1 00:00:00,240 --> 00:00:07,270 For now, we have been showing you this binary accuracy, false positives up to the AUC. 2 00:00:07,290 --> 00:00:13,300 What if we plot out the coefficient matrix and also the Rosie plot? 3 00:00:13,320 --> 00:00:19,320 So we'll impart cycle, learn and seaborne. 4 00:00:19,600 --> 00:00:29,250 Right here we have scikit learn and from scikit learn dot matrix, we are going to impart coefficient 5 00:00:29,250 --> 00:00:30,210 metrics. 6 00:00:30,510 --> 00:00:37,530 And then for Seaborn, we need to import Seaborn and SAS. 7 00:00:37,530 --> 00:00:38,480 So that's it. 8 00:00:38,490 --> 00:00:48,930 So yeah, we are importing I learn and Seaborn, which we'll both use in plotting out this confusion 9 00:00:48,930 --> 00:00:49,710 matrix. 10 00:00:49,710 --> 00:00:55,140 So let's run this and then we'll visualize our conversion metrics. 11 00:00:55,320 --> 00:01:03,960 So we'll have to get the level that's the true values of one of the outputs, and then we'll also get 12 00:01:03,960 --> 00:01:05,070 the predicted values. 13 00:01:05,070 --> 00:01:09,030 So yeah, we want to get the labels and the predicted values. 14 00:01:09,060 --> 00:01:11,430 Now let's start with the labels. 15 00:01:11,430 --> 00:01:13,650 So yeah, we're going to have labels. 16 00:01:13,770 --> 00:01:17,850 We have this list, there we go. 17 00:01:17,850 --> 00:01:31,860 And then for X, Y, or let's say for, for x, Y in the test data set, we're going to take this as 18 00:01:31,890 --> 00:01:33,570 non py iterator. 19 00:01:33,570 --> 00:01:44,280 So for x, Y and this we have the x and the Y, we have that and then we append every output to this 20 00:01:44,280 --> 00:01:44,790 labels. 21 00:01:44,790 --> 00:01:50,070 So here we have labels that append y and we run that. 22 00:01:50,160 --> 00:01:51,150 So that's fine. 23 00:01:51,150 --> 00:01:53,430 We could now print out labels. 24 00:01:53,430 --> 00:01:54,630 What do we get? 25 00:01:55,140 --> 00:01:56,100 There we go. 26 00:01:56,250 --> 00:01:57,720 You see, we have this labels. 27 00:01:57,720 --> 00:02:07,440 Now let's convert it into a simpler form where we just have only this values in here and not this arrays 28 00:02:07,440 --> 00:02:08,760 which we have here. 29 00:02:08,760 --> 00:02:16,170 So in order to get that, we have labels will be equal labels or rather will be equal an array. 30 00:02:16,320 --> 00:02:18,690 And in here we're going to create this list. 31 00:02:18,690 --> 00:02:25,110 So in this list here, for every element of this labels list. 32 00:02:25,110 --> 00:02:36,470 So for here, for I in labels, we are going to do I the zero. 33 00:02:38,230 --> 00:02:38,860 Index. 34 00:02:38,860 --> 00:02:42,220 So we're going to take always this element right here. 35 00:02:42,370 --> 00:02:45,460 So that said, we have that and now we have levels. 36 00:02:45,460 --> 00:02:48,760 Let's print out a levels after doing this transformation. 37 00:02:49,390 --> 00:02:50,260 There we go. 38 00:02:50,290 --> 00:02:51,750 Here is our levels now. 39 00:02:51,760 --> 00:02:54,670 So we have that fine could take that off. 40 00:02:56,110 --> 00:03:01,810 We have this error actually because we tried we've converted this levels already and then we're trying 41 00:03:01,810 --> 00:03:02,680 to re convert them. 42 00:03:02,680 --> 00:03:06,250 So let's command this right here. 43 00:03:06,250 --> 00:03:09,610 We command that and run this and we have all levels. 44 00:03:09,700 --> 00:03:11,560 OC Now we have those levels. 45 00:03:11,560 --> 00:03:15,490 We could move ahead and get the predicted values. 46 00:03:15,490 --> 00:03:22,350 So to get this predicted values, we have our low net model Loonette learned model. 