1 00:00:00,020 --> 00:00:03,800 Hello everyone, and welcome to this session on evaluation and Testing. 2 00:00:03,830 --> 00:00:09,770 So so far we've built and trained our model and we have, um, reasonable performance. 3 00:00:09,770 --> 00:00:12,170 Let's go ahead and evaluate this model. 4 00:00:12,170 --> 00:00:14,180 So here we have model evaluate. 5 00:00:14,180 --> 00:00:17,960 And then we could pass in um our test data. 6 00:00:17,960 --> 00:00:20,990 So test data set we run that. 7 00:00:22,130 --> 00:00:26,960 The result we get now is about 93% for the test data set. 8 00:00:26,960 --> 00:00:30,410 So that's the accuracy for our test data set. 9 00:00:30,410 --> 00:00:32,090 Now let's dive into prediction. 10 00:00:32,090 --> 00:00:34,160 So we've trained our model. 11 00:00:34,160 --> 00:00:35,090 We've evaluated it. 12 00:00:35,090 --> 00:00:39,680 Let's suppose that we have some data drawn from our um test set. 13 00:00:39,680 --> 00:00:44,210 And we want to be able to say whether it's parasitized or uninfected. 14 00:00:44,210 --> 00:00:51,530 Now before moving on we'll define this parasite or not method, parasite or not or not. 15 00:00:51,530 --> 00:00:53,360 And then it takes an x. 16 00:00:53,360 --> 00:01:05,030 And if x if x is um less than 0.5, if x is less than 0.5, then what we want to do is return the string 17 00:01:05,030 --> 00:01:11,780 um, p that's parasitized and else we want to return the string. 18 00:01:11,900 --> 00:01:12,740 Um, you. 19 00:01:12,740 --> 00:01:13,520 So that's it. 20 00:01:13,520 --> 00:01:16,700 So it's either parasitized or uninfected. 21 00:01:16,700 --> 00:01:21,290 Now let's run that and then we have model. 22 00:01:21,410 --> 00:01:22,490 Or we could actually say. 23 00:01:22,640 --> 00:01:23,540 Net model. 24 00:01:23,540 --> 00:01:24,500 Let's define this as. 25 00:01:24,620 --> 00:01:25,730 Net model. 26 00:01:25,730 --> 00:01:29,300 Well since we already had model let's just leave it that way. 27 00:01:29,300 --> 00:01:34,610 But you could define it as later model to separate it from all other models we'll be working with in 28 00:01:34,610 --> 00:01:35,690 this course. 29 00:01:35,690 --> 00:01:41,960 So we have this model and the model we use make use of model predict does the predict method. 30 00:01:41,960 --> 00:01:44,510 And it takes in a test data. 31 00:01:44,510 --> 00:01:46,550 So we have test data set. 32 00:01:46,550 --> 00:01:49,160 We could take um a single one. 33 00:01:49,160 --> 00:01:55,760 And then we get the output now once note that here we have a batch of 32 elements. 34 00:01:55,760 --> 00:02:01,970 So if you output this you would have an output of of 32 elements. 35 00:02:01,970 --> 00:02:06,200 So after this we could pick out a single one out of this. 36 00:02:06,200 --> 00:02:08,150 Well let's just get the output first. 37 00:02:09,200 --> 00:02:11,840 As you could see, we have the 32 different outputs. 38 00:02:11,870 --> 00:02:17,060 Now we could pick out just the, the very first one so we could run that. 39 00:02:17,060 --> 00:02:20,270 And you see we have our output. 40 00:02:20,300 --> 00:02:24,350 Now we want to get it as parasitized or not. 41 00:02:24,350 --> 00:02:29,960 So parasitize parasite um or not. 42 00:02:29,960 --> 00:02:31,190 Parasite or not. 43 00:02:31,190 --> 00:02:32,780 And we get that. 44 00:02:32,780 --> 00:02:34,670 So let's print this out. 45 00:02:35,570 --> 00:02:39,020 Print that and there we go. 46 00:02:39,020 --> 00:02:42,200 So let's run this again and see what we get. 47 00:02:42,590 --> 00:02:46,280 You see we have p meaning that it's parasitized. 