1
00:00:00,330 --> 00:00:00,930
All right.

2
00:00:00,960 --> 00:00:05,080
So in this lesson we will load our pre trained model.

3
00:00:05,100 --> 00:00:09,840
This is gonna be the same model that we've saved in the previous lesson and we're now loaded into a

4
00:00:09,840 --> 00:00:16,710
fresh notebook and use that model to make predictions on the amnesty test dataset.

5
00:00:16,710 --> 00:00:24,060
If you head over to the lesson resources you should see a downloadable zip file and if you extract the

6
00:00:24,060 --> 00:00:30,430
contents of this zip file you should see this Jupiter notebook right here which is going to be a stub

7
00:00:30,450 --> 00:00:35,310
project and you should see to see as V files with some test data.

8
00:00:35,310 --> 00:00:42,330
So after you've downloaded and extracted these files take these test files to see as V files and drag

9
00:00:42,330 --> 00:00:46,950
them over into your amnesty folder just Adam right here.

10
00:00:46,950 --> 00:00:53,280
Now this amnesty folder is a folder that you would have created for the previous module but if you don't

11
00:00:53,280 --> 00:00:58,570
have it then just create a new folder and call it amnesty in all caps.

12
00:00:58,680 --> 00:01:03,540
Once you've done that drag the stop project over into your projects folder.

13
00:01:03,540 --> 00:01:08,880
Now you can go back into Jupiter notebook and you should see the stop project that I've given you right

14
00:01:08,880 --> 00:01:09,180
here.

15
00:01:10,690 --> 00:01:18,120
So just click on it and I'll open in a new window and what you should see here are a couple of markdown

16
00:01:18,130 --> 00:01:23,470
cells that delineate the different sections of the stop project and I find a little bit of code here

17
00:01:23,470 --> 00:01:26,350
already to help us load the test data.

18
00:01:26,650 --> 00:01:32,680
We've used num pi extensively already to load the text files so I've just saved you the work and added

19
00:01:32,680 --> 00:01:33,880
this code here.

20
00:01:33,910 --> 00:01:39,340
The important thing to note about the stop project is that it expects to find these load underscore

21
00:01:39,400 --> 00:01:47,380
x test see as V and load underscore why test dot CSP files in the end this sub folder and it has to

22
00:01:47,380 --> 00:01:49,350
be spelled exactly like this.

23
00:01:49,390 --> 00:01:54,040
And the file names also have to match this exactly.

24
00:01:54,040 --> 00:01:57,340
Now that we're all set up how shall we get started.

25
00:01:57,340 --> 00:02:01,460
Well to do anything useful in tensor flow we need two things.

26
00:02:01,510 --> 00:02:04,040
First we need a graph.

27
00:02:04,060 --> 00:02:07,000
And second we need a session.

28
00:02:07,000 --> 00:02:09,760
The graph is something like this.

29
00:02:09,760 --> 00:02:16,420
It's what holds onto all our tensor is in all the operations and represents how the data flows from

30
00:02:16,420 --> 00:02:23,600
the inputs to the outputs so let's create a blank graph in our Jupiter notebook.

31
00:02:23,600 --> 00:02:26,840
So right here we're just going to create a variable called Graph.

32
00:02:26,880 --> 00:02:34,150
Although our case and we're going to set it equal to T F dot graph with some parentheses at the end.

33
00:02:34,370 --> 00:02:40,400
And what this will do is it will create a placeholder for us a graph object placeholder.

34
00:02:41,060 --> 00:02:46,880
And this at the moment is empty but soon it's going to hold onto the graph that we will load from our

35
00:02:46,880 --> 00:02:47,780
model.

36
00:02:47,780 --> 00:02:53,900
The next thing that we need is a session object and this session if you recall is what actually runs

37
00:02:53,900 --> 00:02:56,090
our operations and tensor flow.

38
00:02:56,090 --> 00:03:00,230
We can only really do calculations intensive flow when we run a session.

39
00:03:00,230 --> 00:03:01,400
So let's create a session.

40
00:03:01,530 --> 00:03:04,230
I'm going to call it says an all lowercase.

41
00:03:04,550 --> 00:03:10,970
And once again it's TAF dot session with parentheses at the end.

42
00:03:10,970 --> 00:03:17,210
Now this will create a session object but we actually should link this session to our graph.

