1
00:00:01,140 --> 00:00:01,620
Hello.

2
00:00:02,190 --> 00:00:07,920
In this lecture, we will learn how to run rate and the laso in Python.

3
00:00:09,810 --> 00:00:16,530
If you remember from the curity lecture, we first need to standardize our data before running the region.

4
00:00:16,590 --> 00:00:17,280
The LASO.

5
00:00:19,010 --> 00:00:24,890
So let's first start by centralizing over data for standardizing.

6
00:00:25,520 --> 00:00:28,420
We will first import be processing from Skillern.

7
00:00:28,960 --> 00:00:32,540
So we will write from Escalon Import Preprocessing.

8
00:00:41,210 --> 00:00:48,830
Then we cleared the scalar object, which will store the Skilling information for what x variable rate

9
00:00:48,920 --> 00:00:52,960
is scalar equate to reprocessing dot ascendants scalar.

10
00:00:53,660 --> 00:00:57,410
Here is standard scalar is the function of preprocessing.

11
00:01:02,830 --> 00:01:06,910
And then we will fit our extreme into this scale at.

12
00:01:17,500 --> 00:01:20,310
For this hour, Scaler is ready.

13
00:01:21,160 --> 00:01:26,200
Now let's first transform our extreme and to extreme scale.

14
00:01:26,530 --> 00:01:28,790
So we'll write X, underscore creon.

15
00:01:28,850 --> 00:01:30,010
Underscore s.

16
00:01:33,070 --> 00:01:35,690
He had our repellent is skilled.

17
00:01:37,140 --> 00:01:39,420
Right is skill learned, not transform.

18
00:01:41,550 --> 00:01:44,390
Remember, we created this scalar object double.

19
00:01:45,210 --> 00:01:46,470
Now we are transforming.

20
00:01:46,500 --> 00:01:47,430
What extreme?

21
00:01:54,450 --> 00:01:57,650
But on this hour, extreme is scarily variable is that.

22
00:02:00,070 --> 00:02:02,480
Now let's transform our test set also.

23
00:02:02,550 --> 00:02:06,380
So in right X underscored test underscore s.

24
00:02:08,870 --> 00:02:13,070
Equate to a scalar, not transform.

25
00:02:23,870 --> 00:02:29,960
X underscore test, just to give you a quick summary.

26
00:02:30,230 --> 00:02:37,370
Here we are first creating the scaler, which is exporting the information on how to divide each rule

27
00:02:37,400 --> 00:02:38,510
and each cell.

28
00:02:39,380 --> 00:02:46,340
Then we are using this is scalar object to transform our extreme variable so that we can create our

29
00:02:46,440 --> 00:02:47,570
Escorial variables.

30
00:02:48,710 --> 00:02:52,970
Similarly, we are using the same is scalar for our test set as well.

31
00:02:58,270 --> 00:03:01,390
Now we have we standardize over independent variables.

32
00:03:02,140 --> 00:03:04,720
Now let's move on to red integration.

33
00:03:07,150 --> 00:03:10,930
For religious aggression, we first need to import grit from Escala.

34
00:03:11,500 --> 00:03:17,650
So right from a Skillern, not linear model.

35
00:03:24,830 --> 00:03:25,790
Import grade.

36
00:03:32,190 --> 00:03:35,990
Now, let's run a simple, rich regression.

37
00:03:37,970 --> 00:03:41,520
Well, first cleared to object, LMR underscored odd

38
00:03:44,300 --> 00:03:45,750
liquid to build Ridge.

39
00:03:49,000 --> 00:03:53,520
And in record, we will mention our Lamela and Biton.

40
00:03:53,590 --> 00:03:54,550
We call it Dulfer.

41
00:03:54,670 --> 00:03:58,810
So we will write a five point two zero point five.

42
00:04:01,960 --> 00:04:04,330
So here I will memorize zero point five.

43
00:04:06,310 --> 00:04:08,760
And we are creating a Attridge object.

44
00:04:10,770 --> 00:04:14,670
Let's fit this model, Bill, right, elements of good are not fit.

45
00:04:19,700 --> 00:04:22,760
Now we will use over skill variables.

46
00:04:22,830 --> 00:04:26,170
We'll write X, underscore crane, underscore s.

47
00:04:32,960 --> 00:04:34,660
Comma, why underscore Craine?

48
00:04:39,060 --> 00:04:40,540
We have food data, what the first model.

49
00:04:40,890 --> 00:04:41,860
So let's find out.