48 00:02:46,580 --> 00:02:50,300 Now the next step will be to visualize our predictions. 49 00:02:50,300 --> 00:02:56,570 That's to see what the model was supposed to predict and what the model actually predicted. 50 00:02:56,570 --> 00:02:59,900 So let's go for image label. 51 00:02:59,900 --> 00:03:03,410 We'll just do nine predictions or yeah nine images. 52 00:03:03,410 --> 00:03:09,920 So for that in test um data set, let's take a single batch. 53 00:03:10,340 --> 00:03:13,430 Now for I in range. 54 00:03:13,430 --> 00:03:15,290 Um nine. 55 00:03:15,890 --> 00:03:19,460 What we're going to do is we're going to have this subplot. 56 00:03:19,760 --> 00:03:26,480 So we have our subplot and it's three by three and then we increment I. 57 00:03:27,620 --> 00:03:31,040 Then from here we have plot image show. 58 00:03:31,070 --> 00:03:37,880 We have the image um, image I actually because we have a batch. 59 00:03:37,880 --> 00:03:39,290 So it's image I. 60 00:03:39,290 --> 00:03:46,880 And then once we have that image we are going to have a title which is going to be made of what the 61 00:03:46,880 --> 00:03:49,550 model was supposed to predict and what the model predicted. 62 00:03:49,550 --> 00:03:55,730 So we have this title which is made of, um, well, yeah, we have parasitized or not. 63 00:03:55,730 --> 00:03:58,220 So we just say parasitized. 64 00:03:58,520 --> 00:04:01,610 Well, the function name is parasite or not okay. 65 00:04:01,610 --> 00:04:04,400 So it's parasite or not. 66 00:04:04,400 --> 00:04:09,800 And then it takes in the output the label. 67 00:04:09,800 --> 00:04:13,760 So we have here label send that to numpy. 68 00:04:14,150 --> 00:04:15,410 NumPy. 69 00:04:15,440 --> 00:04:16,580 There we go. 70 00:04:16,580 --> 00:04:22,070 And then we pick I because this is that specific um index. 71 00:04:22,070 --> 00:04:27,080 And then now we have like that is after getting this label. 72 00:04:27,080 --> 00:04:28,850 So here is a label. 73 00:04:28,850 --> 00:04:32,240 The next thing we want to have is the what the model predicts. 74 00:04:32,240 --> 00:04:42,500 So we have this here then now we have plus then we could have this slash. 75 00:04:42,860 --> 00:04:45,260 Well plus there we go. 76 00:04:45,260 --> 00:04:55,370 And then we have again um parasite or not which takes in the model or the output of the model. 77 00:04:55,370 --> 00:04:58,190 So model predict we have model predict. 78 00:04:58,400 --> 00:05:00,920 And this model predict takes in the image. 79 00:05:00,920 --> 00:05:05,150 So here we have image I and then we. 80 00:05:05,150 --> 00:05:05,690 And that's it. 81 00:05:05,690 --> 00:05:06,800 So there we go. 82 00:05:06,800 --> 00:05:08,690 We have a text image. 83 00:05:08,690 --> 00:05:12,200 And then we have well let's do this way instead. 84 00:05:12,680 --> 00:05:16,370 When this takes in the image we have I and then we have zero. 85 00:05:16,370 --> 00:05:17,150 There we go. 86 00:05:17,150 --> 00:05:19,370 So that should be fine. 87 00:05:19,640 --> 00:05:30,500 Then finally we have let's take this this way we have plot axis and set that to off. 88 00:05:30,770 --> 00:05:31,820 And there we go. 89 00:05:31,820 --> 00:05:34,910 We could now run this and see what we obtain. 90 00:05:35,840 --> 00:05:41,630 You could see from here that, um, we have most of the predictions right out of the the nine predictions. 91 00:05:41,630 --> 00:05:43,310 We only have one wrong. 92 00:05:43,340 --> 00:05:44,090 That's this one. 93 00:05:44,090 --> 00:05:50,720 So essentially here, um, the model was supposed to predict, uh, this as parasitized, but it predicts 94 00:05:50,720 --> 00:05:52,010 this as uninfected.