1
00:00:00,120 --> 00:00:01,590
So yeah, we have this killers.

2
00:00:01,590 --> 00:00:09,600
And then if you click on this graphs here or have this graph coming up, let's reduce this slightly.

3
00:00:09,600 --> 00:00:10,000
Okay?

4
00:00:10,050 --> 00:00:16,230
So we have this notice that the graph type here is up graph and this actually means we're viewing the

5
00:00:16,230 --> 00:00:19,250
graph from the operations level.

6
00:00:19,260 --> 00:00:26,040
If you could come right here and zoom in, you'll see you have the cough layer, you have the comfort

7
00:00:26,280 --> 00:00:29,640
layer again, dense layer, dense layer.

8
00:00:29,640 --> 00:00:32,820
And then zoom in this other way here.

9
00:00:32,820 --> 00:00:35,010
You see you have the albums optimizer.

10
00:00:35,010 --> 00:00:39,720
Let's double click your where you have this, plus you double click and then you get to see exactly

11
00:00:39,720 --> 00:00:43,830
what goes on in this Adam Optimizer.

12
00:00:43,830 --> 00:00:50,580
Let's reduce that zooming and then let's double click again to reduce this.

13
00:00:50,610 --> 00:00:51,030
Okay?

14
00:00:51,090 --> 00:00:52,770
So you have that.

15
00:00:52,770 --> 00:00:55,620
And for this dense layer, you could double click.

16
00:00:55,650 --> 00:01:01,680
You see, you have that, you see you have this can now you double click to better understand what goes

17
00:01:01,680 --> 00:01:02,220
on.

18
00:01:02,730 --> 00:01:05,610
And as you could see here, we have this regular riser.

19
00:01:05,610 --> 00:01:08,370
So the regular riser we had defined previously.

20
00:01:08,400 --> 00:01:11,790
Now let's reduce this by double clicking.

21
00:01:11,790 --> 00:01:17,460
Let's scroll here, we double click that reduced and then double click this.

22
00:01:17,460 --> 00:01:19,740
So basically, it hangs 2/10 of a bar.

23
00:01:19,740 --> 00:01:27,780
We're able to visualize exactly what goes on under the hood when TensorFlow creates this graphs, which

24
00:01:27,780 --> 00:01:31,230
in turn parameters do computations even faster.

25
00:01:31,380 --> 00:01:36,480
Now, another way we can look at this is by coming right here, the stack and selecting carrots.

26
00:01:36,480 --> 00:01:42,510
Once we select the carrots, we instead of having the operation graph we have just seen, we now have

27
00:01:42,510 --> 00:01:44,040
this conceptual graph.

28
00:01:44,040 --> 00:01:49,110
We'll see that this is going to be quite easy compared to or easy to understand compared to what we

29
00:01:49,110 --> 00:01:50,220
had done previously.

30
00:01:50,220 --> 00:01:53,610
So you have this focuses on the cross model.

31
00:01:53,610 --> 00:01:57,330
We have built a cross sequential model we had specified.

32
00:01:57,330 --> 00:02:00,240
And here what we have is you see the input.

33
00:02:00,240 --> 00:02:06,660
So unlike previously where we had stuff like the Atom Optimizer, the different metrics and the say

34
00:02:06,660 --> 00:02:12,630
loss computations with the up graph here we have just the Keras model.

35
00:02:12,630 --> 00:02:20,610
So here you have the input conf batch normalization max pulling drop out conf batch normalization max

36
00:02:20,610 --> 00:02:26,040
pulling flatten dense batch norm dropout dense batch norm and finally dense.

37
00:02:26,040 --> 00:02:34,110
So that's our conceptual graph and that's it for the section on graphs you see, you get to understand

38
00:02:34,110 --> 00:02:40,950
exactly what goes on under the hood thanks to this visualization.

39
00:02:41,670 --> 00:02:45,990
Or better still, this graphs visualization made available with tensor board.