50
00:04:42,980 --> 00:04:48,680
The artist squared value on the word test data will right out to underscore the school.

51
00:04:53,150 --> 00:04:57,830
Again, are to underscore is score is a function of skill and liablity.

52
00:04:57,860 --> 00:05:01,040
So remember, two important before running this function.

53
00:05:03,360 --> 00:05:04,870
We'll read like best.

54
00:05:08,270 --> 00:05:13,550
Then will write, L.M. underscored are not predict

55
00:05:16,360 --> 00:05:23,430
and in record, we'll write our best independent, weary ones, which is Xander Score Test underscore

56
00:05:23,540 --> 00:05:23,930
as.

57
00:05:25,370 --> 00:05:29,570
Here, our white based variable is a word or regional values.

58
00:05:30,710 --> 00:05:39,430
And the second argument is the predicted values of why on test variables can see the artist square value

59
00:05:40,460 --> 00:05:41,800
is zero point five four.

60
00:05:42,440 --> 00:05:49,580
Now, if you remember, we can also change the value of this LAMBROS or ULFA while running red regression,

61
00:05:51,290 --> 00:06:01,490
and we want to find out the value of all four Lamela for which the artist squared value is maximum for

62
00:06:01,490 --> 00:06:04,820
crying out such multiple values of hyper parameters.

63
00:06:05,270 --> 00:06:10,240
There is another function in Ashkelon, which is validation curves.

64
00:06:11,450 --> 00:06:20,020
So we'll first import the validation curve from Escalon Library and then I will show you how to select

65
00:06:20,020 --> 00:06:24,380
your optimum model from multiple values of hyper parameters.

66
00:06:27,490 --> 00:06:31,980
Well, first, import validation good right from Skillern, Todmorden, selection,

67
00:06:36,030 --> 00:06:37,330
import validation, good.

68
00:06:52,620 --> 00:06:57,250
Now, let's first see the parameters we need for this validation, good function.

69
00:06:57,530 --> 00:06:58,780
So, all right, validation.

70
00:06:59,710 --> 00:07:04,420
And then we will put a question mark to open the help for this function.

71
00:07:16,280 --> 00:07:20,270
If you see our first parameter should be an estimated.

72
00:07:23,080 --> 00:07:31,290
So now if we want to run rigid aggression, we should great, great, our second parameter is X variable.

73
00:07:31,540 --> 00:07:33,310
Our third parameter is very, very well.

74
00:07:33,730 --> 00:07:36,730
These are independent and dependent variables.

75
00:07:37,300 --> 00:07:41,020
Our fourth parameter is bad name.

76
00:07:42,010 --> 00:07:43,590
This is the hyper parameter.

77
00:07:43,650 --> 00:07:44,680
We want to change.

78
00:07:45,340 --> 00:07:48,130
In this case, we want to change the alpha value.

79
00:07:48,730 --> 00:07:48,990
So we.

80
00:07:49,040 --> 00:07:49,250
Right.

81
00:07:49,300 --> 00:07:50,080
I'll fire.

82
00:07:51,640 --> 00:07:54,890
Our fifth parameter is bad membranes.

83
00:07:55,570 --> 00:07:58,750
This is the values of this badam named parameter.

84
00:07:59,290 --> 00:08:00,100
You want to give.

85
00:08:01,480 --> 00:08:06,790
So here we will give my devil values all four alpha for our model.

86
00:08:09,870 --> 00:08:18,000
Now, we are not going to discuss groups, C.V. and other parameters, one another parameter we want

87
00:08:18,000 --> 00:08:19,830
to give is is scoring.

88
00:08:20,850 --> 00:08:24,370
We want our moderate to score using our square.

89
00:08:24,750 --> 00:08:28,080
So we will go escorting equate to are to.

90
00:08:30,860 --> 00:08:38,760
Now, to summarize, what we want to do is we want to train our models for multiple values of Lamela.

91
00:08:39,960 --> 00:08:44,230
So we will give multiple values of Lamela to our model.

92
00:08:44,790 --> 00:08:49,110
And we want the best model out of those lamda values.

93
00:08:49,920 --> 00:08:54,060
So we will run our model multiple times and will select the best model.

94
00:08:54,870 --> 00:08:56,250
All of those models.

95
00:09:00,650 --> 00:09:04,940
So first, let's create an array of over Lamela values.

96
00:09:06,830 --> 00:09:09,290
We will write Barum underscore range.

