﻿1
00:00:00,200 --> 00:00:01,866
Hello my friends, and welcome

2
00:00:01,866 --> 00:00:06,333
to this new practical activity
on this time the kernel SVM model.

3
00:00:06,600 --> 00:00:09,866
So please follow me into part
three classification

4
00:00:09,900 --> 00:00:13,100
to implement the kernel SVM model.

5
00:00:13,433 --> 00:00:15,533
And as usual
we're going to start with Python.

6
00:00:15,533 --> 00:00:18,900
And here we are with our two files.

7
00:00:19,066 --> 00:00:20,400
The first one that contains

8
00:00:20,400 --> 00:00:24,600
the whole implementation of the kernel
SVM model in the Ipynb format.

9
00:00:24,900 --> 00:00:29,066
And the second one the data set,
that same data set on which we trained

10
00:00:29,066 --> 00:00:32,066
and will train all the classification
models of this part.

11
00:00:32,233 --> 00:00:34,333
So let's have a quick look at it again.

12
00:00:34,333 --> 00:00:38,033
This data set contains
some information of 400 customers

13
00:00:38,266 --> 00:00:40,300
represented by each row here.

14
00:00:40,300 --> 00:00:42,366
So each row corresponds to a customer.

15
00:00:42,366 --> 00:00:46,633
And for each of these customers, well
we have the age which is a first feature.

16
00:00:46,800 --> 00:00:49,300
The estimated salary
which is a second feature.

17
00:00:49,300 --> 00:00:53,000
And the dependent variable purchased
which tells whether or not

18
00:00:53,166 --> 00:00:57,033
they bought no or yes, a previous SUV.

19
00:00:57,400 --> 00:00:57,900
All right.

20
00:00:57,900 --> 00:00:59,600
And so in our practical activities

21
00:00:59,600 --> 00:01:03,533
of this part three classification,
we build several classification models

22
00:01:03,533 --> 00:01:06,766
to understand the correlations
between these two features here.

23
00:01:06,966 --> 00:01:11,866
And this dependent variable purchased
in order to predict which future customers

24
00:01:11,866 --> 00:01:16,566
will buy brand new luxury SUV
that is advertised on social networks.

25
00:01:16,733 --> 00:01:19,600
All right.
So still the same story. And now.

26
00:01:19,600 --> 00:01:23,333
Well, let's begin
our kernel SVM implementation.

27
00:01:23,533 --> 00:01:25,800
I'm going to open it
with Google Colaboratory.

28
00:01:25,800 --> 00:01:28,800
But you have the choice to open it
with Jupyter Notebook as well.

29
00:01:28,833 --> 00:01:30,033
Choose your favorite.

30
00:01:30,033 --> 00:01:33,966
And now it is opening
the notebook loading it, laying it out.

31
00:01:33,966 --> 00:01:35,533
And there we go.

32
00:01:35,533 --> 00:01:38,800
Welcome to the kernel SVM implementation.

33
00:01:39,500 --> 00:01:39,900
All right.

34
00:01:39,900 --> 00:01:43,500
And of course it results from the exact
same classification template

35
00:01:43,500 --> 00:01:46,666
which we made in the first section
of this part logistic regression.

36
00:01:46,966 --> 00:01:49,733
Indeed, when we built the logistic
regression model,

37
00:01:49,733 --> 00:01:53,300
we made at the same time
a great classification template

38
00:01:53,500 --> 00:01:58,733
which allows us to only change one cell
when building another classification

39
00:01:58,900 --> 00:02:02,733
model, which is exactly,
you know, the cell where we build

40
00:02:02,733 --> 00:02:05,733
and train this classification model
on the training set.

41
00:02:05,733 --> 00:02:08,933
So this cell is the only one
we have to modify here.

42
00:02:09,066 --> 00:02:12,466
However,
this notebook is in read only mode.

43
00:02:12,566 --> 00:02:15,600
So in order to modify this cell
to build a kernel SVM model,

44
00:02:15,800 --> 00:02:21,033
we have to indeed click file here in order
to create a copy of this notebook.

45
00:02:21,233 --> 00:02:24,233
And to do this you just need to click
save a Copy and drive.

46
00:02:24,466 --> 00:02:27,200
This will create a copy of this notebook.

47
00:02:27,200 --> 00:02:31,800
And in a second
you'll have it opened on your machine.

48
00:02:32,100 --> 00:02:34,666
All right so I'm going to put everything
in that order.

49
00:02:34,666 --> 00:02:35,400
All right.

50
00:02:35,400 --> 00:02:36,733
And now this is our copy.

51
00:02:36,733 --> 00:02:40,866
And so inside the only thing we have to
re-implement is indeed that cell

52
00:02:40,866 --> 00:02:44,600
where we build and train
the kernel SVM model on the training set.

53
00:02:44,933 --> 00:02:47,100
So let's not look at it.

54
00:02:47,100 --> 00:02:48,766
Let's put it in the trash.

55
00:02:48,766 --> 00:02:52,066
And now let's re-implement
it from scratch.

56
00:02:52,733 --> 00:02:53,666
All right. Perfect.

57
00:02:53,666 --> 00:02:56,766
So it's actually super, super easy.

58
00:02:57,133 --> 00:03:01,400
We can even do it even more efficiently
than, you know, the usual way.

