1
00:00:00,450 --> 00:00:02,610
Okay you've made it.

2
00:00:02,820 --> 00:00:04,180
Look how far we've come.

3
00:00:04,290 --> 00:00:10,140
We've covered an absolutely massive chunk of the psychic line library and we're still going to finish

4
00:00:10,140 --> 00:00:10,650
it off.

5
00:00:10,650 --> 00:00:13,710
We're up to the last section putting it all together.

6
00:00:13,800 --> 00:00:15,810
We've been through an intense workflow.

7
00:00:15,810 --> 00:00:17,690
We've seen how to get the data ready.

8
00:00:17,700 --> 00:00:21,020
We've seen how to choose the right machine learning model for our problems.

9
00:00:21,030 --> 00:00:22,660
We fill it to the data.

10
00:00:22,770 --> 00:00:24,730
We've used it to make predictions.

11
00:00:24,750 --> 00:00:26,740
We've evaluated a trained model.

12
00:00:26,820 --> 00:00:28,890
We've improved a drain model.

13
00:00:28,890 --> 00:00:30,970
We've saved and loaded a trained model.

14
00:00:30,990 --> 00:00:39,060
Now let's combine all of this all of this could we do it could we do it now let's make a little heading

15
00:00:40,140 --> 00:00:49,800
7 Buono putting it all together first thing give yourself a little pat on the back right.

16
00:00:49,810 --> 00:00:56,410
Because we've covered a lot and so far it seems to be all over the place in which it is but not to worry.

17
00:00:56,420 --> 00:01:02,300
This is how machine learning projects often start out as a whole bunch of different experiments and

18
00:01:02,300 --> 00:01:03,880
code all over the place.

19
00:01:04,130 --> 00:01:09,220
And then once you've found something which works the refinement process begins.

20
00:01:09,430 --> 00:01:12,650
And now what would this refinement process look like.

21
00:01:12,650 --> 00:01:17,960
Well let's use remember our good old car sales regression problem a.k.a. predicting the sale price of

22
00:01:17,960 --> 00:01:18,860
cars.

23
00:01:18,860 --> 00:01:21,970
We'll use that as an example and to tidy things up.

24
00:01:22,040 --> 00:01:24,450
We're going to be using psychic loans.

25
00:01:24,470 --> 00:01:24,980
Let's have a look.

26
00:01:24,990 --> 00:01:28,760
Psychic learn pipeline to tidy things up.

27
00:01:28,760 --> 00:01:31,490
We're going to be using the pipeline class.

28
00:01:31,670 --> 00:01:38,120
Now you can imagine pipeline as being a way to string together a number of different socket line processes

29
00:01:38,450 --> 00:01:41,090
in one hit so similar to like writing a function right.

30
00:01:41,090 --> 00:01:43,400
You're getting a couple of different steps and you're putting them together.

31
00:01:44,860 --> 00:01:50,500
Pipeline of Transformers with the final list minor so a transformer or Transformers a.k.a. transforming

32
00:01:50,500 --> 00:01:50,900
data.

33
00:01:50,920 --> 00:01:54,370
Getting data ready and then a final estimate are estimated.

34
00:01:54,370 --> 00:02:01,210
Remember in psychic loan terms is a machine learning model so as a refresh this is our car sales data

35
00:02:02,130 --> 00:02:05,810
is a thousand rows and there's some missing data here right.

36
00:02:05,830 --> 00:02:07,200
So there's missing here.

37
00:02:07,240 --> 00:02:08,050
Missing here.

38
00:02:08,050 --> 00:02:15,580
And what we're trying to do is use make color odometer doors to predict the sale price of cars.

39
00:02:15,580 --> 00:02:19,480
Now again this data is imperfect because could you do this.

40
00:02:19,480 --> 00:02:21,570
Could you figure this out given enough time.

41
00:02:21,700 --> 00:02:26,650
You'd probably realistically need a few more parameters a few more details about the cars before you

42
00:02:26,650 --> 00:02:29,230
could figure out how much they'd sell for.

43
00:02:29,560 --> 00:02:36,300
So let's stop talking about it let's get hands on with the code to to start off we'll report the data.

44
00:02:36,310 --> 00:02:40,290
Go pay data read see as v dot dot.

45
00:02:40,300 --> 00:02:42,360
We've saved it in our data voter.

46
00:02:43,000 --> 00:02:48,620
Car sales extended online in a data folder is not is not up the chain.

