1 00:00:00,690 --> 00:00:03,720 ‫Let's start by importing our BTC. 2 00:00:05,040 --> 00:00:12,480 ‫As we've discussed earlier we are going to use defection and my SD dataset to classify images of fashion 3 00:00:12,480 --> 00:00:22,760 ‫objects such as those goods boots etc. fashion MONUC Yazdi is a very popular dataset. 4 00:00:23,310 --> 00:00:31,860 ‫It is relatively small and is used to verify that an algorithm books as expected are not efficient. 5 00:00:31,890 --> 00:00:41,440 ‫Amnesty consists of a training set of sixty thousand examples and a test set of 10000 examples. 6 00:00:41,700 --> 00:00:50,640 ‫Each example is a twenty eight by twenty eight grayscale image that is it has twenty eight pixels by 7 00:00:50,640 --> 00:00:55,500 ‫twenty eight pixels dimensions and it is a black and white image. 8 00:00:58,960 --> 00:01:06,000 ‫With each image there is an associated label of in glasses. 9 00:01:06,310 --> 00:01:08,970 ‫I'll show you the images after we import the dataset. 10 00:01:10,290 --> 00:01:13,520 ‫Let's run this line of code to import data say 11 00:01:20,310 --> 00:01:23,280 ‫you know there are multiple ways to import data. 12 00:01:23,370 --> 00:01:28,990 ‫Here we are using the big dataset that comes with us liability. 13 00:01:29,090 --> 00:01:36,750 ‫So if the US package is not installed this line more book if it is installed you will get the dataset 14 00:01:36,780 --> 00:01:46,570 ‫imported into this video building which is fashion underscored emanate SD. 15 00:01:46,790 --> 00:01:55,460 ‫You can see on the right in the environment variable window here exactly does it let us view this dataset 16 00:01:55,510 --> 00:01:59,980 ‫by clicking on it. 17 00:02:00,650 --> 00:02:06,560 ‫Here you can see that this dataset has to pass grain and test. 18 00:02:06,560 --> 00:02:12,680 ‫This means that it is already divided into two parts of planing and pasting. 19 00:02:12,710 --> 00:02:15,220 ‫We do not need to do this separately. 20 00:02:15,370 --> 00:02:22,760 ‫However if you want to learn how to separate any data set into train and test which is not in this format 21 00:02:23,720 --> 00:02:26,220 ‫please take the opening section of this course. 22 00:02:26,660 --> 00:02:30,620 ‫There you will find a lecture theatre test train split. 23 00:02:30,620 --> 00:02:35,840 ‫With that you will be able to split any day does it make sense. 24 00:02:35,840 --> 00:02:39,100 ‫Here are data set is already split. 25 00:02:39,440 --> 00:02:46,690 ‫Let's go for the train set for that has two parts X and Y. 26 00:02:46,910 --> 00:02:52,730 ‫X is the set of productivity levels and y is the list of output values. 27 00:02:52,730 --> 00:02:55,100 ‫That is the class of defection object 28 00:02:58,940 --> 00:03:06,860 ‫you can see the structure of X and Y also here X is a set of sixty thousand images. 29 00:03:06,860 --> 00:03:09,430 ‫We talked to indeed pixel by two indeed pixels. 30 00:03:10,790 --> 00:03:16,820 ‫So for each image we have a value between 0 and 255. 31 00:03:18,080 --> 00:03:22,460 ‫If the value is 0 that pixel is black. 32 00:03:22,460 --> 00:03:25,850 ‫If it is 255 that pixel is white. 33 00:03:27,440 --> 00:03:34,440 ‫So each individual pixels data for all the sixty thousand images is stored in this exhibit even 34 00:03:37,580 --> 00:03:38,700 ‫similarly. 35 00:03:38,750 --> 00:03:43,420 ‫Why has the glassware lose of 60000 images. 36 00:03:43,970 --> 00:03:48,680 ‫For example the first image has the glass nine. 37 00:03:48,830 --> 00:03:55,520 ‫What this 90 percent look at that in something similar to the train data we have this data. 38 00:03:56,090 --> 00:03:57,930 ‫Only difference is in the training set. 39 00:03:57,950 --> 00:04:01,050 ‫We have sixty thousand images data in it. 40 00:04:01,130 --> 00:04:03,820 ‫We have 10000 image data. 41 00:04:04,250 --> 00:04:07,390 ‫We will use this train data to bring that model. 