8. Build CNNs in Python using Keras/3. Building a Handwriting Recognition CNN.srt

1
00:00:00,720 --> 00:00:06,790
Hi and welcome to chapter eight point two we are now about ready to start building our handwriting recognition.

2
00:00:06,780 --> 00:00:08,570
CNN In kurus.

3
00:00:08,760 --> 00:00:14,040
But before we actually dive into the code let's actually talk about how we approach this problem and

4
00:00:14,040 --> 00:00:16,190
what the problem actually is.

5
00:00:16,230 --> 00:00:20,430
It's a very good practice in data science to actually think about your problem.

6
00:00:20,430 --> 00:00:22,100
Think of what you assert.

7
00:00:22,140 --> 00:00:22,700
No a bit.

8
00:00:22,720 --> 00:00:28,640
But you did us at first before trying anything crazy and treating a model on your data.

9
00:00:29,100 --> 00:00:30,850
So we're going to use the meanest.

10
00:00:30,870 --> 00:00:32,690
This is a famous dataset.

11
00:00:32,730 --> 00:00:40,770
Basically it comprises of 60000 images with 10000 test images and 60000 images comprised basically of

12
00:00:40,770 --> 00:00:41,970
10 classes.

13
00:00:42,180 --> 00:00:51,840
So you have a Towson's of zeros ones two is all of them here in this training dataset and you can read

14
00:00:51,840 --> 00:00:52,830
more about it here.

15
00:00:52,890 --> 00:00:54,860
You click these links afterward.

16
00:00:54,870 --> 00:00:57,750
Basically this is the guy Yan A-Kon.

17
00:00:57,900 --> 00:01:00,970
Basically he was a guy who trained Linette I believe.

18
00:01:01,260 --> 00:01:07,760
And that was due basically to themis CNN had actually scored 99 percent accuracy on this data set.

19
00:01:07,770 --> 00:01:14,400
It's quite interesting so the problem let's look at the original of the amnesty the set it was developed

20
00:01:14,400 --> 00:01:20,820
by the U.S. Postal Service because they needed a way to automatically read postcards handwritten postcards

21
00:01:20,910 --> 00:01:22,080
on mail.

22
00:01:22,200 --> 00:01:28,170
So the for classifier now is to take these digits and basically establish build a model that will tell

23
00:01:28,170 --> 00:01:38,350
us this is a 5 6 2 7 4 and 8 and this is a CNN reboot we are online to CNN before I showed you how to

24
00:01:38,350 --> 00:01:38,830
build it.

25
00:01:38,830 --> 00:01:43,140
And Chris we're actually now about to build the CNN in Paris.

26
00:01:43,240 --> 00:01:44,500
So it's going to be quite exciting.

27
00:01:44,500 --> 00:01:46,160
So let's move on to your iPod.

28
00:01:46,210 --> 00:01:46,720
In the book.