AI_DL_Assignment / 8. Build CNNs in Python using Keras /5. Getting our data in ‘Shape’.srt
Prince-1's picture
Add files using upload-large-folder tool
d157f08 verified
1
00:00:00,760 --> 00:00:01,300
Right.
2
00:00:01,320 --> 00:00:08,130
So we've just imported the data into carious However that is not in the ideal or the correct format
3
00:00:08,130 --> 00:00:10,410
that cares needs to train on.
4
00:00:10,440 --> 00:00:15,370
So this chapter is going to be a tiny chapter in how to get us data into the correct shape.
5
00:00:15,450 --> 00:00:17,020
So let's look at the next slide.
6
00:00:17,330 --> 00:00:21,780
So as I said before we've brought this in here and help with it it's here.
7
00:00:21,780 --> 00:00:22,650
All right.
8
00:00:22,650 --> 00:00:29,340
How ever terrorists requires it to be in a special shape and that she basically as number of rows columns
9
00:00:29,910 --> 00:00:33,040
a number of samples rows of columns and Dept.
10
00:00:33,120 --> 00:00:40,880
So this is how extreme it looks here but healthcare is actually once it is 16000 to 28 and 1.
11
00:00:40,950 --> 00:00:46,090
So we sort of have to add a fourth dimension antoh data and that's how we do it here.
12
00:00:46,090 --> 00:00:47,580
So now let's do this in Iraq.
13
00:00:47,910 --> 00:00:52,980
And the reason Chris is looking for a fourth dimension here is because when you load a greyscale image
14
00:00:52,980 --> 00:00:57,930
dataset it doesn't add the dimension of the depth to up.
15
00:00:58,230 --> 00:01:00,550
Mainly because it's a greyscale and is Nakul adept.
16
00:01:00,630 --> 00:01:04,850
If it was a color image dataset you would have the tree popping up at the end here.
17
00:01:05,160 --> 00:01:10,390
However we each need to add this one just to indicate to Cara's that it's a greyscale image data set.
18
00:01:10,590 --> 00:01:17,070
Otherwise it's going to return an error is an incorrect format so to change this you would just use
19
00:01:17,070 --> 00:01:20,210
non-place reship function which is quite simple to use.
20
00:01:20,220 --> 00:01:23,170
You just use extreme thought reshape.
21
00:01:23,170 --> 00:01:24,760
Take the first 60000 digits.
22
00:01:24,810 --> 00:01:27,620
That's when you just when you use shape.
23
00:01:27,780 --> 00:01:30,640
This is a first I mentioned at 0 addresses.
24
00:01:30,750 --> 00:01:37,310
Then you enter 28 28 or image or image columns and add 1 and it's quite simple to use.
25
00:01:37,320 --> 00:01:38,880
So let's get into it.
26
00:01:39,190 --> 00:01:42,350
I buy that book now.
27
00:01:42,740 --> 00:01:43,120
OK.
28
00:01:43,160 --> 00:01:49,220
So like we just saw this is a section of code where we reshape and add the fourth dimension onto these
29
00:01:49,790 --> 00:01:50,610
treating datasets.
30
00:01:50,620 --> 00:01:52,580
Untested sets.
31
00:01:52,620 --> 00:01:57,650
I didn't point this out to you before but we just get image rows and image columns by simply addressing
32
00:01:57,650 --> 00:02:00,450
the first mention of the shape here extreme.
33
00:02:00,540 --> 00:02:08,030
Now we use extreme zero and extreme in one we don't need to use Xorn we can use X test as well because
34
00:02:08,030 --> 00:02:09,850
they're the same dimensionality.
35
00:02:09,890 --> 00:02:15,530
However this just gives us the rows and columns so let's actually print this out so you can actually
36
00:02:15,530 --> 00:02:16,990
see what's going on here.
37
00:02:17,480 --> 00:02:21,800
So this run print that you'll see it turns 28.
38
00:02:21,800 --> 00:02:23,590
So this is getting to 28 here.
39
00:02:23,960 --> 00:02:32,210
And if we were to just use this for columns two you'll see will also get 28 right.
40
00:02:32,210 --> 00:02:38,130
So moving on this is fairly simple if you guys make it nice and neat.
41
00:02:38,150 --> 00:02:41,010
Again this is how we make the input shape.
42
00:02:41,010 --> 00:02:48,340
Now you've seen before we needed input shape dimension in the first layer of the convolutional on that.
43
00:02:48,450 --> 00:02:50,870
And this is how we just create our image shape here.
44
00:02:51,060 --> 00:02:52,540
It's a tuple That's combined.
45
00:02:52,550 --> 00:02:57,560
What rows columns and the dimensionality of the depth of the image.
46
00:02:57,710 --> 00:02:58,540
This would be tree.
47
00:02:58,590 --> 00:03:04,400
Again if it was a color image dataset and because Kurus expected it to be in a floated to.
48
00:03:04,730 --> 00:03:08,090
Right now I believe it's going to be in some sort of integer format here.
49
00:03:08,490 --> 00:03:10,360
So we just need to change this.
50
00:03:10,830 --> 00:03:15,410
So we just change extra in as type flow to the two and do it for X test as well.
51
00:03:15,570 --> 00:03:18,280
We don't need to do this for the labels just the turning of data.
52
00:03:18,780 --> 00:03:25,290
And this is how we normalized data we normalize all data by dividing it by 255 because remember images
53
00:03:25,380 --> 00:03:27,270
range from 0 to 255.
54
00:03:27,530 --> 00:03:34,720
So if we divide it all the image data by 255 we basically bring it into range 0 to 1.
55
00:03:34,830 --> 00:03:36,470
So that's quite simple here now.
56
00:03:36,720 --> 00:03:40,350
And basically we just print this out again if we need to do that because we did it before.
57
00:03:40,630 --> 00:03:43,680
But just do a sanity check on our data.
58
00:03:43,770 --> 00:03:45,480
So let's run this again.
59
00:03:45,480 --> 00:03:49,770
If you want to see what extreme It actually looks like now this is it.
60
00:03:49,860 --> 00:03:53,930
You can see basically the decimal points because it's going to hidden here.
61
00:03:54,720 --> 00:03:59,870
But all the data ranges from 0 to 256 0 to 1 right now.
62
00:04:00,430 --> 00:04:00,670
OK.
63
00:04:00,690 --> 00:04:03,440
So let's move on to heart encoding labels.