43
00:03:17,390 --> 00:03:18,120
How do we do that.

44
00:03:18,980 --> 00:03:21,500
Well let's look at the documentation.

45
00:03:21,500 --> 00:03:28,370
If you search for tensor flow session then you should be able to go right here to this link and of course

46
00:03:28,370 --> 00:03:35,720
I want to put all of this into the lesson resources when you scroll down you'll see this section labeled

47
00:03:35,870 --> 00:03:36,670
in it.

48
00:03:36,770 --> 00:03:44,120
This is the code that will allow us to actually create the session object and then we see as the second

49
00:03:44,150 --> 00:03:47,490
argument we can supply a graph by default.

50
00:03:47,570 --> 00:03:52,090
This graph has the value none so it is indeed an optional parameter.

51
00:03:52,310 --> 00:03:59,540
But here is how we can link the graph object that we've created to the session object so let's modify

52
00:03:59,690 --> 00:04:07,640
this line of code right here so that it will read graph is equal to graph between the parentheses.

53
00:04:07,640 --> 00:04:09,740
So this is the name of the parameter.

54
00:04:10,130 --> 00:04:12,610
And here is the actual value.

55
00:04:12,650 --> 00:04:15,160
This object right here that we're passing in.

56
00:04:15,470 --> 00:04:16,190
Brilliant.

57
00:04:16,190 --> 00:04:19,250
So now we have a session and a graph and the two are linked.

58
00:04:19,820 --> 00:04:23,450
So the next step is actually loading our saved model.

59
00:04:23,570 --> 00:04:29,340
So how do we load a model well let's pull up the documentation again for tensor flow.

60
00:04:29,540 --> 00:04:35,700
If you google something like load model tensor flow then you should be taking to this page right here.

61
00:04:36,020 --> 00:04:37,800
If we look at this section right here.

62
00:04:37,850 --> 00:04:40,820
Loading a saved model in Python.

63
00:04:41,030 --> 00:04:44,240
Then we see a little excerpt with this bit of code.

64
00:04:44,480 --> 00:04:50,870
There we see that there's something called a load function and it seems to require three arguments a

65
00:04:50,870 --> 00:04:56,780
session some sort of tag and an export directory.

66
00:04:56,780 --> 00:05:00,810
Let's click on this link right here and pull up a more detailed explanation.

67
00:05:00,830 --> 00:05:05,540
So when I click on it and navigate to this load function right here and there what I see is this full

68
00:05:05,540 --> 00:05:07,640
name of the load function.

69
00:05:08,030 --> 00:05:13,580
But just like in the last lesson with simple Save there were some aliases that make it compatible with

70
00:05:13,580 --> 00:05:15,980
future versions of tensor flow.

71
00:05:15,980 --> 00:05:21,500
So in order to get our code to work with the upcoming tensor flow version 2.0 we're going to use this

72
00:05:21,590 --> 00:05:23,120
full name right here.

73
00:05:23,120 --> 00:05:30,680
TMF dot com Pat dot v one dot saved underscore model dot load so let's import this load function in

74
00:05:30,680 --> 00:05:32,150
our Jupiter notebook.

75
00:05:32,150 --> 00:05:38,900
I want to head back in here and at the very very top I'm going to import the load function and that

76
00:05:38,900 --> 00:05:49,880
means importing it from tensor flow dot compact dot V1 dot saved underscore model and from there we

77
00:05:49,880 --> 00:05:52,190
can import load.

78
00:05:52,190 --> 00:05:58,340
Now what I can do is maybe run all the cells in my notebook after you've run all the cells.

79
00:05:58,460 --> 00:06:00,420
You can call your load function.

80
00:06:00,560 --> 00:06:02,090
So I've made a little sexual note here.

81
00:06:02,090 --> 00:06:08,770
Load the model and I've pasted in a little bit of the documentation as we can see in this example write

82
00:06:08,780 --> 00:06:10,910
him we need three things.

83
00:06:10,910 --> 00:06:13,260
The first is that session object.

84
00:06:13,610 --> 00:06:14,910
So let's add that right here.

85
00:06:15,020 --> 00:06:21,800
We'll say load open parentheses and then for the session we'll supply our session object.

86
00:06:22,060 --> 00:06:26,420
The session object that we've created right here in this cell.

87
00:06:26,420 --> 00:06:29,720
The second thing that we need to provide is some sort of tag.