47
00:02:48,750 --> 00:02:51,640
I want car sales.

48
00:02:52,030 --> 00:02:52,890
There we go.

49
00:02:52,900 --> 00:02:55,790
Car sales extended data missing.

50
00:02:56,200 --> 00:02:57,520
That's what we want.

51
00:02:57,520 --> 00:02:59,060
And then we'll have a look.

52
00:02:59,530 --> 00:03:02,470
We'll have a quick peek at it wonderful.

53
00:03:02,500 --> 00:03:10,820
So this is just reporting this CSB fall that we have here comeback now.

54
00:03:10,920 --> 00:03:12,360
What did we say.

55
00:03:12,360 --> 00:03:17,670
We said we want to use these columns so these four columns to predict the price column we'll have a

56
00:03:17,670 --> 00:03:19,350
look at our data types.

57
00:03:19,410 --> 00:03:24,450
So you've got make which is an object that's a string their color which is an object that's a string

58
00:03:24,450 --> 00:03:31,330
their odometer doors and price which are we've kind of talked about doors being a category as well.

59
00:03:31,680 --> 00:03:34,050
So odometer is our numerical column.

60
00:03:34,050 --> 00:03:36,810
And do you remember what we have to do for our data.

61
00:03:36,810 --> 00:03:41,430
We have to make sure that it's all in numerical format before we can build a machine learning model

62
00:03:41,430 --> 00:03:46,160
on it and there's another little roadblock that we have as well before we can build our machine learning

63
00:03:46,160 --> 00:03:48,530
model and that figure is an A.

64
00:03:48,590 --> 00:03:55,850
So is now not some we've got no values in some of our while all of our columns we've got 50 missing

65
00:03:55,970 --> 00:03:57,860
missing data points in each of our columns.

66
00:03:58,430 --> 00:04:01,090
So that's also a big no no with machine learning models.

67
00:04:01,100 --> 00:04:07,340
We need to make sure that the data is numerical and that there's no missing values we refer back to

68
00:04:07,340 --> 00:04:10,930
our keynote or data should be numerical.

69
00:04:10,970 --> 00:04:12,270
This is things to remember right.

70
00:04:13,010 --> 00:04:14,910
And there should be no missing values.

71
00:04:15,090 --> 00:04:17,800
Okay let's check it out.

72
00:04:17,810 --> 00:04:23,210
So what we'll have to do is if we want to put this all together we've done this before we've seen filling

73
00:04:23,210 --> 00:04:30,570
data we've seen converting it but this time we want to use a pipeline to do so so the socket lines pipelines

74
00:04:30,600 --> 00:04:39,890
main import is steps which is a list that contains a topple which has step name and action to take.

75
00:04:39,890 --> 00:04:45,220
We're gonna see this in a second what we're just talking through it and in our case our steps are what

76
00:04:45,220 --> 00:04:46,090
are our steps.

77
00:04:46,090 --> 00:04:54,530
Well first we need to fill the missing data and then convert the data to numbers and then build a machine

78
00:04:54,530 --> 00:04:56,020
learning model on it.

79
00:04:56,090 --> 00:04:57,150
So let's do it.

80
00:04:57,360 --> 00:04:59,090
We'll go here.

81
00:04:59,090 --> 00:05:01,440
I want to do this one so it's all in one cell.

82
00:05:01,460 --> 00:05:02,560
Let's put our steps here.

83
00:05:02,570 --> 00:05:06,130
What we have to do steps.

84
00:05:06,310 --> 00:05:09,850
We want to do all in one cell.

85
00:05:09,850 --> 00:05:11,170
Can we do it all in one cell.

86
00:05:11,170 --> 00:05:15,520
Because before to do this it took a fair bit of work but let's say this is what we're doing.

87
00:05:15,520 --> 00:05:18,970
This is our refinement process using pipeline.

88
00:05:19,000 --> 00:05:28,590
So step one is fill missing data and then step two is convert data to numbers.

89
00:05:28,690 --> 00:05:29,620
Wonderful.

90
00:05:29,710 --> 00:05:32,700
And then step 3 is build a model on the data.

91
00:05:35,180 --> 00:05:35,810
All right.

92
00:05:35,970 --> 00:05:39,250
Cracks knuckles gets fingers ready let's do them.

93
00:05:39,300 --> 00:05:44,220
And if it seems like I'm going through this pretty fast let me tell you there's nothing here that we

94
00:05:44,220 --> 00:05:49,440
haven't covered except for the pipeline but what we've learned in the previous videos you'll be able

95
00:05:49,440 --> 00:05:52,910
to deduce as well as reading the documentation here.