97
00:09:09,440 --> 00:09:10,670
This is our variable.

98
00:09:16,500 --> 00:09:20,970
We will equate it to and be not lock space.

99
00:09:26,470 --> 00:09:33,810
And then record will rate minus two, comma eight, comma hundred.

100
00:09:34,570 --> 00:09:43,990
Let me explain you this lock space function, this lock space function will create hundred values between

101
00:09:44,020 --> 00:09:48,220
Penrith's toward minus two and Ben Reas to deport eight.

102
00:09:51,650 --> 00:09:52,970
Now, if I just.

103
00:09:53,150 --> 00:10:01,930
The values of my parents change, you can see we have created a hundred different values between Penrith's

104
00:10:01,930 --> 00:10:05,350
who are minus two and 10 Race to the Power eight.

105
00:10:07,790 --> 00:10:13,090
We will use these values as the values of our lamda for our model.

106
00:10:13,730 --> 00:10:17,390
And then we will select the best Lamela value out of this.

107
00:10:17,420 --> 00:10:18,800
Different level of values.

108
00:10:19,910 --> 00:10:23,330
Now let's run the accretions for our modern.

109
00:10:26,440 --> 00:10:28,610
We saw validation curve earlier.

110
00:10:29,080 --> 00:10:33,430
The output of validation curve is in the form of two arrays.

111
00:10:33,650 --> 00:10:35,050
First is their training score.

112
00:10:35,140 --> 00:10:36,640
And second is their test score.

113
00:10:38,980 --> 00:10:42,240
I told you earlier that we will be scoring on Artist Square.

114
00:10:43,020 --> 00:10:48,700
So training score and test score will contain the values of our score for all these different models.

115
00:10:51,050 --> 00:10:53,450
Let's write Green underscored escort's.

116
00:10:55,840 --> 00:10:57,400
This other what variables?

117
00:11:03,950 --> 00:11:05,540
Equal to valuation of.

118
00:11:10,610 --> 00:11:15,360
Remember, the first barometer of validation good was the estimate.

119
00:11:15,830 --> 00:11:23,140
So here it assuming that is that it will write grade, second parameter was X value.

120
00:11:23,330 --> 00:11:26,720
So we'll write X and at school Kreen underscore to S.

121
00:11:30,490 --> 00:11:33,670
Next, but I mean, that is our way value.

122
00:11:38,430 --> 00:11:39,990
Next, but ometer was over.

123
00:11:40,050 --> 00:11:40,920
But I'm returning home.

124
00:11:44,070 --> 00:11:52,440
We will write ULFA in double quotation marks since we want to change the value of this Alpha.

125
00:11:56,690 --> 00:11:59,680
And then the next parameter was by Rembrandt.

126
00:12:01,760 --> 00:12:08,340
We created a bad Grange variable using lock space functions.

127
00:12:08,420 --> 00:12:09,950
We'll use this video.

128
00:12:10,370 --> 00:12:14,210
But next, we wonder what it is scoring to be off.

129
00:12:14,790 --> 00:12:15,980
Artist Kerswill, right.

130
00:12:16,670 --> 00:12:20,840
Scoring equal to a square.

131
00:12:23,560 --> 00:12:24,400
Fit on this.

132
00:12:26,740 --> 00:12:30,460
We have created this model for all this lamanno values.

133
00:12:31,900 --> 00:12:35,550
Now we will just print our CRANEY score and test the score.

134
00:12:39,620 --> 00:12:45,800
So notice we have not given our test set and to this validation.

135
00:12:47,690 --> 00:12:51,710
So how this validation, Gervase, calculating their test scores?

136
00:12:52,910 --> 00:12:59,930
This is actually running a cross validation or give for validation on our training data.

137
00:13:03,740 --> 00:13:08,150
So for each value of lambda, we are getting three values of our square.

138
00:13:09,290 --> 00:13:15,680
That is because the validation group is running Kiefel validation behind the scene.

139
00:13:18,740 --> 00:13:22,370
Let's just take a mean score of these three values.

140
00:13:22,640 --> 00:13:25,040
So, all right, dream underscored mean

141
00:13:28,120 --> 00:13:29,690
equal to and paid out mean.

142
00:13:33,510 --> 00:13:34,890
Will light green escort.

143
00:13:41,190 --> 00:13:46,920
Coma access equate to one, since we want the word every just to be along the roads.

144
00:13:49,590 --> 00:13:52,330
Similarly, we take the mean well lose over Tesla School.