42 00:04:07,790 --> 00:04:15,230 ‫And later on we will predictive y values for this basic using the x values of this data. 43 00:04:15,890 --> 00:04:22,550 ‫Then we will compare the actual y values in this test set with the predicted y values from my model 44 00:04:23,120 --> 00:04:26,930 ‫to find out the accuracy of our model. 45 00:04:26,930 --> 00:04:36,860 ‫Now let's go back to our board will be assigning the x and y brain values to separate variables to do 46 00:04:36,860 --> 00:04:37,490 ‫that. 47 00:04:37,490 --> 00:04:43,440 ‫This line of code is the standard way in which we assign value to a variable. 48 00:04:43,820 --> 00:04:46,900 ‫You can run this line also and it will give you the same result. 49 00:04:47,000 --> 00:04:54,780 ‫It will assign the x value of deplaning set of fashion amnesty variable into the train images. 50 00:04:54,790 --> 00:05:02,930 ‫We would however get us allows us to do that in a different way in this format. 51 00:05:02,930 --> 00:05:09,230 ‫You can assign the two variables brain images and three labels at the same time. 52 00:05:09,860 --> 00:05:17,070 ‫If you run this line of code this will assign the x values of brain to brain images and divide value 53 00:05:17,070 --> 00:05:20,570 ‫of brain to train labels. 54 00:05:20,570 --> 00:05:21,950 ‫Let's give them this line of code. 55 00:05:21,960 --> 00:05:31,760 ‫Now you can see that we have a brain images variable and then labels variable brain images has the X 56 00:05:31,760 --> 00:05:33,210 ‫part and bring labels. 57 00:05:33,200 --> 00:05:35,960 ‫I did my part. 58 00:05:35,990 --> 00:05:39,020 ‫Same goes with the test images and test labels. 59 00:05:39,020 --> 00:05:39,410 ‫Next one. 60 00:05:39,410 --> 00:05:44,070 ‫This code also and we have two more variables. 61 00:05:45,140 --> 00:05:52,190 ‫Although we have seen the structure of training data and this data if you still want to check out the 62 00:05:52,190 --> 00:05:57,700 ‫structure of the new variables you can know these two lines of code. 63 00:05:58,910 --> 00:06:05,990 ‫Them and within brackets variable name gives you the dimension of this variable to this variable has 64 00:06:06,920 --> 00:06:07,910 ‫three dimensions. 65 00:06:07,910 --> 00:06:16,520 ‫First is the sixty thousand values of different images and then 28 across 28 for all the individual 66 00:06:16,520 --> 00:06:18,290 ‫pixels. 67 00:06:18,410 --> 00:06:24,710 ‫If you're on the SDR command which gives you structure there'll be some additional information that 68 00:06:24,950 --> 00:06:31,920 ‫it has integer type of values and the initial few values are do do do do do do. 69 00:06:33,080 --> 00:06:40,250 ‫So both of these are used for the same thing to understand what is the structure of this variable that 70 00:06:40,250 --> 00:06:42,270 ‫we have. 71 00:06:42,270 --> 00:06:47,190 ‫Now let me show you the images so that you get a feel of what kind of data we have here. 72 00:06:49,160 --> 00:06:56,240 ‫We can store the information of one image into a variable called F object. 73 00:06:56,240 --> 00:07:05,150 ‫So when I've done this line of code it will assign the information of the fifth image all the pixels 74 00:07:05,750 --> 00:07:08,230 ‫into this object which is f object. 75 00:07:09,560 --> 00:07:12,200 ‫Let's end this. 76 00:07:12,200 --> 00:07:18,560 ‫You can see that f object is a wounded cross to indicate two dimensional edit containing all the pixel 77 00:07:18,560 --> 00:07:22,640 ‫data of this fifth image. 78 00:07:22,660 --> 00:07:31,280 ‫Now if you want to plot this image you can then this line of code which has blocked function in plot 79 00:07:31,280 --> 00:07:31,840 ‫function. 80 00:07:31,880 --> 00:07:37,730 ‫We are telling that we have to plot this variable as a raster image. 81 00:07:37,760 --> 00:07:41,510 ‫That estimate is basically a pixilated image. 82 00:07:41,510 --> 00:07:43,040 ‫So when we done this line of code. 83 00:07:43,520 --> 00:07:46,650 ‫So here you can see the image on the right. 