1
00:00:00,420 --> 00:00:08,910
So in the last video we created over Libya, as we are more than for sequel to zero point zero one.

2
00:00:10,300 --> 00:00:17,230
But how to ensure that this is the best sea which we can choose for our linear model?

3
00:00:18,100 --> 00:00:26,110
We can provide any value of sea and for each value of sea, we will get a separate model and the accuracy

4
00:00:26,320 --> 00:00:29,740
escort will be different for all such models.

5
00:00:30,700 --> 00:00:41,790
So one way is to manually change this sea so I can run the same model with sequel to one with sea equal

6
00:00:41,790 --> 00:00:51,820
to zero point one and with sequel two hundred and manually compare the accuracy results for all such

7
00:00:51,940 --> 00:00:52,570
models.

8
00:00:53,200 --> 00:00:54,610
This is a lengthy process.

9
00:00:56,440 --> 00:01:04,100
To overcome this lengthy process, we can use grid search option of a scale on here.

10
00:01:05,020 --> 00:01:05,950
We can provide.

11
00:01:06,970 --> 00:01:08,620
Different values of SI.

12
00:01:10,110 --> 00:01:15,470
And Biton will automatically run our model with all these values of see.

13
00:01:16,500 --> 00:01:22,500
And later on, we can use the best model out of this given models.

14
00:01:25,980 --> 00:01:26,970
Could do this.

15
00:01:27,090 --> 00:01:29,300
We have to use grid search c.v.

16
00:01:29,880 --> 00:01:34,420
So first we have to import grid search seewhy from Escalon modern selection.

17
00:01:35,790 --> 00:01:44,570
Now, next step is to create a dictionary in which you have to mention the name of your hyper parameter

18
00:01:45,780 --> 00:01:46,590
as a key.

19
00:01:48,400 --> 00:01:53,440
And the value it will be taking as a values of your dictionary.

20
00:01:54,890 --> 00:01:58,370
Here, since I want to change values of SI.

21
00:01:59,700 --> 00:02:05,280
I will provide see as a key and then these are the values.

22
00:02:07,510 --> 00:02:10,510
I am passing this values in the form of poll.

23
00:02:11,380 --> 00:02:16,750
You can see I have selected values from zero point zero zero one to tosing.

24
00:02:17,850 --> 00:02:22,260
So I will run my model for all these 13 values of see.

25
00:02:24,080 --> 00:02:31,040
If I want to change and the other hyper parameter, I can just broyd another key value pair.

26
00:02:32,690 --> 00:02:36,100
Now, we have created a weather parameter dictionary.

27
00:02:36,800 --> 00:02:42,560
The method of using grid search is almost similar to that of training a model.

28
00:02:43,580 --> 00:02:45,170
There is one additional step.

29
00:02:45,620 --> 00:02:49,850
First, we need to clear the model that we are going to use in grid search.

30
00:02:50,180 --> 00:02:57,530
So suppose if we are using support vector classification, we have to create an object for support vector

31
00:02:57,530 --> 00:03:00,470
classification, a field using linear regression.

32
00:03:00,500 --> 00:03:03,020
We have to create an object for linear regression.

33
00:03:03,350 --> 00:03:09,650
If we are using support vector regression, we have to create an object for support vector regression.

34
00:03:10,640 --> 00:03:13,070
So first, the surplus to create that object.

35
00:03:13,760 --> 00:03:21,320
Then just like any other model training, we have to provide parameters and grid search.

36
00:03:21,470 --> 00:03:23,990
The first parameter is the word model.

37
00:03:25,490 --> 00:03:30,170
Then the second parameter is the list of parameters that we have already created.

38
00:03:32,330 --> 00:03:40,880
Note that this should be in the form of a dictionary with the parameters of your model as scheiße and

39
00:03:40,880 --> 00:03:42,860
their values as values.

40
00:03:44,450 --> 00:03:46,790
Then there are a few other parameters.

41
00:03:47,060 --> 00:03:49,280
The next parameter is and jobs.

42
00:03:50,240 --> 00:03:52,340
This should be equal to minus one.

43
00:03:53,540 --> 00:03:57,890
This is used to maximize the performance of your laptop or machine.

44
00:03:58,850 --> 00:04:05,990
So minus one means your computer or your system will use all the processing power for training this

45
00:04:05,990 --> 00:04:06,350
model.

46
00:04:07,580 --> 00:04:09,460
Then the next parameter is CV.

47
00:04:09,460 --> 00:04:11,300
CV is for cross validation.

48
00:04:12,590 --> 00:04:19,730
So if you don't know what cross-validation is, Biton will create in our model.

49
00:04:20,150 --> 00:04:21,640
All this will lose of C.

50
00:04:22,550 --> 00:04:30,650
And then it will automatically use a part of of, uh, training data to get their test result and find

51
00:04:30,650 --> 00:04:34,880
out the accuracy scored on this many values of parameter.

52
00:04:36,650 --> 00:04:38,890
So CV is ten here.

53
00:04:39,140 --> 00:04:41,660
You can change this number if you want.

54
00:04:42,260 --> 00:04:44,870
Then verbals is equal to one.

55
00:04:46,250 --> 00:04:50,420
This represents the amount of message we are going to get.

56
00:04:50,540 --> 00:04:52,460
While we are training or model.

57
00:04:54,140 --> 00:04:58,820
So you saw that we received messages in our output.

58
00:04:59,750 --> 00:05:00,990
So a few give.

59
00:05:01,040 --> 00:05:01,580
What was it?

60
00:05:01,610 --> 00:05:05,220
Well, the one you will get some message while you're cleaning your data.

