﻿1
00:00:00,200 --> 00:00:02,500
Hello and welcome to this art tutorial.

2
00:00:02,500 --> 00:00:06,900
So we just connect it to our H2O instance
and now things are going to get easy.

3
00:00:07,466 --> 00:00:08,833
That's another reason why.

4
00:00:08,833 --> 00:00:11,166
This H2O package is great.

5
00:00:11,166 --> 00:00:15,233
Because in a few lines of code,
we're going to be able to build a complex

6
00:00:15,233 --> 00:00:17,700
deep learning model.
And this will be very simple.

7
00:00:17,700 --> 00:00:20,933
We'll just use one function
and then input different parameters.

8
00:00:21,066 --> 00:00:26,400
I'm going to first define our classifier
that I call classifier then equals.

9
00:00:26,566 --> 00:00:29,266
Then that's.
Where we use the H2O. Function.

10
00:00:29,266 --> 00:00:31,200
So to take this function we need to take.

11
00:00:31,200 --> 00:00:34,200
First the H2O package.

12
00:00:34,300 --> 00:00:36,166
And then the function
that we're going to use

13
00:00:36,166 --> 00:00:38,133
right now is one of the functions here.

14
00:00:38,133 --> 00:00:38,533
And you are

15
00:00:38,533 --> 00:00:41,866
going to like the name of this function
since it is called deep learning.

16
00:00:42,300 --> 00:00:43,700
So that here.

17
00:00:43,700 --> 00:00:45,600
And then. Deep.

18
00:00:45,600 --> 00:00:47,200
And here it is deep learning.

19
00:00:47,200 --> 00:00:48,133
That's the function we'll.

20
00:00:48,133 --> 00:00:50,833
Use to build. Our deep learning model.

21
00:00:50,833 --> 00:00:53,100
Pressing enter and here we go.

22
00:00:53,100 --> 00:00:56,100
Now we just need
to input the different arguments.

23
00:00:56,700 --> 00:00:57,933
So what I'm going to do now.

24
00:00:57,933 --> 00:01:00,933
Is go here. And press F1.

25
00:01:01,366 --> 00:01:03,100
To have a better look at all the.

26
00:01:03,100 --> 00:01:05,100
Arguments. Of this H2O.

27
00:01:05,100 --> 00:01:06,800
Deep learning function.

28
00:01:06,800 --> 00:01:07,766
So let's see.

29
00:01:07,766 --> 00:01:10,766
Let's scroll down here
to find the arguments.

30
00:01:10,933 --> 00:01:12,266
And here we go.

31
00:01:12,266 --> 00:01:15,533
So the first argument is x vector
containing the character

32
00:01:15,533 --> 00:01:17,700
names of the predictors in the model.

33
00:01:17,700 --> 00:01:19,933
That's actually not a required argument.

34
00:01:19,933 --> 00:01:21,000
We don't need to input it.

35
00:01:21,000 --> 00:01:23,000
So we'll move on to the next one.

36
00:01:23,000 --> 00:01:24,233
So the next one is y.

37
00:01:24,233 --> 00:01:27,166
The name of the response
variable in the model.

38
00:01:27,166 --> 00:01:27,900
So we will have.

39
00:01:27,900 --> 00:01:31,733
To use this one because actually
the next argument is training frame.

40
00:01:31,900 --> 00:01:33,400
And this training frame argument is.

41
00:01:33,400 --> 00:01:35,100
Basically the. Training set.

42
00:01:35,100 --> 00:01:39,466
That means that for this training frame
argument will input our training set here.

43
00:01:39,900 --> 00:01:43,400
And since the training set contains
both the independent variables

44
00:01:43,400 --> 00:01:45,300
and the dependent variable, we need to.

45
00:01:45,300 --> 00:01:46,800
Specify with this y.

46
00:01:46,800 --> 00:01:50,733
Argument here, what is the dependent
variable in this training frame.

47
00:01:50,966 --> 00:01:52,533
That is our training set.

48
00:01:52,533 --> 00:01:55,533
And so with these two informations
training frame and y

49
00:01:55,566 --> 00:01:58,766
the model will then understand
what is the training set.

50
00:01:58,800 --> 00:02:01,933
What are the independent variables
and what is the dependent variable.

51
00:02:02,333 --> 00:02:03,466
So let's start.

52
00:02:03,466 --> 00:02:05,366
By inputting these two arguments.