96
00:05:52,950 --> 00:05:54,570
What exactly is going on.

97
00:05:54,710 --> 00:06:02,510
So import panders is payday because you want to do this all in one cell and then from SBA loan dot com

98
00:06:02,510 --> 00:06:03,840
pose.

99
00:06:03,900 --> 00:06:07,320
Remember the column transformer.

100
00:06:07,320 --> 00:06:09,110
And then from SBA loan.

101
00:06:09,120 --> 00:06:13,190
This is where we're going to import the pipeline class that we're going to take advantage of here.

102
00:06:13,230 --> 00:06:21,250
Import pipeline and then from SBA loan not impute we're going to import simple computer.

103
00:06:21,250 --> 00:06:24,260
This is what we're going to use to impute our data.

104
00:06:24,270 --> 00:06:26,330
So fill the missing data.

105
00:06:26,810 --> 00:06:33,830
And then from as K line dot pre processing import one hot encoder.

106
00:06:34,080 --> 00:06:39,180
This is what we're going to do to convert our data to numbers or sorry we're going to use to convert

107
00:06:39,180 --> 00:06:40,070
our data to number.

108
00:06:41,010 --> 00:06:44,140
So getting data ready.

109
00:06:44,160 --> 00:06:52,480
These are the inputs we're going to use for that and the next one is modeling so from Skyline dot ensemble

110
00:06:53,760 --> 00:06:58,090
import our trusty Random Forests regress.

111
00:06:58,260 --> 00:07:07,240
Because we wanted to predict a number and then from SBA loan model selection import over there right.

112
00:07:07,510 --> 00:07:08,320
Yeah that looks right.

113
00:07:08,350 --> 00:07:13,870
I've been known to be prone to typos try and test blimp and we want grid search CV because we've seen

114
00:07:14,140 --> 00:07:19,690
grid search being used before which is gonna search for some optimal hover parameters for our Random

115
00:07:19,690 --> 00:07:21,040
Forest aggressor.

116
00:07:21,040 --> 00:07:32,510
Okay now we're going to set up random seed so we'll import num pi as NDP MP random seed 42.

117
00:07:32,950 --> 00:07:39,690
Wonderful now for completeness for doing it all in one cell we're going to reinstall the dinosaur import

118
00:07:39,690 --> 00:07:50,530
data and we're going to drop the rows with missing labels so data equals PD CSB.

119
00:07:50,860 --> 00:07:53,470
We're going to go to our data folder.

120
00:07:53,500 --> 00:07:55,700
This needs to be a string to go.

121
00:07:55,720 --> 00:07:59,910
Car sales extended missing data.

122
00:07:59,920 --> 00:08:07,490
That's the one we're after the thousand Rose version and then to drop the rows that have missing labels.

123
00:08:07,570 --> 00:08:14,440
We'll go data dot drop in a subset we'll set that equal to the price column because that's what we want

124
00:08:14,440 --> 00:08:18,910
to predict and that's our label and we're going to go in place equals true.

125
00:08:19,030 --> 00:08:19,930
Okay.

126
00:08:19,930 --> 00:08:28,360
So what we've done is we've just replicated this here and now we've dropped the rows which contain missing

127
00:08:28,540 --> 00:08:32,810
price values so a.k.a. missing labels.

128
00:08:32,830 --> 00:08:38,830
Now what we're going to do here is we're going to start to define the different features of the dataset

129
00:08:39,370 --> 00:08:46,300
and the different transformer pipelines that we want to take place on that data.

130
00:08:46,300 --> 00:08:52,360
So if we come back up here what transformations do we need to make we need to change our making color

131
00:08:53,110 --> 00:08:54,520
into numbers.

132
00:08:54,850 --> 00:09:01,840
We need to fill the missing values of all the columns in here and we need to change our doors because

133
00:09:01,840 --> 00:09:05,530
it's also a category we need to change it to a category of value.

134
00:09:05,530 --> 00:09:09,940
So we're going to actually know forget what I said about doors.

135
00:09:10,130 --> 00:09:12,700
Let's see what we're actually doing in the code.

136
00:09:12,920 --> 00:09:18,740
So define different features and transformer pipelines.

137
00:09:19,460 --> 00:09:26,870
So for our categorical features doors doors is a special case because it's already numerical categorical

138
00:09:26,870 --> 00:09:32,090
features equals make and color.