88
00:06:29,930 --> 00:06:36,230
In the example in the documentation they've provided this tag underscore constants dot training.

89
00:06:36,230 --> 00:06:39,130
Now we know that we're not gonna be looking to train our model.

90
00:06:39,380 --> 00:06:42,120
We're looking to serve our model instead.

91
00:06:42,140 --> 00:06:43,550
So what's the tag for serving.

92
00:06:44,090 --> 00:06:51,110
Well actually the really neat thing about having used these simple save function in the last lesson

93
00:06:51,650 --> 00:06:58,700
is that it automatically adds a tag to the model and we can see here that it automatically uses the

94
00:06:58,700 --> 00:07:07,760
serving tag which is exactly what we wanted the easiest way to get hold of this serving tag in our notebook

95
00:07:07,760 --> 00:07:14,220
here is to import the tag constants so much scroll back up here where we're importing load.

96
00:07:14,480 --> 00:07:22,340
And I'm going to tack on tag underscore constants to the import limit shift enter on the cell and then

97
00:07:22,340 --> 00:07:28,910
I can come down here where I'm calling my load function and for the tags I'm going to simply set them

98
00:07:28,910 --> 00:07:34,410
equal to tag underscore constants dot serving.

99
00:07:34,460 --> 00:07:40,880
So we're going exactly with the same format as with the documentation right here.

100
00:07:41,000 --> 00:07:44,390
The last thing we need to provide is this export directory.

101
00:07:44,390 --> 00:07:49,460
If we go to the detailed documentation for this load function we can see that in the description the

102
00:07:49,490 --> 00:07:55,280
export directory is simply the directory in which the saved model protocol buffer and variables to be

103
00:07:55,280 --> 00:07:57,260
loaded are located.

104
00:07:57,260 --> 00:08:02,840
All that means is that the load function needs to know where to find this PDB file and these variable

105
00:08:02,840 --> 00:08:03,800
files.

106
00:08:03,860 --> 00:08:09,740
So we created a sub director here called saved model where all these files were saved and that means

107
00:08:09,740 --> 00:08:16,190
for the third argument in the load function we're simply going to say export underscored Diaw is equal

108
00:08:16,190 --> 00:08:25,220
to single quotes saved model and all we have to do is spell that folder name exactly the same way as

109
00:08:25,220 --> 00:08:29,300
we've gone in Windows Explorer or in Mac finder.

110
00:08:29,330 --> 00:08:31,790
Now let's that shift enter on the cell and see if it works.

111
00:08:33,860 --> 00:08:35,070
Brilliant.

112
00:08:35,390 --> 00:08:40,330
So now that we've successfully loaded our model we can use it to make predictions.

113
00:08:40,610 --> 00:08:41,690
But how do we do that.

114
00:08:41,720 --> 00:08:47,870
How do we make a prediction using this model that we've just loaded for that we need to remember.

115
00:08:47,870 --> 00:08:51,330
What are the inputs that we actually need to make a prediction.

116
00:08:51,470 --> 00:08:54,490
And what are the outputs of a prediction.

117
00:08:54,500 --> 00:08:57,450
We've talked a little bit about this in the previous lesson.

118
00:08:57,590 --> 00:08:59,920
The inputs where our exits.

119
00:08:59,990 --> 00:09:07,790
So I've got the graph for a model here and we can see here that the features namely the pixel values

120
00:09:07,880 --> 00:09:13,650
of the image that we're going to predict are gonna be stored in this tensor right here.

121
00:09:13,700 --> 00:09:21,170
We call this place holder capital X. So let's retrieve this tensor from the graph and we can do that

122
00:09:21,410 --> 00:09:28,460
by saying capital X is equal to some creating a new Python variable here and then I'm going to store

123
00:09:28,880 --> 00:09:31,560
a tensor that I'm going to grab from the graph.

124
00:09:31,610 --> 00:09:40,580
So when I take our graph object and then use the function get tensor by name and within the parentheses

125
00:09:40,660 --> 00:09:46,200
mean to supply single quotes and then I'm going to have to supply the name of the tensor.

126
00:09:46,310 --> 00:09:54,380
And in this case it's capital X because that's what we've got write him but then I have to add a colon

127
00:09:54,860 --> 00:09:56,300
and a zero.

