1
00:00:00,000 --> 00:00:02,725
Let's take a look at
training cats versus dogs

2
00:00:02,725 --> 00:00:05,380
with a smaller dataset
over a longer cycle.

3
00:00:09,780 --> 00:00:11,760
I'm going to start
the training and we'll see it

4
00:00:11,760 --> 00:00:13,650
picked up the 2,000
training images,

5
00:00:13,650 --> 00:00:16,800
but note that we're
training for 100 epochs.

6
00:00:21,000 --> 00:00:22,500
After the first epoch,

7
00:00:22,540 --> 00:00:25,800
We'll watch the accuracy
and validation accuracy figures.

8
00:00:31,500 --> 00:00:33,500
Let's watch them for
a few more epochs.

9
00:00:34,500 --> 00:00:36,000
Keep an eye on those figures.

10
00:00:40,500 --> 00:00:44,595
After eight epochs, the accuracy
is approaching 0.8,

11
00:00:44,595 --> 00:00:47,625
but the validation accuracy
has slowed its growth.

12
00:00:48,625 --> 00:00:51,130
So now let's skip
ahead to the end.

13
00:00:51,130 --> 00:00:53,290
I'm going to plot
the accuracy and

14
00:00:53,290 --> 00:00:55,795
loss overall 100 epochs.

15
00:00:55,795 --> 00:00:57,410
We can see from this figure that

16
00:00:57,410 --> 00:00:58,940
the training reach close to

17
00:00:58,940 --> 00:01:02,575
a 100 percent accuracy in
a little over 20 epochs.

18
00:01:02,575 --> 00:01:04,820
Meanwhile, the validation topped

19
00:01:04,820 --> 00:01:06,950
out at around 70 percent,

20
00:01:06,950 --> 00:01:10,100
and that's overfitting
clearly been demonstrated.

21
00:01:10,100 --> 00:01:12,320
In other words,
the neural network was

22
00:01:12,320 --> 00:01:15,230
terrific at finding
a correlation between

23
00:01:15,230 --> 00:01:17,210
the images and labels
of cats versus

24
00:01:17,210 --> 00:01:20,720
dogs for the 2,000 images
that it was trained on,

25
00:01:20,720 --> 00:01:22,700
but once it tried to
predict the images

26
00:01:22,700 --> 00:01:24,665
that it previously hadn't seen,

27
00:01:24,665 --> 00:01:26,660
it was about 70 percent accurate.

28
00:01:26,660 --> 00:01:28,370
It's a little bit
like the example

29
00:01:28,370 --> 00:01:30,350
of the shoes we
spoke about earlier.

30
00:01:30,350 --> 00:01:31,730
So in the next video,

31
00:01:31,730 --> 00:01:33,290
we'll take a look
at the impact of

32
00:01:33,290 --> 00:01:35,840
adding augmentation to this.