SUV.

24
00:01:02,100 --> 00:01:04,633
So let's look at, for example, the well.

25
00:01:04,633 --> 00:01:04,933
Okay.

26
00:01:04,933 --> 00:01:08,400
So all these first guys here
where predicted not to buy the SUV.

27
00:01:08,400 --> 00:01:11,400
We have all zeros until the 103.

28
00:01:11,433 --> 00:01:14,933
But here as you can see
we have some guys who actually bought

29
00:01:14,933 --> 00:01:19,333
the SUV in reality is to 18,
19, 20 and 22.

30
00:01:19,400 --> 00:01:21,966
So our classifier predicted
that they didn't buy it.

31
00:01:21,966 --> 00:01:24,400
And the truth is
that they actually bought it.

32
00:01:24,400 --> 00:01:26,400
So that's incorrect predictions.

33
00:01:26,400 --> 00:01:30,166
But let's look at the incorrect
predictions in a more efficient way

34
00:01:30,766 --> 00:01:32,866
by looking at the confusion matrix.

35
00:01:32,866 --> 00:01:35,933
So let's select
this line here and execute.

36
00:01:36,133 --> 00:01:40,200
And now let's find out about
the real number of incorrect predictions.

37
00:01:40,600 --> 00:01:43,600
So let's type
CM here in the council and press enter.

38
00:01:43,933 --> 00:01:48,300
And wow that's actually a quite large
number of incorrect predictions.

39
00:01:48,633 --> 00:01:52,500
And by the way we don't obtain the same
results as in Python because we have some

40
00:01:52,533 --> 00:01:53,933
random factors in the model.

41
00:01:53,933 --> 00:01:55,666
And here we didn't specify C.

42
00:01:55,666 --> 00:01:57,600
So you might actually have
some different results.

43
00:01:57,600 --> 00:01:59,766
But the idea is here.

44
00:01:59,766 --> 00:02:03,133
So okay so let's look at now
the graph to see how it's doing.

45
00:02:03,600 --> 00:02:08,000
So for those of you who actually didn't
watch the Python tutorial about SVM,

46
00:02:08,533 --> 00:02:11,766
a good exercise is to try to guess
what's going to happen.

47
00:02:11,966 --> 00:02:14,566
That is what are going to be
the prediction regions.

48
00:02:14,566 --> 00:02:17,033
What is going to be the prediction
boundary?

49
00:02:17,033 --> 00:02:19,566
What do you guess you will see?

50
00:02:19,566 --> 00:02:21,100
So I will let you think for a second.

51
00:02:21,100 --> 00:02:22,500
You can pause on the video.

52
00:02:22,500 --> 00:02:24,133
And right now I'm going to tell you.

53
00:02:24,133 --> 00:02:27,766
So as you notice we chose a linear kernel,

54
00:02:27,766 --> 00:02:32,033
which means that our SVM classifier
is a linear classifier.

55
00:02:32,366 --> 00:02:37,133
So as I explained in the logistic
regression tutorials on Python and R,

56
00:02:37,600 --> 00:02:42,500
well, a linear classifier in 2D
dimensional space is a straight line.

57
00:02:42,533 --> 00:02:46,466
So here I'm telling you this right now
we are going to get a straight line.

58
00:02:46,500 --> 00:02:50,333
Not to get any disappointment because
I know we improved our model with the K.

59
00:02:50,400 --> 00:02:54,300
And before we obtained a good prediction
boundary that got,

60
00:02:54,600 --> 00:02:59,166
you know, those users on the bottom
right corner that actually bought the SUV,

61
00:02:59,266 --> 00:03:03,066
but who were incorrect predictions
for logistic regression?

62
00:03:03,066 --> 00:03:07,033
Well, here it's also going to be
some incorrect prediction for the linear

63
00:03:07,033 --> 00:03:10,166
SVM because it's actually
a linear classifier.

64
00:03:10,500 --> 00:03:12,133
So let's look at the results right now.

65
00:03:12,133 --> 00:03:15,133
Select this and press Command
and Control.

