1
00:00:00,510 --> 00:00:01,320
Welcome back.

2
00:00:01,860 --> 00:00:06,630
I hope you have installed get us and intensive flow in your system.

3
00:00:08,270 --> 00:00:13,490
Now, as a practice project, we are going to create an inmate classifier.

4
00:00:15,400 --> 00:00:23,380
We will be classifying this kind of images into ten different categories, such as T-shirts, trousers,

5
00:00:23,410 --> 00:00:26,870
pullover dresses, bags, boots, et cetera.

6
00:00:29,130 --> 00:00:36,330
For this, we are going to use a very famous database that is known as Fashion Emmis Database.

7
00:00:37,800 --> 00:00:45,900
Here we have our own 70000 great scale images of ten different fashion categories, objects.

8
00:00:47,500 --> 00:00:49,700
A training site will be off.

9
00:00:49,850 --> 00:00:55,580
60000 images which we are going to use to screen over more than.

10
00:00:57,640 --> 00:01:05,680
We have another site of 10000 images which we will be using as a test site to evaluate the performance

11
00:01:05,680 --> 00:01:06,430
of our model.

12
00:01:07,870 --> 00:01:16,480
These images are in the form of twenty eight by twenty eight, pixel is square and each pixel is represented

13
00:01:16,690 --> 00:01:17,740
on a gray skin.

14
00:01:17,890 --> 00:01:20,590
On a scale of zero to 255.

15
00:01:21,840 --> 00:01:29,400
The great thing about this database is that it is available within us and we can directly imported from

16
00:01:29,400 --> 00:01:29,940
caite us.

17
00:01:30,210 --> 00:01:34,750
We don't have to upload a separate file to access this database here.

18
00:01:35,160 --> 00:01:39,810
You can see the ten different objects that are present in this database.

19
00:01:43,330 --> 00:01:50,470
Now, to access this database from Get US, we are first creating a fashion underscored amnesty object

20
00:01:51,360 --> 00:01:59,950
where we are calling this database and after that we are loading this database into our X and via datasets.

21
00:02:01,160 --> 00:02:07,490
We are calling Overtrain dataset as X, underscore green, underscore the full and via underscore craine,

22
00:02:07,520 --> 00:02:13,790
underscore full and our best dataset as X underscore best and Švanda scored best.

23
00:02:14,570 --> 00:02:17,180
So just load this database.

24
00:02:19,210 --> 00:02:24,700
No, we can use my library to view images in this database.

25
00:02:24,880 --> 00:02:31,990
So, for example, if I want to access the first image, I can just write Xander score green under score

26
00:02:31,990 --> 00:02:32,320
full.

27
00:02:32,650 --> 00:02:34,840
And we're accessing the first element.

28
00:02:34,930 --> 00:02:38,490
That is the picture that is present at position zero.

29
00:02:42,940 --> 00:02:48,010
You can see that this is our first image if we want to exit the second image.

30
00:02:48,340 --> 00:02:50,470
I'm just changing the location to one.

31
00:02:52,720 --> 00:02:58,040
In this way, you can access the different images that are present in this database.

32
00:03:01,020 --> 00:03:03,240
Now, this is our X variable.

33
00:03:04,290 --> 00:03:08,640
This out of the pixels that we are going to use to predict the object.

34
00:03:10,570 --> 00:03:13,300
Not to view the actual category of this object.

35
00:03:14,110 --> 00:03:21,010
We have to call the wide screen dataset so you can just call the element that is present at the first

36
00:03:21,010 --> 00:03:21,550
position.

37
00:03:21,760 --> 00:03:22,960
So if I run this.

38
00:03:25,210 --> 00:03:27,820
You can see that the output is zero.

39
00:03:29,530 --> 00:03:38,290
To view the category that is responding to this zero label, we can refer this about table here, zero

40
00:03:38,290 --> 00:03:39,940
sense for t shirt and tops.

41
00:03:40,150 --> 00:03:44,980
One is sense for Krosa, who stands for lower and so on.

42
00:03:47,450 --> 00:03:54,830
So this image is of a T-shirt and the output is also representing that this is a T-shirt.