53
00:02:05,366 --> 00:02:07,600
First y equals.

54
00:02:07,600 --> 00:02:12,066
And as you can see it is just asking
for the name of the response variable.

55
00:02:12,066 --> 00:02:13,500
That is the dependent variable.

56
00:02:13,500 --> 00:02:15,133
So here we just need to.

57
00:02:15,133 --> 00:02:18,233
Input in quotes
the name of the dependent variable.

58
00:02:18,333 --> 00:02:20,433
Which is exited.

59
00:02:20,433 --> 00:02:23,433
All right done
for the first argument comma.

60
00:02:23,500 --> 00:02:26,233
And then moving on to the second argument.

61
00:02:26,233 --> 00:02:30,633
So the second argument
is training underscore frame.

62
00:02:31,700 --> 00:02:32,700
Here we go.

63
00:02:32,700 --> 00:02:34,233
And so that's our training set.

64
00:02:34,233 --> 00:02:37,800
And therefore we input here training set.

65
00:02:38,233 --> 00:02:38,533
All right.

66
00:02:38,533 --> 00:02:41,133
But now be. Careful because as it is said.

67
00:02:41,133 --> 00:02:43,866
In this training frame argument info here.

68
00:02:43,866 --> 00:02:46,866
This is an H2O frame object.

69
00:02:47,033 --> 00:02:50,100
And right now our training set
is not an H2O frame object.

70
00:02:50,366 --> 00:02:51,866
It is just. A training set.

71
00:02:51,866 --> 00:02:53,233
That is a data frame.

72
00:02:53,233 --> 00:02:55,266
But that's not an H2O frame.

73
00:02:55,266 --> 00:02:57,166
So we need to convert.

74
00:02:57,166 --> 00:02:57,600
This.

75
00:02:57,600 --> 00:03:00,633
Data frame training set into an H2O frame.

76
00:03:01,000 --> 00:03:03,433
And to do this,
there is one very simple way.

77
00:03:03,433 --> 00:03:07,533
It is to use the as dot h2o.

78
00:03:07,666 --> 00:03:08,400
Function.

79
00:03:08,400 --> 00:03:12,200
That is a function
that will convert its input here training

80
00:03:12,200 --> 00:03:16,400
set that so far as a data frame
into a H2O frame,

81
00:03:16,966 --> 00:03:20,666
and therefore this training frame argument
is getting what it is expecting.

82
00:03:21,166 --> 00:03:21,633
Great.

83
00:03:21,633 --> 00:03:23,433
Second argument done come up.

84
00:03:23,433 --> 00:03:25,900
Moving on to the third argument.

85
00:03:25,900 --> 00:03:27,633
So what is the third argument now?

86
00:03:27,633 --> 00:03:31,133
Well, we have other arguments
like model ID but we won't use it.

87
00:03:31,433 --> 00:03:33,000
Overwrite. With best models.

88
00:03:33,000 --> 00:03:34,466
So that's some other options.

89
00:03:34,466 --> 00:03:35,966
But let's not focus on that right now.

90
00:03:35,966 --> 00:03:38,000
That's not the most important.

91
00:03:38,000 --> 00:03:43,200
Then simple validation frame checkpoints
autoencoder pre-trained autoencoder.

92
00:03:43,200 --> 00:03:45,666
So that's again not the most important.

93
00:03:45,666 --> 00:03:48,266
But then moving on to
what's most important.

94
00:03:48,266 --> 00:03:49,700
We get to.

95
00:03:49,700 --> 00:03:52,466
Activation, which of course corresponds

96
00:03:52,466 --> 00:03:55,633
to the activation function
you want to use for your network.

97
00:03:56,066 --> 00:04:00,866
And so that argument is very important
because you saw an intuition

98
00:04:00,866 --> 00:04:04,266
tutorial that some activation functions
are better than others.

99
00:04:04,633 --> 00:04:06,366
And actually, as Kirill explained,

100
00:04:06,366 --> 00:04:10,166
the best activation function to use
to train an artificial neural network

101
00:04:10,333 --> 00:04:13,800
is the rectifier function,
which is this function here on this slide.

102
00:04:14,400 --> 00:04:18,266
And that's the function we'll use
right now for artificial neural network.

103
00:04:18,900 --> 00:04:21,000
All right. So let's input activation.

104
00:04:22,033 --> 00:04:22,933
Equals.

