1
00:00:00,510 --> 00:00:06,000
In this lecture, we will learn how to train our SVM model with radial kernel.

2
00:00:07,800 --> 00:00:12,010
Again, these steps are the same as linear and polynomial kernel.

3
00:00:12,690 --> 00:00:20,580
We just serve to mention Cardinal Liquidy to order B.F. and then we need to provide additional parameters

4
00:00:20,700 --> 00:00:25,920
of Gamma, since you know that Gamma is the impact of nearby points.

5
00:00:27,190 --> 00:00:31,630
Therefore, we need to provide value of Guama along with see.

6
00:00:34,480 --> 00:00:41,580
And again, if you want some more details, you can always look at the scale as we see documentation.

7
00:00:45,780 --> 00:00:48,250
So, again, first we are creating odor objects.

8
00:00:49,470 --> 00:00:53,000
Then we are fighting over Ekstrand invite trend data.

9
00:00:53,370 --> 00:00:57,570
Then we are predicting our Avivah Lewis for tests and rain.

10
00:00:57,930 --> 00:01:04,080
And at last, we are finding that curacy scored two compared with what, watt lendee man model?

11
00:01:05,880 --> 00:01:07,050
Let's run this.

12
00:01:07,480 --> 00:01:08,830
Here we are using GOMI.

13
00:01:08,930 --> 00:01:11,770
Well, 2.5 and seek well 10.

14
00:01:12,990 --> 00:01:15,020
This is just our initial values.

15
00:01:16,130 --> 00:01:22,330
Usually we choose the gumline, see where loose after applying research for our model.

16
00:01:25,260 --> 00:01:27,180
Oh, let's run this.

17
00:01:28,080 --> 00:01:32,220
We have created this object and we have free trade extending by train data.

18
00:01:33,990 --> 00:01:39,040
Now we can use this object to predict y values of tests and train certain.

19
00:01:39,150 --> 00:01:40,400
So let's do that.

20
00:01:44,510 --> 00:01:49,760
To find the accuracy score, we have to first mention the actual values and then the predicted values.

21
00:01:51,320 --> 00:01:55,310
So on our test data, the predatory score is point sixty one.

22
00:01:56,720 --> 00:02:02,570
Now, let's do a grid search to find the optimize value of C and Gamma.

23
00:02:04,140 --> 00:02:07,300
Here, I want to test these many values of see.

24
00:02:07,800 --> 00:02:14,990
I am testing zero point zero one zero point zero five point one point five one five, 10 and 50.

25
00:02:15,930 --> 00:02:20,340
You can expand this range also by adding some more values.

26
00:02:21,900 --> 00:02:22,890
Next, Gamma.

27
00:02:23,040 --> 00:02:30,390
I want to test this five values of gamma, which is zero point zero zero one zero point zero one zero

28
00:02:30,390 --> 00:02:32,880
point one zero point five and one.

29
00:02:35,280 --> 00:02:37,880
Since I want to test this, many will lose.

30
00:02:39,040 --> 00:02:42,800
I have kept this values in the form of dictionary.

31
00:02:43,780 --> 00:02:52,000
The key of this dictionary is the high but barometer's of my as we are mortal, which is God mine.

32
00:02:52,060 --> 00:02:55,450
See, that's why I didn't see Ingomar as a keys.

33
00:02:56,230 --> 00:03:02,890
And the values I want to test are done in the form of double in this dictionary.

34
00:03:04,630 --> 00:03:12,670
We have parameter names as ski's and we have all the parameter values as a value of this dictionary.

35
00:03:13,570 --> 00:03:18,100
Then we need to create the model on which we want to apply the search.

36
00:03:18,880 --> 00:03:19,720
So since.

37
00:03:20,840 --> 00:03:23,900
For us, the cut in the lives, Lizotte, if we if we will just use.

38
00:03:25,470 --> 00:03:28,980
SVM daughter, as we see, and then got liquid do ought to be.