139
00:09:32,110 --> 00:09:37,230
Now what we're going to do here is create a categorical transformer.

140
00:09:37,270 --> 00:09:41,500
So this just means modifying the data and to do so we're going to use a pipeline.

141
00:09:41,500 --> 00:09:47,530
This is the first time we're seeing pipeline but the steps in the pipeline are going to be first we're

142
00:09:47,530 --> 00:09:51,240
going to fill with missing values within puter.

143
00:09:51,430 --> 00:09:55,540
So this is where the pipeline inputs take steps to take.

144
00:09:55,660 --> 00:09:55,920
Right.

145
00:09:55,930 --> 00:10:00,550
One after the other as a list of topples so let's see this happening.

146
00:10:00,550 --> 00:10:06,470
The name is in pewter and we're going to remember this is the action to take.

147
00:10:06,520 --> 00:10:14,070
So simple computer strategy equals constant.

148
00:10:14,530 --> 00:10:17,610
And then the fill value is going to be missing.

149
00:10:19,360 --> 00:10:27,090
So you might be out of juice what's happening there with this particular case where imputing the categorical

150
00:10:27,090 --> 00:10:31,930
features with the constant value of missing string.

151
00:10:31,950 --> 00:10:41,780
Now we're going to create one hot and go one hot encoder and to handle the unknowns the columns that

152
00:10:41,780 --> 00:10:45,580
hasn't seen before we're going to tell it to ignore them.

153
00:10:46,740 --> 00:10:47,910
Wonderful.

154
00:10:48,210 --> 00:10:55,050
And then we'll make sure that we've all lined up or lined up these things that we're gonna need to we're

155
00:10:55,050 --> 00:10:56,620
missing something here.

156
00:10:56,640 --> 00:11:00,520
Need to check where our brackets are at.

157
00:11:00,710 --> 00:11:02,730
We're missing one here.

158
00:11:02,750 --> 00:11:04,130
There we go.

159
00:11:04,160 --> 00:11:09,800
See half of this is our figuring out what you're missing racquets are and then door feature.

160
00:11:09,800 --> 00:11:15,530
We're going to define our door feature Eagles doors.

161
00:11:15,780 --> 00:11:23,230
And so the same thing again so we're gonna create a door transformer door.

162
00:11:23,240 --> 00:11:29,650
Transformer equals pipeline again we're using a pipeline because this will all make sense in a second

163
00:11:29,650 --> 00:11:29,870
right.

164
00:11:29,900 --> 00:11:38,400
Why we're setting out pipelines so to do this we have to press impute.

165
00:11:39,470 --> 00:11:48,360
We want to do the same thing again simple computer and the strategy is going to be constant and the

166
00:11:48,360 --> 00:11:54,900
full value for our doors is we're just going to fill it out with force because four is the majority

167
00:11:54,900 --> 00:11:56,910
number in the doors column.

168
00:11:56,910 --> 00:12:01,040
Then next we want numeric features.

169
00:12:01,170 --> 00:12:06,370
So our numeric features is our domain a common Cam.

170
00:12:06,380 --> 00:12:07,190
Wonderful.

171
00:12:07,190 --> 00:12:15,620
And then we're going to create a numeric transformer equals first saying we're going to do this pipeline

172
00:12:16,670 --> 00:12:22,900
steps equal and then again we need to impute values here.

173
00:12:22,930 --> 00:12:32,820
So impute a simple computer strategy equals mean.

174
00:12:32,890 --> 00:12:39,910
So this means it's going to fill the numerical columns in our case the odometer with the strategy means

175
00:12:39,910 --> 00:12:46,090
is gonna take the main value of the odometer column and fill all of the missing rows in odometer with

176
00:12:46,090 --> 00:12:49,300
the main of the rest of the values.

177
00:12:49,390 --> 00:12:49,920
That was a lot.

178
00:12:50,140 --> 00:12:51,880
Now still not done.

179
00:12:51,880 --> 00:12:53,170
That's right.

180
00:12:53,710 --> 00:12:57,630
We're gonna set up the pre processing steps.

181
00:12:58,180 --> 00:13:01,180
So remember what we need to do we need to fill missing data.

182
00:13:01,240 --> 00:13:05,330
So this is part of our pre processing these to a pre processing our data.

183
00:13:05,590 --> 00:13:07,250
And this is modelling the data.

