1
00:00:01,740 --> 00:00:08,370
Now, in this video, we are going to bring the model with a radial cannon as we go and indeed your

2
00:00:08,390 --> 00:00:12,540
elected radial camel as to hyper parameters.

3
00:00:13,440 --> 00:00:14,460
One is the gamma.

4
00:00:15,680 --> 00:00:22,860
Which 60 days later, importance of the distance between the retail company and other point.

5
00:00:24,500 --> 00:00:26,480
And the second hyper barometer is the cost.

6
00:00:27,730 --> 00:00:30,320
This we have been discussing since Línea comment.

7
00:00:32,480 --> 00:00:36,770
So fitting that a deal cardinal is the same.

8
00:00:37,370 --> 00:00:43,970
It uses the SBM function only that the cardinal value will be radial here instead of linear or binomial.

9
00:00:45,830 --> 00:00:47,910
We will store this in SVM fit.

10
00:00:48,230 --> 00:00:49,970
Ah, it will be for radio.

11
00:00:51,480 --> 00:00:52,480
And if I run this.

12
00:00:52,490 --> 00:01:01,310
Come on, you can see that I have a similar SBM print on variable which has the information of radial

13
00:01:01,310 --> 00:01:02,960
gunnin as a model.

14
00:01:04,340 --> 00:01:05,480
Now you can use this.

15
00:01:06,890 --> 00:01:12,000
Variable to predict the values on the best set and take out its performance.

16
00:01:13,020 --> 00:01:19,260
But it does not really make sense because we need to find out that value of these hyper barometers at

17
00:01:19,260 --> 00:01:22,080
which we expect that the test performance will be better.

18
00:01:22,950 --> 00:01:27,360
So we want to find out the value of glamour and cost.

19
00:01:28,260 --> 00:01:32,490
Or that we will use the function and we will give it these values of cost.

20
00:01:33,270 --> 00:01:35,970
You can see that I have provided logarithmic range.

21
00:01:36,340 --> 00:01:42,750
It start from zero point zero zero one, then zero point zero one and goes up to 1000.

22
00:01:44,110 --> 00:01:53,800
And these values of Gama let from Datapoint to one to 50, I'll reduce some of these, I'll remove this

23
00:01:54,700 --> 00:02:00,470
to this forward so that it does not take too much time while running.

24
00:02:01,090 --> 00:02:04,150
And I've also added this glossy article to Ford.

25
00:02:05,320 --> 00:02:09,480
Which means that there should be full cross validation sets instead of thing.

26
00:02:10,660 --> 00:02:12,010
So everything is the same.

27
00:02:12,150 --> 00:02:13,540
Just got radial.

28
00:02:14,350 --> 00:02:21,250
And in the hybrid barometer's, instead of the degree we have Gamma Gloucester's the same as earlier

29
00:02:22,150 --> 00:02:24,130
and we have a cross-validation take out for.

30
00:02:25,710 --> 00:02:26,680
So I Londis, come on.

31
00:02:31,410 --> 00:02:37,440
And we have this tuned out out of what evil created, this video will condense the information of all

32
00:02:37,440 --> 00:02:43,590
the combination of models that could be created using these values of cost and gamma.

33
00:02:48,800 --> 00:02:54,020
If you want to look at the information in this, you're not out.

34
00:02:54,170 --> 00:02:54,470
Out.

35
00:02:55,190 --> 00:02:55,690
You can go.

36
00:02:55,790 --> 00:03:02,300
Somebody come on on this and you can see that it contains the information of all the different combinations.

37
00:03:02,510 --> 00:03:08,230
So for cost of this, Garmo, of this, we get a cross ventilation at add value.

38
00:03:08,300 --> 00:03:08,720
This.

39
00:03:09,870 --> 00:03:10,410
And so on.

40
00:03:10,470 --> 00:03:19,860
So all the different models that ran across cross-validation error values did this best model variable

41
00:03:19,980 --> 00:03:28,300
of do not old are contains that particular combination of cost and darma, which has the minimum crossway

42
00:03:28,300 --> 00:03:28,980
religion in it.

43
00:03:30,090 --> 00:03:34,830
So I run this command to store the information of that best model.

44
00:03:34,940 --> 00:03:36,250
Invest more are.

45
00:03:38,660 --> 00:03:47,060
Now, to look at this best mark up and underscore man, and you can see that we ran a radial SVM.

46
00:03:48,760 --> 00:03:56,110
The cost at which we got minimum cross-validation error is Cees equal to 10.

47
00:03:57,460 --> 00:04:04,450
The number of support vectors are coming out to be and a 19 out of three eight one sixty two of them

48
00:04:04,450 --> 00:04:06,940
are on one side, 157 out on the other.

49
00:04:09,350 --> 00:04:13,050
And we have a total accuracy of nearly sixty one point eight.

50
00:04:13,530 --> 00:04:15,450
It's a four fold cross-validation.

51
00:04:17,190 --> 00:04:18,630
Now using this model.

52
00:04:20,300 --> 00:04:27,620
We are going to predict the values on the deficit and those values would be stored in wiped out.

53
00:04:29,740 --> 00:04:36,310
And now we are going to compare how our predicted values fared against the actual values in the essay.

54
00:04:36,940 --> 00:04:37,870
So we'll run this.

55
00:04:38,210 --> 00:04:38,750
It will.

56
00:04:41,390 --> 00:04:48,350
And you can see that we are correctly predicting these 29 values for which we predict that Oscar will

57
00:04:48,350 --> 00:04:50,240
not be given to the movie.

58
00:04:50,570 --> 00:04:53,210
And actually the movie did not get the Oscar.

59
00:04:53,980 --> 00:04:59,450
And for at least 37 movies for which we predicted that it will get an Oscar.

60
00:04:59,690 --> 00:05:00,890
And actually, it the.

61
00:05:02,310 --> 00:05:08,930
So we get an accuracy of 66 out of when they do it in percentage terms.

62
00:05:08,970 --> 00:05:12,570
That is 66 divided by one, they do it.

63
00:05:14,170 --> 00:05:16,320
It is nearly sixty one point one percent.

64
00:05:20,350 --> 00:05:26,890
So you can see that when we then are polynomial Cottonelle also, it was suggesting that there is more

65
00:05:26,890 --> 00:05:31,720
of a linear relationship between the predictive variables and the response would even.

66
00:05:33,630 --> 00:05:41,250
And when we tried to feed a canal, cardinal swim on variables, we tell more of a linear relationship.

67
00:05:41,940 --> 00:05:45,680
It does not perform better than the polynomial order linear cardinal.

68
00:05:47,280 --> 00:05:51,130
So using the right gomel is very important.

69
00:05:51,780 --> 00:05:55,080
If that is the new relationship, a linear cardinal will do better.

70
00:05:55,620 --> 00:05:59,990
If there is a non-linear relationship, really a lot more normal cardinals would do better.

71
00:06:02,550 --> 00:06:04,450
And this was all about classification.

72
00:06:04,960 --> 00:06:12,070
That is that the response variable of the variable that we were trying to predict had glasses and we

73
00:06:12,070 --> 00:06:17,890
were assigning glasses using the created model on a test set or a music.

74
00:06:19,710 --> 00:06:24,090
In the next video, we will learn how to do regression using SBM.

75
00:06:24,430 --> 00:06:29,950
That is how to predict continuous values instead of classes using the SBM technique.
