1 00:00:00,570 --> 00:00:01,470 ‫Welcome back. 2 00:00:02,130 --> 00:00:09,120 ‫In the last lecture, we saw how to combine and train our model in this lecture. 3 00:00:09,210 --> 00:00:16,780 ‫We will learn how to evaluate performance of our model and how to predict classes on the new dataset 4 00:00:17,100 --> 00:00:27,150 ‫using over this train model first to evaluate performance of our model than we have to use the evaluate 5 00:00:27,150 --> 00:00:27,600 ‫method. 6 00:00:28,230 --> 00:00:33,440 ‫So just write, you have model object name, which in our case is model 7 00:00:33,840 --> 00:00:39,550 ‫And then we are calling evaluate method and inside evaluate. 8 00:00:39,720 --> 00:00:42,740 ‫We have to give our test datasets. 9 00:00:43,320 --> 00:00:49,910 ‫So we are writing X test first the independent test dataset and then dependent test set 10 00:00:51,150 --> 00:00:53,310 ‫So let's just run this. 11 00:00:58,910 --> 00:01:04,700 ‫So the output here is first the loss and then the accuracy. 12 00:01:05,330 --> 00:01:12,990 ‫This is accuracy because this is the metrics which we mention when compiling a lot more than a few remember. 13 00:01:13,510 --> 00:01:17,720 ‫While compiling, we mentioned metrics equal to accuracy. 14 00:01:18,350 --> 00:01:22,700 ‫And that's why the second value which we are getting here is of accuracy. 15 00:01:23,600 --> 00:01:33,230 ‫So on our test set, the accuracy which we are getting is 86 percent, which is really good as compared 16 00:01:33,230 --> 00:01:37,450 ‫to logistic liquidations or decision trees on the same dataset. 17 00:01:38,600 --> 00:01:42,860 ‫So with those models, you will get around 70 to 80 percent accuracy. 18 00:01:42,920 --> 00:01:47,810 ‫But here with ann we are getting 86 percent accuracy. 19 00:01:49,730 --> 00:01:55,170 ‫So once again, to check the accuracy score or just to check the performance of your model. 20 00:01:56,030 --> 00:01:57,190 ‫You have to use evaluate method 21 00:01:57,190 --> 00:01:57,770 ‫. 22 00:01:58,670 --> 00:02:06,560 ‫Now let's learn how to predict the probabilities and how to predict the classes on our new unseen data 23 00:02:07,900 --> 00:02:08,150 ‫since. 24 00:02:08,670 --> 00:02:10,460 ‫Right now we do not have any new data. 25 00:02:11,030 --> 00:02:20,690 ‫We just taking first three samples from our test dataset and consider that as your new unseen data now 26 00:02:21,080 --> 00:02:22,250 ‫just run this. 27 00:02:23,840 --> 00:02:31,340 ‫So we have Stored the first three records from our test dataset and we are considering it as our new record. 28 00:02:32,600 --> 00:02:35,060 ‫So there are two parameters which you can predict. 29 00:02:35,420 --> 00:02:42,320 ‫First one is the probability of each class and second one is the class. 30 00:02:42,380 --> 00:02:42,850 ‫Itself 31 00:02:44,790 --> 00:02:49,460 ‫Let's first calculate the probability score assigned to each class. 32 00:02:50,810 --> 00:02:53,740 ‫To do that, we can use dot predict method. 33 00:02:54,980 --> 00:02:59,840 ‫So we'll write our object name model then dot predict method 34 00:03:00,440 --> 00:03:04,830 ‫And as an argument, we are passing our new unseen data. 35 00:03:04,910 --> 00:03:06,440 ‫Which is X_new 36 00:03:08,730 --> 00:03:10,080 ‫Let's just run this. 37 00:03:11,930 --> 00:03:15,770 ‫You can see our new that contains three records. 38 00:03:16,100 --> 00:03:18,830 ‫That's why we are getting output for three records. 39 00:03:19,250 --> 00:03:28,360 ‫And there are ten values in each of these elements representing the probability values of corresponding 40 00:03:28,360 --> 00:03:28,820 ‫class. 41 00:03:31,090 --> 00:03:33,100 ‫So for the first record. 42 00:03:34,870 --> 00:03:41,670 ‫The maximum probability is for label equal to 10 and the probability value is 0.99 . 43 00:03:43,190 --> 00:03:49,910 ‫For the second record, probability is one, for the label corresponding to the third object and for the 44 00:03:49,910 --> 00:03:54,500 ‫third record, we have probably equal to hundred for the second object. 45 00:03:56,560 --> 00:04:02,840 ‫We are also using round equal to 2 just to round off these probabilities values to 2 decimal digits. 46 00:04:03,080 --> 00:04:03,400 ‫. 47 00:04:05,980 --> 00:04:07,360 ‫Now with dot predict 48 00:04:07,660 --> 00:04:14,530 ‫We are predicting probabilities, but if we want to directly predict the class and not the probability scores, 49 00:04:15,010 --> 00:04:17,540 ‫we can use dot predict underscore. 50 00:04:17,700 --> 00:04:18,600 ‫Classes method 51 00:04:20,170 --> 00:04:21,630 ‫So Syntex is almost same. 52 00:04:21,850 --> 00:04:28,080 ‫First, you have to write your model object name and then call the predict underscore classes method. 