128
00:09:56,390 --> 00:10:00,460
Now that's the full name of the tensor that I might get from the graph.

129
00:10:00,670 --> 00:10:04,640
Now this point you might ask why is there this colon and a zero at the end.

130
00:10:04,640 --> 00:10:05,790
Why is this tacked on.

131
00:10:05,930 --> 00:10:07,550
Because we don't see it intensively right.

132
00:10:08,690 --> 00:10:13,160
Well the answer is that this has something to do with how tensor flow works.

133
00:10:13,280 --> 00:10:21,980
If we look at 10 support what we actually see are a whole bunch of operations so labels is an operation

134
00:10:23,110 --> 00:10:29,930
and if I expand my first layer then clicking on any of these things we see that these are all operations

135
00:10:30,680 --> 00:10:38,540
and all of these operations are linked with arrows and the arrows are the actual tensor is the thing

136
00:10:38,540 --> 00:10:44,240
to understand about tensor flow is that all of these operations have inputs and outputs.

137
00:10:44,300 --> 00:10:50,960
Take this ad operation here in our first layer here we're simply just adding the biases to the weights

138
00:10:51,530 --> 00:10:55,640
and what we get from this addition is a single output the sum.

139
00:10:55,700 --> 00:11:00,220
Now say I want to access the sum of the weights and the biases in the first layer.

140
00:11:00,680 --> 00:11:02,270
How would I do that.

141
00:11:02,420 --> 00:11:05,850
What I would do is I would take the name of the operation.

142
00:11:05,900 --> 00:11:13,940
So this is layer underscore one slash add and then if I wanted to get a hold of the sum in this line

143
00:11:13,940 --> 00:11:20,180
of code here I would simply have graph don't get tensor by name and then in the single quotes I would

144
00:11:20,180 --> 00:11:27,860
have layer underscore one slash add there's only one little twist again I'm going I have to add a colon

145
00:11:28,250 --> 00:11:29,090
and add zero.

146
00:11:29,090 --> 00:11:33,440
This will give me access to the actual sum of the addition.

147
00:11:33,740 --> 00:11:38,840
So this will give me access to the values that are the result of adding two things together.

148
00:11:38,930 --> 00:11:46,640
If I leave out this colon on zero then I get access to the operation itself to the ADD operation on

149
00:11:46,640 --> 00:11:49,860
the other hand if I have the suffix colon zero.

150
00:11:50,060 --> 00:11:51,350
Then I get access to the sum.

151
00:11:52,040 --> 00:11:55,340
So this is a little bit of a quirk of tensor flow.

152
00:11:55,340 --> 00:12:00,890
By adding this colon zero to an operation we get access to an end point of an operation.

153
00:12:00,890 --> 00:12:06,350
So in this case this add operation and the reason we have colon zero is because there might be more

154
00:12:06,350 --> 00:12:08,950
than one output in the case of addition of course.

155
00:12:08,990 --> 00:12:13,430
If you add two things you only have one output but if you had some other operation that has more than

156
00:12:13,430 --> 00:12:15,600
one output then these things are indexed.

157
00:12:15,620 --> 00:12:17,280
The first one is at index 0.

158
00:12:17,300 --> 00:12:21,900
The second one is in index 1 The third one is set in next to and so on.

159
00:12:21,920 --> 00:12:25,710
So when it changes back to X colon 0.

160
00:12:25,910 --> 00:12:29,800
So this is how we're going to get hold of our inputs to make our predictions.

161
00:12:29,870 --> 00:12:36,080
So the next question is how do we actually access the predictions from our model and actually want to

162
00:12:36,080 --> 00:12:38,220
put this over to you as a challenge.

163
00:12:38,330 --> 00:12:44,870
Can you get hold of the tensor that will hold the models predictions from the graph store these predictions

164
00:12:45,320 --> 00:12:48,440
in a variable called Y underscore pred.

165
00:12:48,710 --> 00:12:57,180
I'll give you a few seconds to pause the video before I show you the solution.

166
00:12:57,290 --> 00:12:58,510
Ready.

167
00:12:58,520 --> 00:13:00,410
Here's the solution.

168
00:13:00,410 --> 00:13:05,110
Now if you recall the output of our model was this operation right here.

169
00:13:05,120 --> 00:13:08,890
It had the name accuracy underscore calc slash prediction.