84 00:07:46,730 --> 00:07:54,410 ‫It's a small twenty eight cross twenty eight pixel image so the image quality is not good but you can 85 00:07:54,410 --> 00:07:56,210 ‫make out the object. 86 00:07:56,210 --> 00:07:58,180 ‫It probably looks like a top. 87 00:07:58,610 --> 00:08:05,480 ‫If you want to check what it is we need to see the image label which is stored in the green label variable 88 00:08:08,970 --> 00:08:10,530 ‫in the train label variable. 89 00:08:10,530 --> 00:08:13,470 ‫We saw that the values are in decoded format. 90 00:08:13,470 --> 00:08:18,020 ‫That is it is written from 0 to 9. 91 00:08:18,300 --> 00:08:23,390 ‫So to get the actual name of the class B first grade class name. 92 00:08:23,410 --> 00:08:32,790 ‫Edit this edit contains the list of names in the order in which we have coded these names so Zito stands 93 00:08:32,790 --> 00:08:42,510 ‫for t shirt so if you see nine hit nine stands for ankle boot to stand spot pullover. 94 00:08:42,600 --> 00:08:44,070 ‫It starts with D2. 95 00:08:44,100 --> 00:08:45,440 ‫This is the second element. 96 00:08:45,660 --> 00:08:48,260 ‫This is the ninth element. 97 00:08:48,570 --> 00:08:56,130 ‫Once we created this array we can find out the name of this object which all t. 98 00:08:56,130 --> 00:09:07,340 ‫Fifth image indeed trainee labels variable so the label of the fifth image plus one because decoding 99 00:09:07,350 --> 00:09:08,930 ‫started with 0. 100 00:09:09,340 --> 00:09:16,090 ‫So we just want the plus Vernet to eliminate from this edit. 101 00:09:16,750 --> 00:09:25,530 ‫So let's first create the setting and now find out the name of this fifth image. 102 00:09:25,530 --> 00:09:29,100 ‫You can see that the 50 made is a t shirt slash top 103 00:09:32,160 --> 00:09:35,880 ‫you can check this again for an entertainment. 104 00:09:36,360 --> 00:09:47,940 ‫So let's try it out for ninth image on this come on my mean slaughtered I mean this looks like a sandal 105 00:09:50,180 --> 00:09:51,450 ‫not if we take the liberty 106 00:09:55,480 --> 00:10:08,210 ‫to be updated and dig delivered it comes out to it so this is our data we have created for variables 107 00:10:09,690 --> 00:10:17,930 ‫dream images contains all the predictive variables dream levels contains the output variable using these 108 00:10:17,930 --> 00:10:26,690 ‫two variables will be bringing our model then we will be using that model to predict on the test images 109 00:10:27,230 --> 00:10:29,870 ‫and we will compare the prediction of the test levels 110 00:10:32,870 --> 00:10:41,710 ‫the last thing I'm going to discuss in this video is normalization of data when we have heterogeneous 111 00:10:41,710 --> 00:10:48,310 ‫data a learning model takes a lot of time to converge to handle this problem we do normalization of 112 00:10:48,310 --> 00:10:58,300 ‫data to normalize data usually a general formalize we subtract the mean of that variable from that variable 113 00:10:59,140 --> 00:11:02,420 ‫and divided by the standard deviation. 114 00:11:02,920 --> 00:11:11,350 ‫So this is the general formula but since our training data is not that heterogeneous every value is 115 00:11:11,350 --> 00:11:15,370 ‫of a pixel having a value between 0 to 255. 116 00:11:17,380 --> 00:11:23,760 ‫So we can just divide all the values in the pixels by 255. 117 00:11:23,800 --> 00:11:31,030 ‫This will result in values between 0 to 1 and we can input these values into our operating model. 118 00:11:32,230 --> 00:11:37,390 ‫So normalization is required when we have different types of variables in a dataset. 119 00:11:37,600 --> 00:11:44,590 ‫If that is the case use this formula to normalize here since our model is already very homogeneous we 120 00:11:44,590 --> 00:11:53,040 ‫can just divide the the pixel values by the highest value to get the simple normalized value. 121 00:11:53,530 --> 00:11:59,130 ‫Now using these train and test values we'll be creating a model in the next 12.