1
00:00:02,450 --> 00:00:10,880
In this session, we will understand the maths behind Cannon can use nearest neighbor.

2
00:00:12,530 --> 00:00:19,870
Came here, a neighbor concept is used both in a classification supervised learning algorithm and also

3
00:00:19,880 --> 00:00:21,560
an unsupervised learning.

4
00:00:21,980 --> 00:00:29,390
OK, so let's explain this using an example I am trying to find in this case whether this green ball

5
00:00:29,390 --> 00:00:34,830
belongs to the group of blue balls or the group of Red Bulls.

6
00:00:35,510 --> 00:00:36,870
So how will I find that out?

7
00:00:37,340 --> 00:00:43,450
I will find the distance between this green ball and the nearest red and blue balls.

8
00:00:44,390 --> 00:00:47,070
OK, so that is all I do.

9
00:00:47,810 --> 00:00:48,950
And what is this key?

10
00:00:51,390 --> 00:00:58,320
He is the number of data points that I'm considering for the purpose of determining whether this blue

11
00:00:58,320 --> 00:01:03,370
ball belongs to a green ball, belongs to the Blue Group or to the red group.

12
00:01:03,930 --> 00:01:08,580
So what I do is I'm taking four points, Suki's four.

13
00:01:09,000 --> 00:01:13,950
So I compute the distance between this point and this Red Point.

14
00:01:14,100 --> 00:01:20,850
Similarly, I compute the distance between this point on this point and with each of the two blue points.

15
00:01:20,910 --> 00:01:25,080
So I compute the distance for four scenarios.

16
00:01:25,080 --> 00:01:31,020
Right, because I'm considering for nearest neighbors, how will I compute the distance?

17
00:01:31,230 --> 00:01:33,440
I use the equilibrium distance formula.

18
00:01:33,840 --> 00:01:34,150
Right.

19
00:01:34,440 --> 00:01:38,070
Suppose I have to find the distance between this and this point.

20
00:01:38,340 --> 00:01:43,680
OK, this point is x1 come on way run on this point, this extra come out.

21
00:01:44,280 --> 00:01:51,190
So the formula is x2 minus x1 the all squared plus Y to minus Y one the world square.

22
00:01:51,780 --> 00:01:54,780
So I compute for such distances.

23
00:01:56,200 --> 00:01:56,530
Clear.

24
00:01:58,510 --> 00:01:58,950
OK.

25
00:02:00,900 --> 00:02:08,610
The formula that I use this Euclidian, I can also use what is known as Manhattan distance, that is,

26
00:02:09,060 --> 00:02:15,960
instead of finding the distance through a straight line like this, I compute the distance from here

27
00:02:16,140 --> 00:02:16,840
to here.

28
00:02:16,860 --> 00:02:19,020
That is the base and then the height.

29
00:02:19,930 --> 00:02:28,000
OK, that's another way of computing the distance you can use any one of the two methods for computing

30
00:02:28,000 --> 00:02:28,720
the distance.

31
00:02:29,530 --> 00:02:34,420
OK, now let's explain this further using an example.

32
00:02:34,900 --> 00:02:40,000
I have seven data points for height, age and the corresponding weight.

33
00:02:40,980 --> 00:02:48,330
I'm being asked to predict the rate for the eighth data point that only Heighten agent provided.

34
00:02:49,210 --> 00:02:51,230
OK, so how will I compute?

35
00:02:52,270 --> 00:03:02,110
I compute the distance between this point five point eight height and age 37 with respect to each of

36
00:03:02,110 --> 00:03:03,370
these five data points.

37
00:03:03,850 --> 00:03:07,120
For that, I use the equilibrium formula right.

38
00:03:07,630 --> 00:03:12,410
In the case of six and 40, how will they compute Euclidean distance?

39
00:03:12,610 --> 00:03:15,910
It will be six minus five point eight.

40
00:03:15,910 --> 00:03:20,900
The whole square plus 40, minus thirty seven, the whole square.

41
00:03:21,610 --> 00:03:24,310
I add the two and then I take a square root.

42
00:03:24,910 --> 00:03:25,260
Right.

43
00:03:25,480 --> 00:03:27,940
I get a value of three point zero one.

44
00:03:28,450 --> 00:03:31,930
Similarly, I compute five point eight and thirty seven.

45
00:03:31,930 --> 00:03:34,670
I compare against six point one one twenty six.

46
00:03:35,110 --> 00:03:36,250
So on and so forth.

47
00:03:36,490 --> 00:03:39,970
I compute this equilibrium distance, ok.

48
00:03:40,900 --> 00:03:45,530
However, I am considering nearest neighbor of three only.

49
00:03:46,450 --> 00:03:49,120
OK, so I take.

50
00:03:50,240 --> 00:03:57,470
These three points in the nearest neighbors right in the distance is lawyer, that means it is closer

51
00:03:57,470 --> 00:03:57,890
to the.

52
00:03:58,960 --> 00:04:01,280
Five point eight and thirty seven data point, right?

53
00:04:01,480 --> 00:04:09,910
So I consider the three lawyers the distances and the corresponding rate, I think I add the three week

54
00:04:09,940 --> 00:04:14,400
that is at 78 and 60, I am divided by three.

55
00:04:14,740 --> 00:04:19,440
I'm taking an average and the wait I get is seventy two point six seven.

56
00:04:20,050 --> 00:04:27,220
So the weight corresponding to the height of five point eight and thirty seven is seventy two point

57
00:04:27,340 --> 00:04:28,060
six seven.

58
00:04:28,630 --> 00:04:29,830
That is my prediction.

59
00:04:30,870 --> 00:04:42,300
This is how I compute the predicted value using Cannon, just like in the case of regression and decision

60
00:04:42,300 --> 00:04:49,650
tree, the entire process of computing the nearest neighbors is done using a prebuilt library that is

61
00:04:49,650 --> 00:04:51,990
available both in Python as well as in R.

62
00:04:53,220 --> 00:04:55,020
OK, is this clear?