170
00:13:08,900 --> 00:13:13,850
So what I want to do is I'm going to copy this name right here and I'm going to create this variable

171
00:13:13,880 --> 00:13:18,130
by underscore pred and I want to use the same code as above graph.

172
00:13:18,140 --> 00:13:25,040
Don't get tensor by name and then as an argument I'm going to provide in single quotes the name of the

173
00:13:25,040 --> 00:13:28,200
operation and then a colon and then a zero.

174
00:13:28,220 --> 00:13:34,600
This line of code will store the tensor that holds the actual predictions in this variable right here.

175
00:13:35,000 --> 00:13:41,600
So now we have everything we need the inputs and the outputs to make predictions using our pre trained

176
00:13:41,600 --> 00:13:47,910
model so the next step is making predictions on our testing dataset.

177
00:13:47,990 --> 00:13:49,510
How do we do that.

178
00:13:49,520 --> 00:13:54,560
Well any time we actually want to use tensor float to make a calculation or give us an output of some

179
00:13:54,560 --> 00:13:55,220
sort.

180
00:13:55,220 --> 00:14:02,270
Then we have to run a session and we can run our session by writing says Dot run.

181
00:14:02,380 --> 00:14:06,280
This takes our session object and we'll call the run functional.

182
00:14:07,130 --> 00:14:09,100
And it will give us some sort of output usually right.

183
00:14:09,370 --> 00:14:17,320
So I'm going to store this output in a python variable called prediction and now I have to provide some

184
00:14:17,320 --> 00:14:19,170
arguments to this run function.

185
00:14:19,180 --> 00:14:22,660
Let's take a look at the documentation again for this session.

186
00:14:22,660 --> 00:14:26,490
Here we've got the run function and it's got a number of arguments that it can take.

187
00:14:26,500 --> 00:14:32,320
Now I'm not interested in the options and the metadata I'm only interested in these two the fetches

188
00:14:32,770 --> 00:14:34,130
and the feed dictionary.

189
00:14:34,360 --> 00:14:36,200
So what are the fetches in the feed dictionary.

190
00:14:36,910 --> 00:14:41,570
Well the fetches are the values that we want to fetch from the graph.

191
00:14:41,740 --> 00:14:46,010
Here we can specify the kind of output that we want from the session.

192
00:14:46,240 --> 00:14:48,240
In our case we want the predictions.

193
00:14:48,250 --> 00:14:49,000
Right.

194
00:14:49,420 --> 00:14:51,440
Now what about the feed dictionary.

195
00:14:51,640 --> 00:14:55,810
Well the feed dictionary is the data that the prediction is based on.

196
00:14:55,810 --> 00:14:59,800
So once again I'd like to pose this over to you as a challenge.

197
00:14:59,800 --> 00:15:07,180
Can you figure out how to call this wrong function and supply the values to the fetches and the feed

198
00:15:07,180 --> 00:15:10,620
dictionary that will give us the predictions for our model.

199
00:15:10,660 --> 00:15:17,220
Remember the goal is to predict the handwritten digits that are stored inside X on the score test.

200
00:15:17,350 --> 00:15:21,550
I'll give you a few seconds to pause the video before I show you the solution

201
00:15:24,920 --> 00:15:25,870
ready.

202
00:15:25,880 --> 00:15:27,510
Here's the solution.

203
00:15:27,770 --> 00:15:35,150
So for the finches I want to set those equal to while the output that we're after we're after Y underscore

204
00:15:35,570 --> 00:15:43,810
pret the output right here and for the feed dictionary we're going to have to provide X on the score

205
00:15:43,810 --> 00:15:45,220
test.

206
00:15:45,220 --> 00:15:52,060
So this has to be provided in the form of a python dictionary with a name and a value.

207
00:15:52,510 --> 00:15:54,540
The name is gonna be capital X..

208
00:15:54,610 --> 00:15:59,620
And the value is gonna be X on a score test and that's it.

209
00:15:59,620 --> 00:16:05,140
So let's execute these cells and check if it worked first then when execute this cell right here.

210
00:16:05,860 --> 00:16:11,170
And second I want to execute this cell right here where I'm running my session.

211
00:16:11,170 --> 00:16:18,250
Let's see what the actual values are for the first couple of labels in the y underscored test data set.

212
00:16:18,670 --> 00:16:23,470
So why underscore test and then it's square brackets.

