1
00:00:00,810 --> 00:00:07,000
OK so now let's start building our first CNN model.

2
00:00:07,170 --> 00:00:16,590
We will be using the same data that we use for classification problem in and then here the task is to

3
00:00:16,590 --> 00:00:23,730
identify the fashion article name depending on their images.

4
00:00:23,730 --> 00:00:27,450
We have 10 categories of different objects.

5
00:00:27,450 --> 00:00:38,070
We have t shirts trousers words and for all of these articles we have their 28 by 28 pixel grayscale

6
00:00:38,130 --> 00:00:40,960
images.

7
00:00:41,190 --> 00:00:47,520
We have already created a neural network model for this classification problem.

8
00:00:47,520 --> 00:00:54,090
Now we are going to egg convolution a layered width of it and then more than that.

9
00:00:54,580 --> 00:00:59,630
So let's just start by importing some of the important libraries.

10
00:00:59,640 --> 00:01:08,580
We are importing numbed by a US and my Lord then we are also importing tens of law and get us

11
00:01:11,230 --> 00:01:17,770
then as I have told you earlier we will be using fashion and honest data.

12
00:01:17,770 --> 00:01:22,480
This data is available in get us datasets.

13
00:01:22,510 --> 00:01:26,870
For more information you can click on this link here.

14
00:01:27,100 --> 00:01:35,590
We have our own 60000 images as our training data and another 10000 images as of our test data.

15
00:01:35,590 --> 00:01:42,330
We have ten different categories all these categories are label from 0 to 9.

16
00:01:44,320 --> 00:01:46,300
And this is this index.

17
00:01:46,300 --> 00:01:50,170
We are going to use the import fashion amnesty data.

18
00:01:50,170 --> 00:01:53,770
We have already done this in our aim in tutorial.

19
00:01:54,010 --> 00:01:57,910
So I'm not going to spend our time on this

20
00:02:01,320 --> 00:02:10,120
so we are importing over data and to explain full why train full x test and Vi test variables.

21
00:02:10,120 --> 00:02:16,680
Let us just run this now since we have labels for all these articles.

22
00:02:16,800 --> 00:02:23,650
We are going to create a list with their description so that we can refer to this list whenever we get

23
00:02:23,650 --> 00:02:25,800
a class label.

24
00:02:25,920 --> 00:02:28,280
So let's create a class name.

25
00:02:28,290 --> 00:02:30,900
This also.

26
00:02:31,680 --> 00:02:36,330
Now we have to do better at reshaping this Donnelly change.

27
00:02:36,330 --> 00:02:42,760
We are going to do in three processing of coin delusional neural network as compared to an

28
00:02:45,540 --> 00:02:46,780
if you remember four.

29
00:02:46,820 --> 00:02:54,410
And then we convert it over to the images into a single one dimensional array losing light and function

30
00:02:55,470 --> 00:03:00,780
but for CNN we need a three dimensional array as input.

31
00:03:00,900 --> 00:03:08,180
We need height weight and also another dimension for channels.

32
00:03:08,400 --> 00:03:11,760
Currently we have our X data in the form of this.

33
00:03:11,780 --> 00:03:12,900
Good images.

34
00:03:12,900 --> 00:03:22,020
There is no another dimension for channels since these are simple grayscale images but by default for

35
00:03:22,020 --> 00:03:25,900
CNN layers we need a three dimensional images.

36
00:03:26,070 --> 00:03:34,400
So we are going to reshape what external data and we are going to add another dimension to our data.

37
00:03:34,410 --> 00:03:39,690
So earlier we were using commandeered by 20th pixel images.

38
00:03:39,690 --> 00:03:45,050
Now we are reshaping it and 228 and 228 and 2 1 1.

39
00:03:45,050 --> 00:03:54,090
It stands for channel and again we have 60000 images in our training dataset and 10000 images in the

40
00:03:54,100 --> 00:04:04,960
work as dataset so before doing this reshape the shape of our extreme full dataset was 60000 across

41
00:04:04,970 --> 00:04:11,110
28 across 28 now for convolution on neural network.

42
00:04:11,110 --> 00:04:14,050
We are adding another dimension for the channel as well.

43
00:04:14,440 --> 00:04:21,430
So we are just adding another dimension to make it four dimensional.

44
00:04:21,430 --> 00:04:24,030
Just on this.

45
00:04:24,040 --> 00:04:26,680
Now we have reshaped our data.

46
00:04:26,680 --> 00:04:36,700
The next step is to normalize the data so all lower pixel values are between 0 to 255.

47
00:04:37,600 --> 00:04:42,420
So we are just going to divide our entire dataset by 255.

48
00:04:42,460 --> 00:04:47,460
In that way all lower values will lie between 0 and 1.

49
00:04:48,100 --> 00:04:52,910
We already did the similar thing for a word and then model as well.

50
00:04:52,990 --> 00:04:56,250
So I'm not going to spend much time here.

51
00:04:57,910 --> 00:05:04,470
Similarly we are going to split our dataset and do train and validation sites.

52
00:05:04,570 --> 00:05:13,210
We are keeping 55000 images for our training the desert and dress of 5000 images for our validation

53
00:05:13,210 --> 00:05:13,750
dataset.

54
00:05:15,250 --> 00:05:17,910
So let's just on this as well.

55
00:05:19,090 --> 00:05:28,340
So first 5000 are going into validation and from 5000 to 6000 are going into training data set.

56
00:05:30,080 --> 00:05:39,310
Now let's say the seed for a word then sort of flow and number so that we get the same result every

57
00:05:39,310 --> 00:05:40,410
time we run this code.