105
00:04:22,933 --> 00:04:25,933
And then in quotes. Rectifier.

106
00:04:26,033 --> 00:04:27,866
Make sure to type capital R.

107
00:04:27,866 --> 00:04:30,400
So perfect
we have one of the best choices of.

108
00:04:30,400 --> 00:04:35,000
Activation functions
for artificial neural network okay great.

109
00:04:35,000 --> 00:04:38,266
And then come up 
moving on to the fourth argument.

110
00:04:38,700 --> 00:04:40,800
So the fourth argument is actually
the one just.

111
00:04:40,800 --> 00:04:42,400
Below. Hidden.

112
00:04:42,400 --> 00:04:44,866
And that's the hidden layer sizes.

113
00:04:44,866 --> 00:04:47,666
That is it's a. Two and one argument.

114
00:04:47,666 --> 00:04:49,500
Because specifying this argument.

115
00:04:49,500 --> 00:04:50,900
Will allow you to specify.

116
00:04:50,900 --> 00:04:53,700
Two parameters of your network
at the same time.

117
00:04:53,700 --> 00:04:56,333
The first parameter
is the number of hidden layers,

118
00:04:56,333 --> 00:04:59,933
and the second parameter is
the number of nodes in each hidden layer.

119
00:05:00,300 --> 00:05:01,700
And so how do we input.

120
00:05:01,700 --> 00:05:02,233
These two.

121
00:05:02,233 --> 00:05:04,866
Parameters in this only one argument here?

122
00:05:04,866 --> 00:05:07,566
Well, we are doing this with a vector.

123
00:05:07,566 --> 00:05:11,533
The number of elements of this vector
will be the number of hidden layers,

124
00:05:11,933 --> 00:05:15,833
and the value of each element of this
vector will be the number of nodes

125
00:05:15,833 --> 00:05:17,233
in the hidden layer.

126
00:05:17,233 --> 00:05:19,733
So for example here we have c.

127
00:05:19,733 --> 00:05:22,566
So you know C
is the way. To define a vector in R.

128
00:05:22,566 --> 00:05:25,066
Then first element of the vector 100

129
00:05:25,066 --> 00:05:27,866
and then second element of the vector
100 again.

130
00:05:27,866 --> 00:05:28,833
So that means that here

131
00:05:28,833 --> 00:05:32,366
we have a vector of two elements
and therefore will have two hidden layers.

132
00:05:32,866 --> 00:05:36,100
And in the first hidden layer
will have 100 neurons,

133
00:05:36,400 --> 00:05:39,300
and in the second hidden layer
will have 100 neurons.

134
00:05:39,300 --> 00:05:40,200
Again.

135
00:05:40,200 --> 00:05:42,966
And now time to ask ourselves
a big question,

136
00:05:42,966 --> 00:05:45,966
a question that comes very often
when building a deep learning model.

137
00:05:46,266 --> 00:05:48,833
It's how many layers do we want to choose?

138
00:05:48,833 --> 00:05:51,133
That is, how many elements in this vector,

139
00:05:51,133 --> 00:05:54,633
and then how many neurons
we want to put in each layer.

140
00:05:55,000 --> 00:05:59,100
So unfortunately, the bad news is that
there is no rule of thumb to.

141
00:05:59,100 --> 00:06:01,133
Choose
this. Optimal number of hidden layers.

142
00:06:01,133 --> 00:06:04,133
And these optimal numbers
of neurons in the hidden layers.

143
00:06:04,266 --> 00:06:06,566
But there is a tip. That we can use.

144
00:06:06,566 --> 00:06:10,400
And this tip is not really based on
research, but rather based on experiments.

145
00:06:10,766 --> 00:06:12,366
We observed that.

146
00:06:12,366 --> 00:06:14,066
A convenient choice. Of.

147
00:06:14,066 --> 00:06:17,900
Number of hidden nodes in the hidden
layers not the optimal choice.

148
00:06:17,900 --> 00:06:20,333
But. More. A convenient choice.

149
00:06:20,333 --> 00:06:21,500
Is to have the average.

150
00:06:21,500 --> 00:06:24,900
Of the number of input nodes
and the number of output nodes.

151
00:06:25,400 --> 00:06:29,533
And as you saw in Carol's intuition
tutorial, well, the number of input nodes

152
00:06:29,533 --> 00:06:30,600
in the input layer.