184
00:13:07,480 --> 00:13:12,790
So our pre processing steps are going to be fill missing data and convert data to numbers so we can

185
00:13:12,790 --> 00:13:19,100
do that fill missing values then convert to numbers.

186
00:13:19,550 --> 00:13:29,820
So to create a pre processor will instantiate a pre processor using column transformer tab order complete.

187
00:13:30,820 --> 00:13:38,370
And we're going to pass it column transformer takes Transformers Eagles.

188
00:13:38,430 --> 00:13:43,290
And now this is going to be a list of the transformations we want to do on our data and in our case

189
00:13:43,300 --> 00:13:50,050
we have one called Cat for short and categorical it's going to it the categorical transformer and we

190
00:13:50,050 --> 00:13:56,050
want those transformations that we've defined up here the categorical transformer to fill a values simple

191
00:13:56,050 --> 00:14:00,350
computer with missing and to one hot in code making color.

192
00:14:00,370 --> 00:14:08,580
We want those to be performed on the categorical features and press tab or complete categorical features.

193
00:14:08,580 --> 00:14:09,240
Wonderful.

194
00:14:09,240 --> 00:14:11,070
Now we pass it another topple.

195
00:14:11,130 --> 00:14:21,530
So now we want it to do same steps but this time using the door transformer which is going to fill the

196
00:14:21,530 --> 00:14:24,570
door column with the constant value of four.

197
00:14:25,000 --> 00:14:29,730
So go door feature wonderful.

198
00:14:29,860 --> 00:14:37,000
And finally I wanted to finish up using the nom transformer which is going to take our numeric transformer

199
00:14:38,990 --> 00:14:44,140
and perform the imputation on the numeric features

200
00:14:46,840 --> 00:14:52,600
beautiful when there's only one feature so features is as a plural is probably pulling not correct.

201
00:14:52,600 --> 00:14:58,060
And now now that we've done our pre processing and so our data is getting filled the missing values

202
00:14:58,060 --> 00:14:58,750
are getting filled.

203
00:14:58,750 --> 00:15:02,650
And then it's getting converted to numbers using what we've done here.

204
00:15:03,710 --> 00:15:08,310
The next step is to create a pre processing and modelling pipeline.

205
00:15:08,350 --> 00:15:12,310
We can do this as you maybe had a guess because I've used the word pipeline.

206
00:15:12,340 --> 00:15:19,360
We need to create a pre processing and modelling pipeline so now essentially what we're doing is we're

207
00:15:19,360 --> 00:15:23,550
putting our pre processing steps together with a modelling step.

208
00:15:23,550 --> 00:15:27,880
And because there's multiple steps we'll use a pipeline to combine them.

209
00:15:28,000 --> 00:15:30,970
So model equals pipeline

210
00:15:34,070 --> 00:15:38,980
deposit steps which is a list containing tar balls.

211
00:15:39,010 --> 00:15:41,380
And this is the name.

212
00:15:41,380 --> 00:15:50,440
So this is a pretty process a step and we're going to pass it pre processor so the first step in our

213
00:15:50,440 --> 00:15:58,220
pipeline here our model pipeline is to run through this and then the next step will put a little camera

214
00:15:58,220 --> 00:15:58,490
here.

215
00:15:58,490 --> 00:16:09,400
We need to create another top on which is going to be our model is our Random Forest regress up wonderful.

216
00:16:09,410 --> 00:16:14,780
So now we've got our model pipeline setup which will go through pre processing once the pre processing

217
00:16:14,780 --> 00:16:15,500
is done.

218
00:16:15,530 --> 00:16:18,580
It'll build a random forest classifier on it.

219
00:16:18,590 --> 00:16:23,630
We can get our data ready so split data X equals data don't draw.

220
00:16:23,930 --> 00:16:27,220
We're going to remove the price column because that's what we want to predict.

221
00:16:27,230 --> 00:16:37,370
Access is going to equal one Y is going to be data price then we're going to create X train x test as

222
00:16:37,370 --> 00:16:38,230
we always do.

223
00:16:38,270 --> 00:16:39,490
Why train.

224
00:16:39,830 --> 00:16:46,390
Why test equals try and test split x y test size.

225
00:16:46,390 --> 00:16:58,810
We'll use zero point two and then we own it fit and score the model to model don't fit X train y trying

226
00:16:59,720 --> 00:17:06,580
now model dot school X test y test where we got it wrong.

227
00:17:06,580 --> 00:17:07,700
Handle unknown.

228
00:17:07,780 --> 00:17:10,870
We've got a typo of course we do handle unknown.