53 00:04:28,450 --> 00:04:31,400 ‫And in bracket, you have to mention your new dataset. 54 00:04:32,830 --> 00:04:33,760 ‫Just run this. 55 00:04:36,540 --> 00:04:40,800 ‫You can see here we are getting an array of the labels. 56 00:04:41,160 --> 00:04:42,600 ‫So for the first object. 57 00:04:44,210 --> 00:04:49,430 ‫The class label is nine, so you can also confirm it from here also. 58 00:04:50,720 --> 00:04:58,180 ‫The maximum probability was assigned to this object and here also we are getting the position as 9 59 00:04:58,510 --> 00:05:00,630 ‫remember the numbering 60 00:05:00,650 --> 00:05:01,850 ‫starts from zero 61 00:05:02,000 --> 00:05:03,350 ‫So this one is zero. 62 00:05:03,680 --> 00:05:04,640 ‫This one is one. 63 00:05:04,940 --> 00:05:05,600 ‫And so on. 64 00:05:05,840 --> 00:05:06,950 ‫And this one is nine. 65 00:05:08,620 --> 00:05:13,570 ‫So for the second record, we are getting the predicted class is 2. 66 00:05:13,990 --> 00:05:20,370 ‫And you can also see that the probability was maximum for this record, which is zero one two. 67 00:05:20,710 --> 00:05:24,250 ‫So second object for the third record. 68 00:05:24,960 --> 00:05:28,510 ‫We are getting the predicted class as one. 69 00:05:29,680 --> 00:05:37,800 ‫And in the probabilities also, you can see that the maximum probability correspond to position number 70 00:05:37,810 --> 00:05:38,080 ‫one. 71 00:05:39,950 --> 00:05:45,500 ‫Now, again, it's a little bit difficult to interpret results like this. 72 00:05:45,530 --> 00:05:46,670 ‫Nine to one. 73 00:05:47,240 --> 00:05:55,400 ‫So let's just call the label description instead of this, numerical labels. So will do the same thing we did 74 00:05:55,550 --> 00:05:56,060 ‫earlier. 75 00:05:56,810 --> 00:06:00,260 ‫We have already created a list called class Name. 76 00:06:00,770 --> 00:06:04,340 ‫We are just passing y pred into that list. 77 00:06:05,880 --> 00:06:12,810 ‫So if you run this instead of this labels, you will get the description of each record. 78 00:06:13,500 --> 00:06:20,750 ‫So now let's just confirm this by plotting the image of our first new dataset. 79 00:06:22,100 --> 00:06:24,650 ‫So the position is zero and we are using. 80 00:06:24,740 --> 00:06:27,230 ‫Iamshow plot this. 81 00:06:30,650 --> 00:06:39,260 ‫You can see this photo looks like an ankle boot, and we also predicted ankle boot for the second 82 00:06:39,260 --> 00:06:46,460 ‫record where the position is one and see that this is a pullover. 83 00:06:46,700 --> 00:06:48,610 ‫And we also predicted pullover. 84 00:06:50,830 --> 00:06:53,500 ‫For the third object here, the position is 2 85 00:06:58,090 --> 00:07:04,000 ‫You can see that this is a jeans or a trouser and we also predicted trouser 86 00:07:07,010 --> 00:07:15,710 ‫So that's how we predict with our model we can use evaluate to evaluate performance of model 87 00:07:15,830 --> 00:07:18,950 ‫on test set and to predict new values. 88 00:07:19,010 --> 00:07:26,420 ‫We can either use dot predict method or dot predict underscore classes method. Dot predict 89 00:07:26,500 --> 00:07:32,460 ‫method will give us the probability values assigned to each class for each record. 90 00:07:33,770 --> 00:07:41,720 ‫While predict underscore classes will give us details about the most probable class for that record 91 00:07:46,220 --> 00:07:55,880 ‫So that's how we create train and predict using mlp model. now just for a quick summary. 92 00:08:00,320 --> 00:08:03,670 ‫We started with importing and installing tensorflow. 93 00:08:05,960 --> 00:08:11,240 ‫Then we imported our fashion mnsist dataset with 10 categories. 94 00:08:14,070 --> 00:08:19,220 ‫Then we divided that dataset in to train validation and test set 95 00:08:20,390 --> 00:08:23,160 ‫And before that we also normalize our dataset 96 00:08:25,720 --> 00:08:32,740 ‫Now, after that, we created architecture for a our model this is where you mention the activation 97 00:08:32,740 --> 00:08:37,270 ‫functions, number of neurons you want in your layers. 98 00:08:40,580 --> 00:08:46,700 ‫After that, we compiled our model to give loss function optimizer and the metrics we want 99 00:08:46,700 --> 00:08:47,410 ‫to calculate. 100 00:08:47,870 --> 00:08:51,120 ‫And after that we trained our model using dot fit 101 00:08:51,290 --> 00:08:51,770 ‫Method 102 00:08:54,290 --> 00:09:02,670 ‫And after our model is trained, we can use dot art history method to get all the loss anf accuracy's 103 00:09:02,730 --> 00:09:04,520 ‫Values for each epoch. 104 00:09:05,610 --> 00:09:13,830 ‫After that, we can use dot elevate method to check accuracy scores on our test dataset, and we can 105 00:09:13,830 --> 00:09:20,280 ‫also predict values for new records using dot predict and dot predict underscore classes method.