39
00:03:29,420 --> 00:03:31,290
Ah, we have a sense what radial cardinal.

40
00:03:33,210 --> 00:03:35,760
Then we need to create a grid search object.

41
00:03:36,430 --> 00:03:39,050
The function we are using is grid search S.V..

42
00:03:39,960 --> 00:03:44,060
The first parameter in grid search CV is you model.

43
00:03:45,630 --> 00:03:49,000
Since our model is are not as as we are, we're not scrod.

44
00:03:50,040 --> 00:03:53,460
We have provided it as a false argument.

45
00:03:55,290 --> 00:03:57,960
The next parameter here is the parameters.

46
00:03:57,990 --> 00:03:59,930
We want to change in what model?

47
00:04:01,720 --> 00:04:03,670
And this should be in the form of a dictionary.

48
00:04:04,870 --> 00:04:07,560
We have already created it and we have named.

49
00:04:08,680 --> 00:04:11,100
Therefore, we have provided this bad MS here.

50
00:04:12,180 --> 00:04:17,200
The Nexus and you'll equal to minus one since we want all the processing power of all system.

51
00:04:18,240 --> 00:04:19,570
Then see the equal do three.

52
00:04:19,630 --> 00:04:21,490
There's a sense for cross-validation.

53
00:04:24,630 --> 00:04:26,080
And verbal equal to one.

54
00:04:26,160 --> 00:04:30,480
And we want it scoring on the basis of the accuracy scored.

55
00:04:31,660 --> 00:04:33,550
Let's create this.

56
00:04:36,810 --> 00:04:37,380
Object.

57
00:04:38,400 --> 00:04:41,130
Then we will fit to what extent and why train data?

58
00:04:42,500 --> 00:04:43,700
On this grid search.

59
00:04:49,500 --> 00:04:55,830
So we have trained over more than known to look at the best parameter we can use best.

60
00:04:55,850 --> 00:04:59,950
And it's got bedrooms, underscore a tribute of this grid, search it.

61
00:05:00,060 --> 00:05:05,750
So just write the name off your object and dot best underscore bedrooms.

62
00:05:07,050 --> 00:05:09,070
This should give you the best barometer.

63
00:05:10,080 --> 00:05:13,590
So here the best barometer for our model is see.

64
00:05:13,890 --> 00:05:17,580
250 and Goll might well do the zero point zero zero one.

65
00:05:19,380 --> 00:05:23,730
Now we are saving this model with the best parameter.

66
00:05:24,930 --> 00:05:30,000
And another object that is radial SVM underscored classifier.

67
00:05:31,140 --> 00:05:32,430
We are using this.

68
00:05:33,300 --> 00:05:40,900
A swim grade underscored radial, and then we are using dot best underscore, assuming to underscore,

69
00:05:40,960 --> 00:05:43,310
attribute, just run this.

70
00:05:44,240 --> 00:05:49,410
Now we have the model that correspond to this well use of hyper parameter.

71
00:05:51,050 --> 00:05:57,130
We can use the predicate function and accuracy function to find out the accuracy skort.

72
00:05:57,930 --> 00:05:58,930
Yes, run this.

73
00:05:59,000 --> 00:06:02,610
The accuracy for over this model is point six one.

74
00:06:03,290 --> 00:06:09,200
As I've told you earlier, you might not get the best accuracy score because our data is very small.

75
00:06:09,280 --> 00:06:16,400
We are working with large amount of data, say a million datapoint or a billion observations.

76
00:06:17,060 --> 00:06:23,690
In that case, we'll get a lot better score with its search as compared to some random values of parameter.

77
00:06:24,920 --> 00:06:29,030
So always strive to use grid search to optimize your parameters.

78
00:06:29,690 --> 00:06:31,010
That's all for this lecture.

79
00:06:31,100 --> 00:06:31,550
Thank you.