47 00:03:22,360 --> 00:03:28,960 There we go dot predict what are we going to pass in this predict method is our input. 48 00:03:28,960 --> 00:03:35,800 So we have the input and then right here we add this input list and then we do same as we did with the 49 00:03:35,800 --> 00:03:36,280 labels. 50 00:03:36,280 --> 00:03:37,510 So we have input. 51 00:03:37,510 --> 00:03:41,300 Then append X, which is this input here. 52 00:03:41,320 --> 00:03:45,160 Now we run this and then let's add a cell. 53 00:03:45,160 --> 00:03:49,420 So we add this code cell and then we're going to pre process our input. 54 00:03:49,420 --> 00:03:54,950 So from this we're just going to print out the input to see what it gives us, what kind of shape we 55 00:03:54,950 --> 00:03:56,140 are having right here. 56 00:03:56,290 --> 00:03:58,120 Let's put non power array. 57 00:03:58,870 --> 00:04:01,120 Wow, that's loading shape. 58 00:04:01,870 --> 00:04:02,830 Then we run this. 59 00:04:02,830 --> 00:04:07,930 All the cell right here is what we get as shape. 60 00:04:07,930 --> 00:04:12,040 So we have to take this off, this one off here. 61 00:04:12,040 --> 00:04:18,700 And to do that you see we would have the input will select this first index. 62 00:04:18,700 --> 00:04:23,080 So we take that and then for this next we take only one. 63 00:04:23,080 --> 00:04:27,280 So we're going to chop this off and then we have the rest. 64 00:04:27,280 --> 00:04:28,250 So that's it. 65 00:04:28,270 --> 00:04:35,920 Now we run this again and we should have put a shape here, so we run that again. 66 00:04:37,660 --> 00:04:38,860 And there we go. 67 00:04:38,860 --> 00:04:40,570 So this is what we expect. 68 00:04:40,600 --> 00:04:47,560 Now we've gotten the labels and then we are now ready to get the predictions. 69 00:04:48,070 --> 00:04:53,260 So here we have this input or take this and then we pass it in. 70 00:04:53,260 --> 00:04:56,200 Here we pass our input and we get our predictions. 71 00:04:56,200 --> 00:05:02,590 Let's print out the predicted here or yeah, let's print out a predicted and see what we get. 72 00:05:02,830 --> 00:05:08,590 So what we get right here, we could print out a shape to better understand what this is. 73 00:05:08,920 --> 00:05:11,500 And as you could see, we have this. 74 00:05:11,500 --> 00:05:15,730 So how do we take this off or how do we make it look like the labels? 75 00:05:16,600 --> 00:05:19,590 We could simply take this off. 76 00:05:19,600 --> 00:05:20,860 So we have. 77 00:05:20,860 --> 00:05:25,150 We select the first, and then for this one, we take it off. 78 00:05:25,150 --> 00:05:26,770 So we run that again. 79 00:05:27,520 --> 00:05:28,690 And there we go. 80 00:05:28,690 --> 00:05:36,670 So now we could do this, take off the shape, and then it should look like the labels we've had here. 81 00:05:37,630 --> 00:05:38,370 And that's it. 82 00:05:38,380 --> 00:05:40,030 So now we've gotten this. 83 00:05:40,990 --> 00:05:49,220 Our next step is to use the confusion matrix from circle and matrix and produce our conversion matrix. 84 00:05:49,240 --> 00:05:55,960 Now let's comment this sections first so you will pass in the labels with pass in the predicted and 85 00:05:55,960 --> 00:05:57,630 then we'll specifying this threshold. 86 00:05:57,640 --> 00:06:03,880 So what we're seeing here is all the values greater than the threshold considered uninfected and all 87 00:06:03,880 --> 00:06:05,350 values less than or equal. 88 00:06:05,350 --> 00:06:10,240 The threshold is going to be considered as containing the parasite. 