61
00:05:06,020 --> 00:05:07,930
And the next parameter is scoring.

62
00:05:08,060 --> 00:05:12,740
Here we want to use accuracy scores of this models.

63
00:05:13,400 --> 00:05:17,360
Therefore, we are using scoring well to accuracy.

64
00:05:19,430 --> 00:05:22,600
Then the next step is just like any other model.

65
00:05:22,740 --> 00:05:26,700
We will fit this grid search object using our test, Sandrine.

66
00:05:27,790 --> 00:05:35,470
And after that, since we are training this many models, we can pick the best model and then predict

67
00:05:35,470 --> 00:05:37,330
the values of my variable.

68
00:05:38,950 --> 00:05:41,380
So let's first import the grid search.

69
00:05:42,680 --> 00:05:44,500
Cleared this paramedic study.

70
00:05:46,460 --> 00:05:50,580
Cleared this VLF, a solemn, linnear object.

71
00:05:51,180 --> 00:05:56,550
You can see that we are just using as as we see with current liquidly, too linear.

72
00:05:58,690 --> 00:05:59,170
And then.

73
00:06:00,180 --> 00:06:02,430
We are creating a of grid search object.

74
00:06:05,400 --> 00:06:07,980
If you want more details on the parameters.

75
00:06:09,040 --> 00:06:12,850
You can open the documentation of Eskil and Grid Search.

76
00:06:12,880 --> 00:06:15,540
You will get all the information there as well.

77
00:06:17,830 --> 00:06:20,470
Then the next, the surplus to Fethullah data.

78
00:06:21,610 --> 00:06:27,150
Remember to use standard X data instead of just X variable.

79
00:06:31,110 --> 00:06:33,810
Now, you can see we are getting this kind of messages.

80
00:06:33,990 --> 00:06:36,750
This is because we have used verbal sequel to one.

81
00:06:39,320 --> 00:06:45,920
You can see we have provided turbin values of see, that's what we are getting, the message that we

82
00:06:45,920 --> 00:06:49,340
are fitting then for we model for 13 candidates.

83
00:06:49,850 --> 00:06:52,430
For you to send for cross-validation, equate to 10.

84
00:06:55,420 --> 00:06:56,830
This may take some time.

85
00:07:01,260 --> 00:07:07,800
You can see that we have fitted our models now grid search comes with different attributes.

86
00:07:08,340 --> 00:07:11,970
The first attribute here is best underscored patterns.

87
00:07:12,690 --> 00:07:15,820
This will give you the information of the best model.

88
00:07:16,030 --> 00:07:17,550
Although this 13 models.

89
00:07:19,180 --> 00:07:25,590
So if freed on this, we can get the parameter for which we are getting the best result.

90
00:07:26,470 --> 00:07:31,510
So out of all this sea values, we are getting our best preserved at sea.

91
00:07:31,610 --> 00:07:33,340
Well, two zero point five.

92
00:07:34,770 --> 00:07:38,930
So you can see that there is no need to separately run.

93
00:07:38,950 --> 00:07:40,930
You are more than for each value of sea.

94
00:07:41,620 --> 00:07:49,900
You can just provide a single Padam dictionary and grid search, do random order for all the values

95
00:07:49,900 --> 00:07:52,780
of sea that you have mentioned in badam parameter.

96
00:07:55,430 --> 00:07:58,760
The next secretary butyrate is best under score, estimate estimated.

97
00:08:00,700 --> 00:08:07,030
So with best underscore Perram, we were getting the parameter value corresponding to our best model,

98
00:08:07,990 --> 00:08:09,830
that best center scored Sumita.

99
00:08:10,900 --> 00:08:13,240
We will get our best model.

100
00:08:14,550 --> 00:08:22,440
And here we are simply saving this, our best model, as Len, as we underscore to see Alef later on.

101
00:08:22,470 --> 00:08:26,280
We will use this Lin as a swimmer, not to score CnF.

102
00:08:27,140 --> 00:08:28,850
To predict the values of this.

103
00:08:30,780 --> 00:08:31,830
Let's run this.

104
00:08:32,580 --> 00:08:35,040
So now we have a what best model?

105
00:08:35,730 --> 00:08:38,150
The model with sequel 2.5.

106
00:08:38,760 --> 00:08:39,750
And this variable.

107
00:08:41,520 --> 00:08:43,780
Now, let's find out the accuracy score.

108
00:08:43,980 --> 00:08:52,830
Here we can radically use why test and instead of providing a separate credit variable, we can directly

109
00:08:52,830 --> 00:08:54,600
use Daudt predict function here.

110
00:08:56,130 --> 00:08:57,720
So let's run this.

111
00:08:58,740 --> 00:09:02,310
The accuracy for our model is zero point five nine.

112
00:09:03,270 --> 00:09:10,790
So this is the best accuracy which we can get out of this 13 values since our data is very less.

113
00:09:11,520 --> 00:09:14,250
And we only have around 500 observations.

114
00:09:15,000 --> 00:09:19,620
That's why he a good service is not making much sense.

115
00:09:20,070 --> 00:09:25,230
And we may get accuracy score lower than our initial accuracy score.

116
00:09:26,460 --> 00:09:27,910
So don't worry about that.

117
00:09:27,930 --> 00:09:32,700
If you have large amount of data, grid search will definitely give you a better result as compared

118
00:09:32,700 --> 00:09:35,140
to some random value of parameter.

119
00:09:39,070 --> 00:09:47,050
That's so we use grid search in the next lecture, will green, a polynomial caramel SVM model.
