1
00:00:02,200 --> 00:00:07,080
So now our training is almost complete.

2
00:00:07,300 --> 00:00:15,220
You can see on the 1980 book we were getting a validation accuracy of points and four and a training

3
00:00:15,310 --> 00:00:17,230
accuracy of mine did over the same

4
00:00:20,570 --> 00:00:22,610
and in the last epoch.

5
00:00:22,610 --> 00:00:28,750
That is the 20th epoch we are getting no validation accuracy of 73 percent.

6
00:00:28,750 --> 00:00:41,900
Now let's just blot all this accuracy values on the graph.

7
00:00:42,890 --> 00:00:51,500
Now the orange line here is our training accuracy and the red line is validation accuracy and green

8
00:00:51,500 --> 00:00:58,870
and blue lines are validation lost and training loss respectively

9
00:01:03,270 --> 00:01:09,480
one thing here is to notice that there is a large difference between validation accuracy and training

10
00:01:09,480 --> 00:01:21,570
accuracy validation accuracy is oscillating around 73 to 74 percent whereas training accuracy is increasing

11
00:01:21,630 --> 00:01:25,930
with each epoch currently it is like 93 percent.

12
00:01:26,280 --> 00:01:37,230
And it is increasing this graph such as that that is then what if waiting and no more than with each

13
00:01:37,230 --> 00:01:39,440
epoch or training accuracy is increasing.

14
00:01:39,510 --> 00:01:43,290
But we are not able to increase our validation accuracy.

15
00:01:43,290 --> 00:01:53,470
Does that clear sign of or what fitting in our model now to fix this awful thing we will create some

16
00:01:53,470 --> 00:01:55,030
dummy data.

17
00:01:55,420 --> 00:02:08,410
We will modify our existing data into different forms by applying zoom shear rotation etc. and we will

18
00:02:08,500 --> 00:02:12,820
again train our model after applying all this modifications

19
00:02:15,770 --> 00:02:20,660
then we will compare its impact on our validation accuracy.

20
00:02:20,720 --> 00:02:28,640
Our hypothesis is that by modifying our data and generating some more dummy data we will be able to

21
00:02:28,640 --> 00:02:30,680
increase our validation accuracy

22
00:02:34,620 --> 00:02:36,640
now before doing that.

23
00:02:36,690 --> 00:02:44,430
Let's save over more than we will lose model dot safe than the fight in which we want to save

24
00:02:48,660 --> 00:02:49,180
now.

25
00:02:49,210 --> 00:02:51,090
Let's delude this model.

26
00:02:51,280 --> 00:03:00,520
And clear the session so that we can proceed on with our next model that's all for this lecture in the

27
00:03:00,520 --> 00:03:01,820
next lecture.

28
00:03:01,870 --> 00:03:07,650
We'll modify our data and we will again train this than thank you.