145
00:14:04,000 --> 00:14:06,580
If we view this test mean Krien mean.

146
00:14:15,490 --> 00:14:21,820
So these are the hundred other square values for and Greg Lambros.

147
00:14:25,690 --> 00:14:29,610
Similarly, we have hungered values on our test score also.

148
00:14:30,910 --> 00:14:33,700
You can also compared with our lambda values.

149
00:14:34,480 --> 00:14:39,490
So for each of our Lemonheads values, these are the artist's good values we are getting.

150
00:14:40,150 --> 00:14:44,950
No, we are only interested in the model which have the highest artist's good value.

151
00:14:45,100 --> 00:14:52,180
So will write Max of our best mean.

152
00:14:59,300 --> 00:15:07,580
This will give me the highest artist square venue for my model, which is zero point seven two four

153
00:15:07,700 --> 00:15:08,240
six.

154
00:15:09,260 --> 00:15:13,410
Few remember earlier our artists got to where Lou was our own point.

155
00:15:13,470 --> 00:15:15,790
So I went to one nove.

156
00:15:16,940 --> 00:15:18,120
We have improved it.

157
00:15:18,140 --> 00:15:19,810
Two point seven dofor.

158
00:15:24,870 --> 00:15:31,410
Now, another way to look at this artist square where new data is to plot a graph of this will use.

159
00:15:33,000 --> 00:15:40,480
Let's load the scatterplot of this data along with our lamina values will right write asanas, don't

160
00:15:40,490 --> 00:15:41,220
join plot.

161
00:15:47,270 --> 00:15:50,990
And since I would never lose, you are in exponential form.

162
00:15:51,170 --> 00:15:53,040
Will take a low of those values.

163
00:16:11,030 --> 00:16:12,700
Overbuy is best mean.

164
00:16:23,220 --> 00:16:25,960
So this is the graph for artists good values.

165
00:16:26,580 --> 00:16:29,580
You can see as we are increasing our lamda.

166
00:16:30,820 --> 00:16:33,400
The artists go to Ray Lewis slightly in freezing.

167
00:16:34,680 --> 00:16:39,120
And after this point, the oddest really starts to decrease.

168
00:16:41,040 --> 00:16:44,670
And then our artists go to value as their own.

169
00:16:44,920 --> 00:16:46,120
Zero four.

170
00:16:46,470 --> 00:16:47,940
Very high lamda values.

171
00:16:48,810 --> 00:16:54,390
You can see we are getting the maximum artist's good value around this around this point.

172
00:16:55,530 --> 00:17:02,010
So what the next objective is to find this point and find the lamda value for this point.

173
00:17:03,000 --> 00:17:08,310
We see that the artists go to value of this pointis zero point seven two four six six.

174
00:17:11,170 --> 00:17:15,790
Not to find the location of all of this will lose it open.

175
00:17:15,880 --> 00:17:18,580
So I went to four double six will right.

176
00:17:18,880 --> 00:17:19,670
And B dot.

177
00:17:24,220 --> 00:17:25,510
Then our test mean

178
00:17:28,360 --> 00:17:33,460
we are just finding wherever testimony is is equal to this when you.

179
00:17:36,360 --> 00:17:41,840
We use Equilar design, the right mix of best me.

180
00:17:44,350 --> 00:17:51,940
We are just finding out the index where this best mean value is equal to maximum of best mean value.

181
00:17:53,200 --> 00:17:58,060
If we run this, you can see the location of this, Max.

182
00:17:58,120 --> 00:17:58,750
That's mean.

183
00:17:58,960 --> 00:18:00,760
Is that deterred index?

184
00:18:02,840 --> 00:18:04,610
So let's find out.

185
00:18:05,240 --> 00:18:10,820
The Lamela value of this started targeted index will rate Barum, underscore range

186
00:18:13,940 --> 00:18:15,030
and then a squared record.

187
00:18:15,080 --> 00:18:16,190
We will rate today three.

188
00:18:17,730 --> 00:18:18,540
If we run this.

189
00:18:20,600 --> 00:18:25,280
This is the Lamela value at which we're getting the maximum artist's good value.

190
00:18:27,260 --> 00:18:32,650
Now you have the LAMDA values for which we have the maximum artist's good value.

191
00:18:35,100 --> 00:18:37,900
We will fit our rage model again for this Lamela.

192
00:18:37,960 --> 00:18:39,470
Well, you will, right?

193
00:18:41,990 --> 00:18:44,570
Alem underscore our best.