153
00:06:30,600 --> 00:06:32,866
Is. The number of independent variables.

154
00:06:32,866 --> 00:06:36,000
So the number of input nodes in the input
layer is ten.

155
00:06:36,000 --> 00:06:38,033
Because we have ten independent variables,

156
00:06:38,033 --> 00:06:41,366
and the number of output node
in the output layer is one.

157
00:06:41,666 --> 00:06:44,133
Because again,
you saw in Carol's intuition tutorial,

158
00:06:44,133 --> 00:06:47,033
that's when our dependent variable
has a binary outcome.

159
00:06:47,033 --> 00:06:49,866
Well, there is only one output node
in the output layer.

160
00:06:49,866 --> 00:06:53,333
So that means that the number of neurons
will choose in the hidden layer

161
00:06:53,333 --> 00:06:56,500
is going to be ten
plus one divided by two.

162
00:06:57,166 --> 00:06:58,500
That is 5.5.

163
00:06:58,500 --> 00:07:00,666
But of course
we need to take a round number.

164
00:07:00,666 --> 00:07:02,333
So we'll go for six.

165
00:07:02,333 --> 00:07:04,400
Six neurons in the hidden layer.

166
00:07:04,400 --> 00:07:07,600
And so now to be honest
we are not working with some complex data.

167
00:07:07,600 --> 00:07:08,533
Set like for.

168
00:07:08,533 --> 00:07:11,566
Example images
to find some patterns in the pixels.

169
00:07:11,800 --> 00:07:13,900
Here we have a simple data. Set with.

170
00:07:13,900 --> 00:07:16,800
Some independent variables
and one dependent variable.

171
00:07:16,800 --> 00:07:18,666
And there is no spatial structure in this.

172
00:07:18,666 --> 00:07:21,366
Data set like. It is the case for images.

173
00:07:21,366 --> 00:07:22,000
So to be.

174
00:07:22,000 --> 00:07:24,033
Honest, we don't need many hidden layers.

175
00:07:24,033 --> 00:07:26,600
In fact,
I'm pretty sure our Ann and model.

176
00:07:26,600 --> 00:07:27,600
Would work very well.

177
00:07:27,600 --> 00:07:29,766
With only one hidden layer.

178
00:07:29,766 --> 00:07:32,200
But hey, since we are in the deep
learning part.

179
00:07:32,200 --> 00:07:34,033
Let's go for. Two hidden layers.

180
00:07:34,033 --> 00:07:36,500
That will not be
this kind of deep learning model.

181
00:07:36,500 --> 00:07:38,100
But that will be a start.

182
00:07:38,100 --> 00:07:40,500
All right, so let's input these layers.

183
00:07:40,500 --> 00:07:42,133
With the neurons inside.

184
00:07:42,133 --> 00:07:45,766
So I'm going to add the hidden argument.

185
00:07:46,400 --> 00:07:47,266
Here we go.

186
00:07:47,266 --> 00:07:49,733
And so as you understood we need to.

187
00:07:49,733 --> 00:07:50,666
Specify.

188
00:07:50,666 --> 00:07:53,600
These numbers of layers and these numbers
of neurons in the layers.

189
00:07:53,600 --> 00:07:57,000
With a vector. So c parenthesis.

190
00:07:57,300 --> 00:07:58,800
Then the first element of this vector

191
00:07:58,800 --> 00:08:01,133
is the number of neurons
in the first hidden layer.

192
00:08:01,133 --> 00:08:02,866
So we said six.

193
00:08:02,866 --> 00:08:05,666
Then comma 
and then the second element of this vector

194
00:08:05,666 --> 00:08:08,433
is the number of neurons
in the second hidden layer.

195
00:08:08,433 --> 00:08:10,800
And so let's. Pick six as well.

196
00:08:10,800 --> 00:08:13,633
Keep in mind that in part ten
we will be able to improve.

197
00:08:13,633 --> 00:08:14,600
These choices.

198
00:08:14,600 --> 00:08:17,600
Of parameters
thanks to parameter tuning techniques.

199
00:08:17,600 --> 00:08:19,700
Like k fold cross-validation.

200
00:08:19,700 --> 00:08:20,366
But for now.

201
00:08:20,366 --> 00:08:22,733
Let's just focus. On deep learning.

202
00:08:22,733 --> 00:08:23,300
Okay good.

203
00:08:23,300 --> 00:08:26,400
So that's it for this hidden argument.

