﻿1
00:00:00,566 --> 00:00:02,900
Hello and welcome
back to the course of Machine Learning.

2
00:00:02,900 --> 00:00:05,900
Previously we talked about the Support
Vector Machine algorithm.

3
00:00:05,900 --> 00:00:09,566
And today we're going to talk about the
kernel support Vector machine algorithm.

4
00:00:09,566 --> 00:00:11,233
And the intuition behind it.

5
00:00:11,233 --> 00:00:14,566
So as you recall in the support
vector machine situation,

6
00:00:14,566 --> 00:00:20,700
we had a set of observations, 
which belong to different classes.

7
00:00:20,700 --> 00:00:23,833
And the point was to find this decision

8
00:00:23,833 --> 00:00:27,900
boundary between them
so that any future observations

9
00:00:28,200 --> 00:00:31,200
could be identified
which class they fall into.

10
00:00:31,733 --> 00:00:34,866
And in this case, we can see that,
there's a decision boundary.

11
00:00:34,866 --> 00:00:35,700
And the support vector

12
00:00:35,700 --> 00:00:39,533
machine algorithm tells us
exactly how to find that boundary.

13
00:00:39,866 --> 00:00:44,466
But what happens
when we cannot find a boundary?

14
00:00:44,633 --> 00:00:47,633
What happens in a situation
like this, for example.

15
00:00:47,800 --> 00:00:49,033
What what do we do here?

16
00:00:49,033 --> 00:00:51,933
We can't just draw a line
through these points, right?

17
00:00:51,933 --> 00:00:55,300
We can't just separate them
with a straight line.

18
00:00:55,633 --> 00:00:57,700
We can't separate them
with a horizontal line.

19
00:00:57,700 --> 00:00:59,766
We can't separate them
with a vertical line.

20
00:00:59,766 --> 00:01:04,200
Whatever we try to do,
we cannot separate these points

21
00:01:04,500 --> 00:01:09,666
in the same way that the support vector
machine algorithm, told us to.

22
00:01:10,066 --> 00:01:13,133
And so what happens in this situation
and why does it happen?

23
00:01:13,200 --> 00:01:16,200
Well, this happens because in this case,

24
00:01:16,400 --> 00:01:20,166
the data is not linearly separable.

25
00:01:20,566 --> 00:01:23,233
So here we've got the two examples
side by side on the left

26
00:01:23,233 --> 00:01:24,633
the linearly separable data.

27
00:01:24,633 --> 00:01:27,166
And on the right
the non-linearly separable data.

28
00:01:27,166 --> 00:01:32,033
And what's the support
vector machine algorithm does is it

29
00:01:32,033 --> 00:01:35,633
helps us find that decision boundary

30
00:01:35,633 --> 00:01:38,633
or correctly place that decision boundary.

31
00:01:38,933 --> 00:01:40,633
But it does have an assumption.

32
00:01:40,633 --> 00:01:43,533
The assumption is that
the data is linearly separable.

33
00:01:43,533 --> 00:01:48,533
So basically it's saying that it is
possible to place the decision boundary.

34
00:01:48,533 --> 00:01:52,900
So support vector machine algorithm
helps us choose the optimal

35
00:01:52,900 --> 00:01:56,533
decision boundary out of the multiple
decision boundaries that we can draw.

36
00:01:56,933 --> 00:02:00,733
Whereas in the non linearly separable case
we can't even draw one

37
00:02:00,733 --> 00:02:04,700
single decision
boundary or linear decision boundary.

38
00:02:04,700 --> 00:02:08,666
So therefore the support vector machine
algorithm just won't work by definition.

39
00:02:09,200 --> 00:02:11,300
And so what are we going to do.

40
00:02:11,300 --> 00:02:13,833
Well this is what this section is about.

41
00:02:13,833 --> 00:02:17,933
This section is about coming up
with an algorithm that will help us deal

42
00:02:17,933 --> 00:02:21,900
with situations like that
where we have to extract this class,

43
00:02:22,600 --> 00:02:25,200
when it is surrounded or,

44
00:02:25,200 --> 00:02:28,200
in other situations where you can't just
simply just draw a line.

45
00:02:28,733 --> 00:02:32,733
And the way this section is structured
or the intuition tutorials of this section

46
00:02:32,733 --> 00:02:36,133
is, first of all, we're
going to explore a method which is called,

47
00:02:36,600 --> 00:02:41,900
going to a higher dimensional space
where we will learn how we can take,

48
00:02:42,433 --> 00:02:46,866
this data set and add an extra dimension

49
00:02:47,200 --> 00:02:50,666
into our, space that we're dealing with

50
00:02:50,833 --> 00:02:53,966
and make our data, linearly separable.

51
00:02:53,966 --> 00:02:54,766
So we'll have a look

52
00:02:54,766 --> 00:02:59,266
at a couple of those examples,
and then we will discuss the kernel trick,

53
00:02:59,633 --> 00:03:03,633
which allows us to do these things
in a more computationally efficient way

54
00:03:03,833 --> 00:03:07,233
without having to deal with multiple,
or higher dimensions.

55
00:03:07,500 --> 00:03:10,566
And finally, we will talk about
the different types of kernels that exist.

56
00:03:10,700 --> 00:03:13,700
So quite an interesting section
ahead, and I can't wait

57
00:03:13,700 --> 00:03:16,466
to see you in the next tutorial.
Until then, happy analyzing.