229
00:17:10,870 --> 00:17:11,880
There we go.

230
00:17:12,250 --> 00:17:14,870
Change that.

231
00:17:15,110 --> 00:17:16,190
We did it.

232
00:17:16,190 --> 00:17:23,460
We did the entire pull up line in one cell taking advantage of the pipeline class so let's step through

233
00:17:23,460 --> 00:17:24,420
what we did.

234
00:17:24,630 --> 00:17:31,440
We imported the classes to get our data ready Panda's colon transformer pipeline simple and pewter one

235
00:17:31,440 --> 00:17:38,480
hot encoder we imported the classes to do our modeling so random forest aggressor trying to split grid

236
00:17:38,480 --> 00:17:46,710
search CV we set up a random seed so our results are reproducible and then we define different features

237
00:17:46,710 --> 00:17:48,300
and transform our pipelines.

238
00:17:48,300 --> 00:17:50,680
So we defined the categorical features here.

239
00:17:50,910 --> 00:17:56,430
The door feature the numeric feature and then for each of them we created a few steps for our pipeline

240
00:17:56,430 --> 00:18:03,000
to take to not only impute missing values but to one hot encode so turn all of our data that wasn't

241
00:18:03,000 --> 00:18:13,210
numbers into numbers and then we set up a pre processing step that took all of our pipelines here to

242
00:18:13,720 --> 00:18:19,690
take the steps that we wanted to take on the categorical features on the door feature and on the numerical

243
00:18:19,690 --> 00:18:21,000
feature.

244
00:18:21,010 --> 00:18:28,100
Then we created another pipeline to put all of these steps together to not only pre process our data

245
00:18:29,120 --> 00:18:34,610
and then model it so then we split our data into x and y and then into train and test.

246
00:18:34,970 --> 00:18:40,750
And so when we call model dot fit it's gonna call this variable here which is going to run this pipeline

247
00:18:42,360 --> 00:18:48,210
which inherently is gonna call the pre process a step which is going to run this pipeline which is inherently

248
00:18:48,210 --> 00:18:50,280
going to run all of these three steps.

249
00:18:50,400 --> 00:18:54,060
You know I'd be thinking Daniel while this has got a lot going on but if you go back through what you've

250
00:18:54,060 --> 00:18:57,810
looked at here you'll be had to say you can step through it and go okay.

251
00:18:57,990 --> 00:19:02,370
I'm starting to piece together what's happening and now if you scroll right back up to where we originally

252
00:19:02,370 --> 00:19:07,500
did this filling data with pandas filling data with SDK learn and then modelling it on our car sales

253
00:19:07,500 --> 00:19:12,200
problem you'll see that we did it over a number of different cells and now we're doing it.

254
00:19:12,360 --> 00:19:17,880
We're getting a full blown machine learning model and 55 lines of code minus some comments and we could

255
00:19:17,880 --> 00:19:20,210
probably tidy this up if we're being honest with ourselves.

256
00:19:20,730 --> 00:19:26,790
But now you're starting to see is an end the beginning machine learning is very experimental throughout

257
00:19:26,790 --> 00:19:29,550
the whole thing and it's very experimental.

258
00:19:29,550 --> 00:19:35,820
And so what it takes at the start is just code everywhere trying different things seeing how they work

259
00:19:36,240 --> 00:19:38,980
and then eventually you can bring it down to something like this.

260
00:19:39,030 --> 00:19:43,890
Put it all into one big cell in a pipeline going through the steps that you've figured out along the

261
00:19:43,890 --> 00:19:45,060
way.

262
00:19:45,060 --> 00:19:50,610
Now of course we'd want to probably improve this use hyper parameter tuning and that's what we might

263
00:19:50,610 --> 00:19:57,660
look at in the next video is how to use hyper parameter tuning with a pipeline slightly different to

264
00:19:57,660 --> 00:20:02,570
how we've seen it done with just a regular model so take a little break step back through what we've

265
00:20:02,570 --> 00:20:03,680
done here because it was a lot.

266
00:20:03,680 --> 00:20:05,900
Don't worry if you don't take it in entirely.

267
00:20:05,900 --> 00:20:07,880
Have a read through the pipeline.

268
00:20:07,880 --> 00:20:13,790
Class of socket loan documentation and see if you can line up what we've done in the documentation to

269
00:20:13,790 --> 00:20:19,380
what we've done here otherwise I'll see you in the next video we'll see if we can improve this score.