204
00:08:26,533 --> 00:08:29,533
And so now let's move
on to the next argument.

205
00:08:29,733 --> 00:08:33,266
The next argument is going to be
the one just below hidden.

206
00:08:33,600 --> 00:08:35,233
That is epochs.

207
00:08:35,233 --> 00:08:36,500
So that is of course.

208
00:08:36,500 --> 00:08:39,833
The number of epochs in the stochastic
gradient descent algorithm.

209
00:08:40,000 --> 00:08:42,333
We actually have a. Good
description. Here.

210
00:08:42,333 --> 00:08:45,133
The number of epochs. Is. How many times.

211
00:08:45,133 --> 00:08:47,100
The data set should be iterated.

212
00:08:47,100 --> 00:08:48,000
If we go back to.

213
00:08:48,000 --> 00:08:51,500
Our intuition, slide about this
stochastic gradient descent algorithm,

214
00:08:51,900 --> 00:08:54,900
well, we can see this number of epochs
here in step seven,

215
00:08:55,033 --> 00:08:58,566
when the whole training set
completed all the steps from one to. Six.

216
00:08:58,766 --> 00:09:00,333
Passing through the A and then

217
00:09:00,333 --> 00:09:03,933
well that makes an epoch
and we redo many more epochs.

218
00:09:04,333 --> 00:09:06,400
So this number of. Epochs here is.

219
00:09:06,400 --> 00:09:10,033
This number of times we are repeating
the whole process of steps one to.

220
00:09:10,033 --> 00:09:11,033
Six for the.

221
00:09:11,033 --> 00:09:13,600
Whole training set,
either by updating the weights.

222
00:09:13,600 --> 00:09:15,033
After each observation.

223
00:09:15,033 --> 00:09:16,000
Or validating the.

224
00:09:16,000 --> 00:09:18,133
Weights after each batch. Of observations.

225
00:09:18,133 --> 00:09:19,900
For example, ten observations.

226
00:09:19,900 --> 00:09:22,900
So let's add this epochs argument here.

227
00:09:23,166 --> 00:09:24,033
Here we go.

228
00:09:24,033 --> 00:09:27,900
And as in Python
we are going to input 100 epochs.

229
00:09:28,700 --> 00:09:29,533
Great.

230
00:09:29,533 --> 00:09:32,400
And now we have one final argument
to input.

231
00:09:32,400 --> 00:09:33,000
That is.

232
00:09:33,000 --> 00:09:34,200
Actually the one just below.

233
00:09:34,200 --> 00:09:37,100
Again train samples per iteration.

234
00:09:37,100 --> 00:09:39,766
So what is that
you might have guessed. What it is.

235
00:09:39,766 --> 00:09:40,433
Well first the.

236
00:09:40,433 --> 00:09:44,566
Description says that it's the number of
training samples per MapReduce iteration.

237
00:09:44,733 --> 00:09:47,733
But more simply that. Is your batch size.

238
00:09:48,100 --> 00:09:51,800
That is the number of observations after
which you want to update the weights.

239
00:09:51,800 --> 00:09:53,533
So this number can be one.

240
00:09:53,533 --> 00:09:55,900
If you want to. Update the weights
after each observation.

241
00:09:55,900 --> 00:09:59,966
Passing through the A and N, and in that
case that's called reinforcement learning.

242
00:09:59,966 --> 00:10:02,966
Exactly as we saw in part six,
reinforcement learning

243
00:10:03,100 --> 00:10:05,000
or this number can be more than one.

244
00:10:05,000 --> 00:10:07,366
That is for example,
when the weights are updated after.

245
00:10:07,366 --> 00:10:07,633
Each.

246
00:10:07,633 --> 00:10:10,533
Batch of ten observations
passing through the end.

247
00:10:10,533 --> 00:10:13,533
And in that case that is called batch
learning.

248
00:10:13,900 --> 00:10:16,200
Okay. So let's input this argument.

249
00:10:16,200 --> 00:10:18,566
Train symbols per iteration.

250
00:10:18,566 --> 00:10:20,400
Here it is pressing enter.

251
00:10:20,400 --> 00:10:23,700
And now the good thing is
that we don't even need to choose.

252
00:10:23,966 --> 00:10:25,333
A batch size.

253
00:10:25,333 --> 00:10:26,433
Because as you can see we.

