1
00:00:02,200 --> 00:00:04,630
So now our training is almost complete.

2
00:00:07,330 --> 00:00:15,220
You can see on the 1980 book, we were getting a validation, accuracy of points and forward and a training

3
00:00:15,310 --> 00:00:17,230
accuracy of mine to do the same.

4
00:00:20,560 --> 00:00:27,700
And in the last epoch, that is the 20th epoch, we are getting the validation accuracy of 73 percent.

5
00:00:28,720 --> 00:00:35,410
Now, let's just blot all this accuracy values on the graph.

6
00:00:42,260 --> 00:00:51,510
Now, the orange line here is our training accuracy and the red line is validation, accuracy and green

7
00:00:51,510 --> 00:00:58,890
and blue lines are validation, loss and training loss respectively.

8
00:01:03,270 --> 00:01:09,480
One thing here is to notice that there is a large difference between relevation accuracy and the training

9
00:01:09,480 --> 00:01:10,110
accuracy.

10
00:01:12,450 --> 00:01:17,400
Validation accuracy is also lurking around 73 to 74 percent.

11
00:01:18,660 --> 00:01:22,770
Whereas training accuracy is increasing with each epoch.

12
00:01:23,220 --> 00:01:28,350
Currently it is 93 percent and it is increasing.

13
00:01:30,150 --> 00:01:32,960
This graph suggests that that isn't what we're doing.

14
00:01:33,320 --> 00:01:34,230
And there were more than.

15
00:01:36,600 --> 00:01:39,420
With each epoch or training, accuracy is increasing.

16
00:01:39,510 --> 00:01:43,110
But we are not able to increase our valuation accuracy.

17
00:01:43,290 --> 00:01:47,490
This is a clear sign of overfitting in our modern.

18
00:01:50,110 --> 00:01:54,130
Now, to fix this awful thing, we will create some dummy data.

19
00:01:55,420 --> 00:02:03,310
We will modify what existing data and two different forms by applying zoom shear.

20
00:02:04,590 --> 00:02:06,120
Rotation acceptor.

21
00:02:07,840 --> 00:02:11,170
And we will again train our Morton after a plane.

22
00:02:11,320 --> 00:02:12,820
All this modifications.

23
00:02:15,800 --> 00:02:20,330
Then we will compare its impact on our valuation accuracy.

24
00:02:20,730 --> 00:02:28,640
Our hypothesis is that by modifying our data and generating some more dummy data, we will be able to

25
00:02:28,640 --> 00:02:30,650
increase lower valuation accuracy.

26
00:02:34,650 --> 00:02:37,700
Now, before doing that, let's say, well, what a.

27
00:02:38,700 --> 00:02:44,430
We will use model, dot, save and then the fight in which we want to save.

28
00:02:48,670 --> 00:02:56,370
Now, let's dilute this model and clear the session so that we can proceed on with our next model.

29
00:02:58,460 --> 00:03:01,390
That's all for this letter in the Legacy Lecture.

30
00:03:01,920 --> 00:03:06,080
We'll modify our data and we will again train this modern.

31
00:03:07,260 --> 00:03:07,640
Thank you.