66
00:03:15,300 --> 00:03:16,800
Press enter to execute.

67
00:03:17,966 --> 00:03:20,033
And here are the results.

68
00:03:20,033 --> 00:03:22,566
Okay. So as you can see
that's exactly what I just told you.

69
00:03:22,566 --> 00:03:26,900
You know those users here
with a lower salary and the higher age.

70
00:03:27,066 --> 00:03:30,066
Well,
they actually bought the SUV in reality,

71
00:03:30,266 --> 00:03:33,266
because the points are green,
the points are the real observations.

72
00:03:33,500 --> 00:03:38,833
But they fell into the red region here
because since the linear classifier

73
00:03:38,833 --> 00:03:42,700
is a straight line, it couldn't,
you know, make some kind of a curve here

74
00:03:42,700 --> 00:03:46,400
to catch all the red points
into the right place.

75
00:03:46,700 --> 00:03:50,866
And therefore it cut some green points,
putting them in the red region.

76
00:03:51,400 --> 00:03:54,400
So yeah, that's exactly
the same as for logistic regression.

77
00:03:54,666 --> 00:03:57,133
Sorry about the disappointment,
but don't worry.

78
00:03:57,133 --> 00:04:00,966
In the next section
we will introduce a new kind of classifier

79
00:04:00,966 --> 00:04:05,300
which will be well the kernel SVM with
a different kernel than linear kernel.

80
00:04:05,566 --> 00:04:07,033
It's going to be a Gaussian kernel.

81
00:04:07,033 --> 00:04:09,000
Or even we can try some more kernels.

82
00:04:09,000 --> 00:04:11,166
Well you can practice that yourself.

83
00:04:11,166 --> 00:04:13,166
It can be very good practice for you.

84
00:04:13,166 --> 00:04:15,766
But here. Yes, it's a linear classifier.

85
00:04:15,766 --> 00:04:18,766
So basically it's
the same as logistic regression.

86
00:04:19,266 --> 00:04:22,800
And if we look at the test set
results is going to be the same.

87
00:04:22,833 --> 00:04:23,866
Let's have a look.

88
00:04:25,033 --> 00:04:26,933
And here is the test set okay.

89
00:04:26,933 --> 00:04:28,466
So yeah same thing here.

90
00:04:28,466 --> 00:04:32,566
We have some users of higher age
and lower estimated salary.

91
00:04:32,566 --> 00:04:37,033
Who bought the SUV
but fell into the red region.

92
00:04:37,033 --> 00:04:39,900
Again because our class for you
is a straight line.

93
00:04:39,900 --> 00:04:43,800
And that's the best
you could do to classify those two points.

94
00:04:43,800 --> 00:04:46,600
And place them into the corresponding
category.

95
00:04:46,600 --> 00:04:50,100
So a few incorrect predictions
here, a few incorrect predictions here.

96
00:04:50,500 --> 00:04:52,966
And if you want, you can count
the number of incorrect predictions.

97
00:04:52,966 --> 00:04:55,266
That is the number of green points
in the red region

98
00:04:55,266 --> 00:04:58,266
plus the number of red points
in the green region.

99
00:04:58,433 --> 00:05:02,900
And you will count the number of incorrect
predictions we found in the confusion

100
00:05:02,900 --> 00:05:05,900
matrix,
which was 20 incorrect predictions.

101
00:05:06,266 --> 00:05:10,200
But that's great that we have this result,
because that gives us the motivation

102
00:05:10,200 --> 00:05:13,433
to improve our classifier,
improve our model,

103
00:05:13,700 --> 00:05:15,766
and that's what
we're going to do in the next section.

104
00:05:15,766 --> 00:05:18,800
So I look forward to seeing you
in the next section

105
00:05:18,800 --> 00:05:22,800
and show you how we can substantially
improve our classification model.

106
00:05:23,033 --> 00:05:25,033
So I look forward
to showing you the next level.

107
00:05:25,033 --> 00:05:26,866
And until then, enjoy machine learning.
