1
00:00:00,330 --> 00:00:04,110
Now, let's take a look at defining or lost function and optimism.

2
00:00:04,590 --> 00:00:08,910
Remember, these are integral components of your CNN training process.

3
00:00:09,360 --> 00:00:13,500
Remember, we remember the lost function that we talked about cross entropy loss.

4
00:00:14,040 --> 00:00:14,910
Well, remember what?

5
00:00:14,910 --> 00:00:22,170
We even use loss for losses use as basically a unit metric measurement that allows the back propagation

6
00:00:22,500 --> 00:00:27,540
to use that as an input so that we can actually calculate the weights we can back, propagate through

7
00:00:27,540 --> 00:00:31,770
the network, calculate whether the weights need to be increased or decreased for all the nodes.

8
00:00:32,220 --> 00:00:37,290
And that process, remember, is called the gradient optimizing optimization process.

9
00:00:37,830 --> 00:00:41,520
And there wasn't where we use algorithms such as stochastic gradient descent.

10
00:00:42,030 --> 00:00:46,590
And remember, we had a parameter lambda as a learning rate parameter which where we use it as zero

11
00:00:46,590 --> 00:00:47,690
point zero zero one.

12
00:00:47,700 --> 00:00:53,610
In this case, in this situation and we use a momentum, remember I recommended a value of zero point

13
00:00:53,610 --> 00:00:54,010
nine.

14
00:00:54,130 --> 00:00:59,040
This bracket shouldn't be the bracket of a value of zero point nine.

15
00:00:59,340 --> 00:01:02,010
That's what we'll be using for momentum value.

16
00:01:02,400 --> 00:01:04,670
So this is how we create a optimizer.

17
00:01:05,230 --> 00:01:10,920
First, we import the optimized object here, so it's torched on Optim and we import it as optimal so

18
00:01:10,920 --> 00:01:12,750
we can use it just like this in the code.

19
00:01:13,380 --> 00:01:21,950
So we specify we want to use the S2 function, which is optimizer belonging to the optimal module or

20
00:01:21,960 --> 00:01:22,410
object.

21
00:01:23,220 --> 00:01:29,420
We use this network parameters here, thus specifying that that we created.

22
00:01:30,000 --> 00:01:32,430
We're going to set the parameters to it.

23
00:01:32,520 --> 00:01:39,750
Being with with this Study Optimizer specified a learning rate of momentum and you would have noticed

24
00:01:39,750 --> 00:01:40,500
I skipped over this.

25
00:01:40,500 --> 00:01:43,710
But this criterion here is just across entropy loss.

26
00:01:43,710 --> 00:01:44,730
We just create this.

27
00:01:44,790 --> 00:01:48,630
I will be using this later on in the code, which you will see shortly.

28
00:01:49,770 --> 00:01:52,230
So that's about it for the optimization function.

29
00:01:52,240 --> 00:01:56,190
Just to some bullet, you want to come this trees and power lines of code.

30
00:01:57,150 --> 00:02:00,090
Now let's take a look at how we train on model.

31
00:02:00,660 --> 00:02:06,360
This is one of the most important parts of the CNN so senior training process.

32
00:02:06,870 --> 00:02:12,960
So I'll stop here for now, and I'll go over this section with you in detail in the next section.

33
00:02:13,110 --> 00:02:13,500
Thank you.