194
00:18:48,070 --> 00:18:50,640
Equated to a rich

195
00:18:53,490 --> 00:18:54,670
and our other fiber.

196
00:18:54,810 --> 00:18:56,220
Lou, is this Lemina, Lou?

197
00:19:08,020 --> 00:19:09,500
We've done this, we have cleared that out.

198
00:19:09,830 --> 00:19:12,250
And them are underscore the best variable.

199
00:19:12,900 --> 00:19:16,030
Now, let's forget this on our training dataset.

200
00:19:22,590 --> 00:19:26,380
Then I would equity, but which is X underscore, Kreen underscore.

201
00:19:26,540 --> 00:19:30,210
S and then I would a variable.

202
00:19:36,950 --> 00:19:39,250
We've done this, we have tried our best.

203
00:19:39,330 --> 00:19:40,030
Rich Morton.

204
00:19:41,400 --> 00:19:42,420
Now let's find out.

205
00:19:42,440 --> 00:19:44,880
The artist's good value on our test data.

206
00:19:51,090 --> 00:19:53,360
Values are to under school to.

207
00:19:54,960 --> 00:20:01,330
Then why test and then Alem underscored are under.

208
00:20:01,620 --> 00:20:02,040
Best.

209
00:20:06,570 --> 00:20:11,130
Not predict then X underscore, test, underscore as.

210
00:20:19,650 --> 00:20:20,980
Now, let's find out.

211
00:20:21,030 --> 00:20:23,160
The artist could of a new photo or training data.

212
00:20:38,330 --> 00:20:46,380
And this hour, artists go to a new photo ID training data ascertains zero points on five for it and

213
00:20:46,380 --> 00:20:49,500
for our test data is zero point five four zero.

214
00:20:52,580 --> 00:20:58,820
You will not see any major difference and not as good a value if we run underage or if we're on simple

215
00:20:58,820 --> 00:21:05,780
linear regression in our case, because our data set is very small and it will hardly make any difference

216
00:21:05,930 --> 00:21:14,550
which algorithm or with technique we are using to train our model in case we have hundreds or thousands

217
00:21:14,550 --> 00:21:17,170
of variable and our dataset is very large.

218
00:21:17,300 --> 00:21:21,680
In that case, this my thoughts would really have an answer.

219
00:21:21,710 --> 00:21:22,280
What a model.

220
00:21:24,020 --> 00:21:29,900
Just to get all the things we did first, we standardized our X variable.

221
00:21:31,100 --> 00:21:40,810
We created this standard scalar and we use this scalar to transform our extrem and expressed data.

222
00:21:42,680 --> 00:21:51,770
After that, we executed the raid more than and there we use a constant other form of zero point five.

223
00:21:53,810 --> 00:21:59,690
Now, after that, we created the Angel full force, which would go into our model.

224
00:22:01,130 --> 00:22:07,720
So we created a range of hundred Lambros from 10 that super to minus two, two Penrith's toward eight.

225
00:22:08,480 --> 00:22:10,250
Then we use the word validation.

226
00:22:10,290 --> 00:22:13,550
Got to find the optimum model.

227
00:22:15,680 --> 00:22:22,560
After that, we calculated the lambda value corresponding to that more.

228
00:22:22,590 --> 00:22:29,570
Then after that, we again executed the order rates more for this value of lambda.

229
00:22:30,770 --> 00:22:38,710
And then we calculated the artist's good values on our training dataset and test their Dussart.

230
00:22:41,030 --> 00:22:44,060
Now running LASO is almost similar.

231
00:22:44,810 --> 00:22:48,410
We just have to import LASO and set off Rich.

232
00:22:52,970 --> 00:23:00,020
We will import LASO from Escalon Linnean more the library, and then we will soon.

233
00:23:00,020 --> 00:23:07,340
The leak cleared Lazor object as we created the LMR and that's what our object here.

234
00:23:07,340 --> 00:23:10,070
We will create LMR underscore object.

235
00:23:21,850 --> 00:23:23,870
Similarly, we will give overvalues.

236
00:23:31,910 --> 00:23:35,730
As you can see, executing LASO is very similar to red.

237
00:23:36,560 --> 00:23:39,170
We are not executing LASO here.

238
00:23:39,710 --> 00:23:48,230
I recommend you to execute it on your own and find the optimum value of Alpha using validation.

239
00:23:48,230 --> 00:23:48,420
Cut.

240
00:23:50,420 --> 00:23:50,870
Thanks.