43
00:03:58,480 --> 00:04:02,270
We just checked this for the first segment.

44
00:04:04,250 --> 00:04:10,010
You can see that this is a boot and the vital label is nine.

45
00:04:11,300 --> 00:04:15,260
If you see the labor line correspond to ankle boot.

46
00:04:16,880 --> 00:04:22,490
So instead of referring people each time, we can create a list of class name.

47
00:04:23,360 --> 00:04:23,660
We are.

48
00:04:23,710 --> 00:04:29,180
We have this search all look at degrees in the order of their labels.

49
00:04:30,820 --> 00:04:38,010
So that if I call the first sentiment of this list, it pretty likely give me the T-shirt, if I call

50
00:04:38,020 --> 00:04:38,890
the second element.

51
00:04:39,220 --> 00:04:41,580
It will give closeups as an old port.

52
00:04:42,390 --> 00:04:45,920
So instead of calling the labels, I can medically call.

53
00:04:45,940 --> 00:04:50,610
The description of those labels using this last name list.

54
00:04:52,490 --> 00:04:57,810
So just check the image off object that is present location.

55
00:04:58,010 --> 00:04:58,370
Ben?

56
00:05:02,610 --> 00:05:04,920
Who checked the white label of this object?

57
00:05:05,250 --> 00:05:10,680
I can medically call last name and then the label of the pollution pan.

58
00:05:12,450 --> 00:05:15,540
You can see the last name here is Shirt Thoughtful.

59
00:05:19,200 --> 00:05:19,390
Few.

60
00:05:19,890 --> 00:05:27,690
Here we are using my plot to plot the data that is stored in the extreme dataset, not to view the content

61
00:05:27,690 --> 00:05:28,440
of this data.

62
00:05:29,100 --> 00:05:29,850
You can.

63
00:05:33,360 --> 00:05:37,460
You can just write extra in full and then call the object.

64
00:05:38,660 --> 00:05:45,290
So earlier I mentioned that this images are of 28 by 28, greyscale, formic.

65
00:05:46,160 --> 00:05:51,950
So here in the data, you are seeing 28 to 28 pixel values.

66
00:05:53,390 --> 00:05:57,710
These are the pixels that are present at the first rule.

67
00:06:00,300 --> 00:06:02,110
These are the pixels that are present.

68
00:06:02,300 --> 00:06:06,330
The second row and so on for the 20th through here.

69
00:06:06,660 --> 00:06:11,180
Zero to present pure black and 255 represents white.

70
00:06:12,220 --> 00:06:16,510
So the location of first pixel, that is the first row and the first pixel.

71
00:06:17,050 --> 00:06:19,060
You can see it's pure black.

72
00:06:19,240 --> 00:06:21,040
That's why we are getting zero.

73
00:06:22,090 --> 00:06:27,130
So our data is present in this form to view the data.

74
00:06:27,250 --> 00:06:28,120
You have to use.

75
00:06:28,210 --> 00:06:34,480
I am sure, mentored and to get the vibe, lose the actual category of this data.

76
00:06:35,140 --> 00:06:42,580
You have to call wide-screen data and you can also use last name list to directly get the descriptions

77
00:06:42,710 --> 00:06:43,470
set off label.

78
00:06:44,440 --> 00:06:49,240
So this is the data that we are going to use for each record.

79
00:06:49,360 --> 00:06:54,460
We have 28 include 28 values, that is 784 values.

80
00:06:55,660 --> 00:07:00,860
And using these values, we are going to predict the description of this.

81
00:07:01,510 --> 00:07:03,130
So that's all for the data.

82
00:07:03,460 --> 00:07:05,800
You have the raw data can view this data.

83
00:07:06,040 --> 00:07:07,360
And you have the last names.

84
00:07:07,510 --> 00:07:10,370
Now, we have to create more than on this data.

85
00:07:11,230 --> 00:07:17,620
In the next lecture, we will normalize our dataset and further divided our train dataset into validation

86
00:07:17,620 --> 00:07:18,720
and train set.

87
00:07:19,200 --> 00:07:19,620
Thank you.
