1
00:00:01,320 --> 00:00:08,220
When we start our discussion on maximal marginal classifier, the first thing we need to understand

2
00:00:08,730 --> 00:00:10,770
is the concept of a hybrid plain.

3
00:00:12,730 --> 00:00:19,540
Simply put, a hyper plane is something that divides up be dimensional space in two parts.

4
00:00:21,800 --> 00:00:24,560
Don't be carried away by the word plane in its name.

5
00:00:25,310 --> 00:00:27,680
It is not always a two dimensional plane.

6
00:00:29,330 --> 00:00:31,850
It depends on how many dimensions we are considering.

7
00:00:33,800 --> 00:00:38,210
Let's consider a one dimensional space or it's a one dimensional space.

8
00:00:38,480 --> 00:00:39,040
It's a line.

9
00:00:40,850 --> 00:00:43,730
And how do we separate this line into two parts?

10
00:00:44,600 --> 00:00:45,890
We just need a point.

11
00:00:47,870 --> 00:00:53,320
Now, there's a left part of the line which is left to this point and there is a right part of the line.

12
00:00:54,980 --> 00:00:57,860
This point is a hyperbole for this piece.

13
00:01:00,650 --> 00:01:10,340
If we take a two dimensional space, we can separate it using a line you can see here in this diagram.

14
00:01:11,000 --> 00:01:13,770
We have a two dimensional space, which is a plane.

15
00:01:14,150 --> 00:01:20,900
And we can separated using a lane and there will be a path to the left and a path to the right of this

16
00:01:20,900 --> 00:01:21,260
lane.

17
00:01:22,760 --> 00:01:27,020
So for our two dimensional space, we have a one dimensional separator.

18
00:01:27,470 --> 00:01:28,430
That is a line.

19
00:01:30,170 --> 00:01:35,450
Similarly, if I continue increasing the dimensions of space, let's say a three dimensional space.

20
00:01:36,050 --> 00:01:38,960
It can be separated using our two dimensional plane.

21
00:01:40,340 --> 00:01:44,420
So this concept can be extended to any p dimensional space.

22
00:01:45,020 --> 00:01:52,430
And we will see that there is a hybrid plane of B minus one dimensions, which separates this B dimensional

23
00:01:52,430 --> 00:01:54,170
space into two parts.

24
00:01:56,060 --> 00:02:01,840
But you may be thinking, why are we talking about, B, dimensional spaces and trying to split these

25
00:02:01,840 --> 00:02:02,420
spaces?

26
00:02:03,890 --> 00:02:05,030
Let me explain it to you.

27
00:02:05,270 --> 00:02:10,190
Using an example, suppose I have three variables.

28
00:02:11,060 --> 00:02:17,010
Excellent x2 and the category or the class to which that observation belongs.

29
00:02:19,150 --> 00:02:28,030
These x1, x2 and classes can be anything related to your business, and Edu, x1 and X2 could be height

30
00:02:28,060 --> 00:02:31,600
and weight so students and classes could be weathered.

31
00:02:31,690 --> 00:02:33,610
The student passes or fails.

32
00:02:33,700 --> 00:02:39,760
A sports test X1 and X2 could be the results of two medical tests.

33
00:02:40,480 --> 00:02:46,120
Besides which, we may be trying to classify whether the patient has a heart disease or not.

34
00:02:46,930 --> 00:02:47,800
It could be anything.

35
00:02:48,880 --> 00:02:55,780
Basically we have to predictor variables and one variable is to be predicted, which is also known as

36
00:02:55,780 --> 00:02:57,100
the response variable.

37
00:02:58,030 --> 00:03:04,180
So X1 and X2 are the predictor variables and category is the response variable.

38
00:03:06,370 --> 00:03:13,060
Since we have two predictor variables, we can plot the observations on a two dimensional space such

39
00:03:13,060 --> 00:03:13,630
as this.

40
00:03:15,460 --> 00:03:25,780
We can have X one on the x axis, x2 on the Y axis, and Galut of each point will tell us to which class

41
00:03:25,930 --> 00:03:27,850
that particular point belongs.

42
00:03:29,820 --> 00:03:37,650
So all the blue points belong to one class and all these purple points belong to another class.

43
00:03:39,660 --> 00:03:46,730
I think you can connect the daughter between our hyper plane discussion and this example.

44
00:03:48,240 --> 00:03:56,310
This is basically a p dimensional space here being equal to two, because we have two predictive variables.

45
00:03:56,970 --> 00:03:59,160
And this is a two dimensional space.

46
00:04:01,210 --> 00:04:06,460
We want a hyper plane, which will be a line for this two dimensional space.

47
00:04:07,360 --> 00:04:16,270
So we want such a hyper plane which can divide this space into two parts such that all the blue points

48
00:04:16,420 --> 00:04:20,920
are in one part and all the purple points are in another part.

49
00:04:22,360 --> 00:04:28,570
If I am able to find this iBOT plane, I will have my classified model with me.

50
00:04:30,640 --> 00:04:38,800
Any point which falls on this side of this hyper plane will be classified as blue or class one.

51
00:04:39,520 --> 00:04:48,000
And anything that falls on this side of the hyper plane will be classified as purple or glass to simple,

52
00:04:48,280 --> 00:04:48,590
right?

53
00:04:50,050 --> 00:04:51,030
It is simple to know.

54
00:04:51,760 --> 00:04:53,290
Let's dig deeper in the next.

55
00:04:53,290 --> 00:04:53,710
We do.