254
00:10:26,433 --> 00:10:29,566
Have these three parameters zero.

255
00:10:29,766 --> 00:10:31,400
Minus one and minus two.

256
00:10:31,400 --> 00:10:32,700
And as you can. See.

257
00:10:32,700 --> 00:10:33,766
This minus two.

258
00:10:33,766 --> 00:10:35,966
Value is very practical because by.

259
00:10:35,966 --> 00:10:41,366
Specifying minus two here, this will
autotune your artificial neural network.

260
00:10:41,700 --> 00:10:44,933
And that corresponds to the third reason
why. H2O.

261
00:10:44,966 --> 00:10:47,366
Was to me one of the. Best packages.

262
00:10:47,366 --> 00:10:48,533
It's because it gives you the.

263
00:10:48,533 --> 00:10:51,300
Option to already apply
some parameter tuning.

264
00:10:51,300 --> 00:10:52,900
Parameter tuning can be way more.

265
00:10:52,900 --> 00:10:56,500
Complicated, but here we have
this great tool that will already help.

266
00:10:56,500 --> 00:10:58,600
Us to improve the model.

267
00:10:58,600 --> 00:11:01,633
So let's definitely input minus two here.

268
00:11:02,033 --> 00:11:06,333
And actually now
artificial neural network is ready.

269
00:11:07,200 --> 00:11:09,933
That's all the parameters
we need. To build.

270
00:11:09,933 --> 00:11:11,833
Our artificial neural network.

271
00:11:11,833 --> 00:11:13,966
And we even have some parameter tuning.

272
00:11:13,966 --> 00:11:16,666
So we have. Everything we need
and even more than we need.

273
00:11:16,666 --> 00:11:18,266
Therefore let's do it.

274
00:11:18,266 --> 00:11:18,733
I'm going.

275
00:11:18,733 --> 00:11:21,733
To select all. These lines here.

276
00:11:21,766 --> 00:11:24,166
That will create the. Classifier.

277
00:11:24,166 --> 00:11:26,400
And now I just want to say something
important for.

278
00:11:26,400 --> 00:11:28,633
Those of you
who follow the. Python tutorials.

279
00:11:28,633 --> 00:11:30,900
Well you saw that. Training
DNN took quite.

280
00:11:30,900 --> 00:11:31,700
A while.

281
00:11:31,700 --> 00:11:32,366
It took about.

282
00:11:32,366 --> 00:11:34,966
One minute,
if I remember correctly, and that was.

283
00:11:34,966 --> 00:11:37,700
Because we used our CPU on our system.

284
00:11:37,700 --> 00:11:41,400
But here we are connected to a server,
and so we have access to a powerful.

285
00:11:41,400 --> 00:11:43,800
System using this. H2O. Instance.

286
00:11:43,800 --> 00:11:45,600
And therefore. You will see how.

287
00:11:45,600 --> 00:11:48,600
The model is going to be trained
much faster.

288
00:11:48,633 --> 00:11:51,500
We are going to see that right now
I'm going to execute.

289
00:11:51,500 --> 00:11:54,433
Ready. Three two. One. Go.

290
00:11:57,533 --> 00:11:58,466
And that's it.

291
00:11:58,466 --> 00:11:59,866
That's already ready.

292
00:11:59,866 --> 00:12:02,400
It took about five seconds.

293
00:12:02,400 --> 00:12:03,700
So that's pretty exciting.

294
00:12:03,700 --> 00:12:05,700
To work. With such powerful tools.

295
00:12:05,700 --> 00:12:07,466
I was very happy to show you this.

296
00:12:07,466 --> 00:12:09,300
And I hope that now you're convinced.

297
00:12:09,300 --> 00:12:11,833
That with this H2O package you are.

298
00:12:11,833 --> 00:12:13,666
In good hands. So great.

299
00:12:13,666 --> 00:12:15,366
And now that our model is trained,

300
00:12:15,366 --> 00:12:18,833
time for the next step
making predictions on the test set.

301
00:12:19,400 --> 00:12:21,200
And that will be very interesting to do.

302
00:12:21,200 --> 00:12:23,133
To see the accuracy on new.

303
00:12:23,133 --> 00:12:25,833
Observations
on which the model wasn't trained.

304
00:12:25,833 --> 00:12:27,633
So we'll see that in the next tutorial.

305
00:12:27,633 --> 00:12:30,633
And until then, enjoy machine learning.