213
00:16:23,480 --> 00:16:28,180
Colon five so that we see that the first digit is seven.

214
00:16:28,180 --> 00:16:29,580
The second is a two.

215
00:16:29,580 --> 00:16:30,540
The third one is a one.

216
00:16:30,970 --> 00:16:34,810
And so on now let's take a look at the predictions.

217
00:16:34,980 --> 00:16:40,200
The predictions that are the result of running our session we can pull up like so just like the cell

218
00:16:40,200 --> 00:16:44,070
above and we can see that the output seems to match.

219
00:16:45,510 --> 00:16:47,550
And that's really encouraging right.

220
00:16:47,580 --> 00:16:53,970
This would suggest that we've successfully loaded our model and we've successfully made some predictions

221
00:16:54,390 --> 00:16:57,900
on X underscore test that match the correct values.

222
00:16:58,230 --> 00:17:02,750
But just because the first five results match doesn't mean that they all match.

223
00:17:02,760 --> 00:17:03,060
Right.

224
00:17:03,840 --> 00:17:10,230
So as a challenge can you calculate what percent of the test data set we predicted correctly in this

225
00:17:10,230 --> 00:17:11,410
notebook.

226
00:17:11,430 --> 00:17:16,350
There's a number of ways you can do this but I'll give you a few seconds to pause the video before I

227
00:17:16,350 --> 00:17:19,140
show you a sample solution.

228
00:17:19,140 --> 00:17:19,440
All right.

229
00:17:19,470 --> 00:17:21,480
So here's how I would do it.

230
00:17:21,690 --> 00:17:27,620
I would store the number of correct predictions in a variable something like an hour on a score.

231
00:17:27,840 --> 00:17:28,590
Correct.

232
00:17:28,680 --> 00:17:35,300
And I'm going to set this equal to while I want a check for quality using num PIs equal function.

233
00:17:35,940 --> 00:17:43,190
And here I can check whether two things are equal namely the prediction and why on a score test.

234
00:17:43,260 --> 00:17:47,220
So let's have a look at what this looks like here.

235
00:17:47,260 --> 00:17:54,050
I get an array of billions so each entry that's equal to true has a value equal to one and each entry.

236
00:17:54,080 --> 00:17:57,730
That's equal to False has a value equal to zero.

237
00:17:57,750 --> 00:18:01,020
So what I can do then it's actually just some them all up.

238
00:18:01,020 --> 00:18:01,670
Right.

239
00:18:01,680 --> 00:18:05,280
This will give us the number of correct predictions.

240
00:18:05,490 --> 00:18:12,420
And in order to calculate the percentage what I need to do is divide the total number of correct predictions

241
00:18:12,720 --> 00:18:15,140
by the number of entries.

242
00:18:15,210 --> 00:18:23,890
So in our case we've got 10000 entries in this dataset so I could either divide it by ten thousand or

243
00:18:24,160 --> 00:18:31,870
I could just divided by the length of this array which is alien parentheses prediction and what I get

244
00:18:31,900 --> 00:18:35,980
in this case is ninety seven point eight six percent.

245
00:18:35,980 --> 00:18:42,720
And that's it there's just one last thing to do and that's a little bit of cleanup because at the moment

246
00:18:43,020 --> 00:18:44,870
our tensor flow session is actually still running.

247
00:18:44,910 --> 00:18:46,080
It's still open.

248
00:18:46,080 --> 00:18:53,730
So we can take our session object and close it and we can take our graph and just reset the default

249
00:18:53,730 --> 00:18:57,330
graph and that's it all done at this point.

250
00:18:57,330 --> 00:19:04,740
We've successfully saved and loaded a saved model using tensor flow and we've also verified that our

251
00:19:04,740 --> 00:19:08,530
saved model still works and can make accurate predictions.

252
00:19:08,610 --> 00:19:14,370
The next step is converting our tensor flow model into something that is able to run in the browser

253
00:19:15,120 --> 00:19:18,720
and for that we need to convert our tensor flow model into tensor flow.

254
00:19:18,790 --> 00:19:19,610
Yes.

255
00:19:19,890 --> 00:19:23,210
And that's exactly what we're going to do in the next lesson.

256
00:19:23,220 --> 00:19:24,260
I'll see you there.

257
00:19:24,270 --> 00:19:24,780
Take care.