59
00:03:01,400 --> 00:03:06,033
When we went to the cycling API,
then got the name of the function

60
00:03:06,366 --> 00:03:10,200
because, as I explained in the previous
section, when building the SVM model,

61
00:03:10,200 --> 00:03:15,066
the classic one, well, the way we built
it was actually to, you know, build

62
00:03:15,066 --> 00:03:18,800
an object of this class as VC
and then choose a linear kernel.

63
00:03:19,133 --> 00:03:21,433
And now actually,
you know, we're about to do the same.

64
00:03:21,433 --> 00:03:25,133
We're about to use the same class
to build this kernel, SVM objects.

65
00:03:25,266 --> 00:03:27,933
But this time,
instead of choosing a linear kernel,

66
00:03:27,933 --> 00:03:29,733
we will choose a nonlinear one.

67
00:03:29,733 --> 00:03:31,333
And that's the best example of it.

68
00:03:31,333 --> 00:03:36,533
You know, the most commonly used it
is the RBF radial basis function kernel.

69
00:03:37,133 --> 00:03:41,133
And you also have some other ones like the
polynomial kernel or the sigmoid kernel.

70
00:03:41,266 --> 00:03:44,100
But really
this one is the best option to start with.

71
00:03:44,100 --> 00:03:48,233
And that's the one we'll choose
for our kernel SVM implementation,

72
00:03:48,266 --> 00:03:49,500
as you might have guessed.

73
00:03:49,500 --> 00:03:52,266
And therefore
remember how I also want to train you on

74
00:03:52,266 --> 00:03:55,866
how to juggle with the different tools
of your machine learning toolkit.

75
00:03:55,866 --> 00:03:59,200
You know, the data preprocessing toolkit
or the regression toolkit, or even now

76
00:03:59,200 --> 00:04:02,200
the classification toolkit
containing all your classification models.

77
00:04:02,500 --> 00:04:08,366
The most efficient way now to build
this kernel SVM model is just to go back

78
00:04:08,366 --> 00:04:13,933
to our SVM implementation, then
find that cell where we built the SVM.

79
00:04:14,100 --> 00:04:18,600
And indeed, since we built that SVM
with the same class as VC,

80
00:04:18,800 --> 00:04:22,733
but choosing a linear kernel,
well now as simple as that.

81
00:04:22,733 --> 00:04:26,900
In order to build a kernel SVM,
we just need to take this code cell

82
00:04:27,166 --> 00:04:30,566
and paste it
right here in a new code cell.

83
00:04:30,733 --> 00:04:31,833
And then guess what?

84
00:04:31,833 --> 00:04:34,833
Just change that kernel parameter.

85
00:04:35,033 --> 00:04:35,733
All right.

86
00:04:35,733 --> 00:04:38,933
And as we said this time
we're not using a linear kernel.

87
00:04:38,933 --> 00:04:41,400
We're going to use an RBF kernel.

88
00:04:41,400 --> 00:04:45,133
The radial basis function which was
explained in the intuition lectures.

89
00:04:45,433 --> 00:04:47,000
And there you go, my friend.

90
00:04:47,000 --> 00:04:49,433
That implementation is over.

91
00:04:49,433 --> 00:04:50,733
You see how I'm doing

92
00:04:50,733 --> 00:04:54,433
all these templates
to make you as most efficient as possible?

93
00:04:54,566 --> 00:04:55,600
Well, that's what I mean.

94
00:04:55,600 --> 00:04:58,700
Now we can juggle between
our different codes, different tools

95
00:04:58,700 --> 00:05:02,400
of machine learning,
and we can build new models in flashlight.

96
00:05:02,633 --> 00:05:05,633
And that's what we just did
with kernel SVM.

97
00:05:05,800 --> 00:05:08,100
All right. So basically we're done.
We're ready.

98
00:05:08,100 --> 00:05:12,433
So the next natural step is to upload
the data set inside the notebook.

99
00:05:12,733 --> 00:05:15,633
And to do this
you need to click this folder button here.

100
00:05:15,633 --> 00:05:20,200
Then your notebook will be connecting
to a runtime to enable file browsing.

101
00:05:20,266 --> 00:05:21,400
Still the same story.

102
00:05:21,400 --> 00:05:25,100
In a few seconds we should be able to see
the upload button.

103
00:05:25,133 --> 00:05:26,766
Perfect. There we go.

104
00:05:26,766 --> 00:05:28,500
And so now we're going to click it

105
00:05:28,500 --> 00:05:31,200
and we will directly get access
to social network.

106
00:05:31,200 --> 00:05:34,200
As if you know
you were here in the SVM section.

107
00:05:34,366 --> 00:05:36,533
But let me show you the path once again.

108
00:05:36,533 --> 00:05:39,900
So please find first your whole
machine learning is that folder.

109
00:05:40,033 --> 00:05:44,000
Then please go to part three,
then section 17 kernel SVM,

110
00:05:44,200 --> 00:05:48,333
then Python
and then social network ads dot CSV.

111
00:05:48,733 --> 00:05:49,300
Perfect.

112
00:05:49,300 --> 00:05:52,300
It will upload as a data
set inside the notebook.

113
00:05:52,533 --> 00:05:53,333
And now.

114
00:05:53,333 --> 00:05:58,100
Now we can run everything
by clicking runtime here and then run
