1
00:00:02,250 --> 00:00:03,690
The training part is complete.

2
00:00:04,470 --> 00:00:12,840
The training ran for 20 epochs, and at the last epoch, we have a training accuracy of 93 percent and

3
00:00:12,840 --> 00:00:15,570
validation accuracy of seventy three point six percent.

4
00:00:17,430 --> 00:00:21,720
You can see that there is a huge difference in validation, accuracy and training accuracy.

5
00:00:22,920 --> 00:00:28,440
This is a sign of overfitting after 10 to 15 debug.

6
00:00:28,460 --> 00:00:33,000
We can see that there is no considerable improvement in the validation accuracy.

7
00:00:33,720 --> 00:00:36,120
However, our training accuracy keeps on increasing.

8
00:00:37,170 --> 00:00:40,560
This is because our model is trying to work on the training data.

9
00:00:41,160 --> 00:00:47,760
However, we are not seeing any considerable improvement in the validation accuracy to see the result.

10
00:00:47,790 --> 00:00:54,420
In the last model, which is the model I did when did Epoch, we can run this command of saving saving

11
00:00:54,420 --> 00:00:56,660
model in the HD five format.

12
00:01:01,640 --> 00:01:05,030
This will create a cat and dog small wonder.

13
00:01:05,240 --> 00:01:07,610
It's five file in the working directory.

14
00:01:09,680 --> 00:01:14,040
This last combine is for plotting the accuracy and lost function.

15
00:01:15,540 --> 00:01:22,940
Odd by default, plotted for us, however, at a later point of pain, if you have removed this image

16
00:01:23,390 --> 00:01:29,780
and you want to plot the model accuracy and loss over 20 epochs, you can run this.

17
00:01:29,780 --> 00:01:30,250
Come on.

18
00:01:33,370 --> 00:01:42,820
So it is the same glove only that each epoch point for validation is marked with a cross and for training

19
00:01:42,920 --> 00:01:45,270
is with a subtler hollowed out.

20
00:01:48,100 --> 00:01:48,700
That's all.

21
00:01:50,800 --> 00:01:59,020
Notice that we have achieved a validation accuracy of 73 percent with this convolutional architecture.

22
00:02:00,880 --> 00:02:08,020
In the next video, we will create artificial data to augment this data that we have.

23
00:02:09,160 --> 00:02:13,480
And instead of having 2000 images, we'll have larger training data.

24
00:02:14,590 --> 00:02:19,480
And in that scenario, we'll have considerable improvement in ad validation accuracy.

25
00:02:20,800 --> 00:02:22,240
Let's see that in the next video.