89 00:06:10,330 --> 00:06:13,360 And then we're going to have this in this conversion matrix here. 90 00:06:13,360 --> 00:06:14,470 So we run that. 91 00:06:14,680 --> 00:06:20,920 And then as you could see, we have our coefficient matrix, which shows us the number of true positives, 92 00:06:20,920 --> 00:06:24,930 true negatives, false positives and false negatives. 93 00:06:24,940 --> 00:06:31,720 Getting back to our model evaluation, we'll see exactly which of those values are the true positives, 94 00:06:31,720 --> 00:06:34,330 false positives, true negatives or false negatives. 95 00:06:34,330 --> 00:06:36,580 So let's just copy this out. 96 00:06:36,610 --> 00:06:42,750 Your if we take this up and we shall see that below. 97 00:06:42,760 --> 00:06:49,810 So here we have this confusion matrix, which we've just seen here is it right here, confusion matrix 98 00:06:49,810 --> 00:06:50,500 right here. 99 00:06:51,190 --> 00:06:52,990 And then let's paste this out here. 100 00:06:52,990 --> 00:07:00,940 So we have the number of true positives this one matches and then number of true negatives. 101 00:07:00,940 --> 00:07:02,560 1298. 102 00:07:02,590 --> 00:07:03,700 Yeah, this is closer. 103 00:07:03,700 --> 00:07:10,360 So what cyclone gives us is quite similar to what we had with TensorFlow, but they're not exactly the 104 00:07:10,360 --> 00:07:10,870 same. 105 00:07:10,870 --> 00:07:13,810 So this tells us that this is a number of true negatives. 106 00:07:13,810 --> 00:07:18,220 This is the number of no, this is the number of true negatives right here. 107 00:07:18,220 --> 00:07:19,810 So here we have two negatives. 108 00:07:19,810 --> 00:07:22,060 Here we have true positives. 109 00:07:22,150 --> 00:07:27,760 Then here we have false positives and then here we have false negatives. 110 00:07:28,190 --> 00:07:36,880 We'll then see how modifying this threshold will either reduce the false negatives or reduce the false 111 00:07:36,880 --> 00:07:37,660 positives. 112 00:07:39,340 --> 00:07:42,790 So let's have, say, 0.25 right here. 113 00:07:42,790 --> 00:07:45,790 We run that again and there we go. 114 00:07:46,330 --> 00:07:47,320 Let's print this. 115 00:07:47,320 --> 00:07:51,640 So we see that we now have a number of false negatives reduced. 116 00:07:51,640 --> 00:07:57,520 When we reduce this threshold, we now go back and then take this to 0.75. 117 00:07:57,520 --> 00:07:58,690 We run that again. 118 00:07:59,170 --> 00:08:00,940 You see that this is now reduced. 119 00:08:00,940 --> 00:08:07,120 So here we have a reduced or an increased number of false negatives and a reduced number of false positives 120 00:08:07,120 --> 00:08:10,150 when this threshold is increased to 0.75. 121 00:08:10,870 --> 00:08:12,100 So that's it. 122 00:08:12,100 --> 00:08:14,410 Let's now print out. 123 00:08:14,410 --> 00:08:14,830 All right. 124 00:08:14,860 --> 00:08:21,570 Let's now plot out this more elegantly looking figure with the Seaborn Library. 125 00:08:21,580 --> 00:08:22,540 So here we go. 126 00:08:22,540 --> 00:08:24,820 We run that and that's it. 127 00:08:24,820 --> 00:08:31,780 So here we have our same conversion matrix, but more elegantly plotted. 128 00:08:31,990 --> 00:08:37,270 Let's take it back to 0.5 here and have that. 129 00:08:37,270 --> 00:08:38,590 Okay, so that's it. 130 00:08:39,580 --> 00:08:45,610 Now, you notice that when we're trying to say reduce more false negatives or number of false positives, 131 00:08:45,610 --> 00:08:49,390 what we're doing is we are just picking up some values here. 132 00:08:49,390 --> 00:08:55,150 Like we could pick up 0.2 and then we see this effect and we could do the same for 0.25. 133 00:08:55,150 --> 00:08:58,960 You see how this takes us to 58 and there's 230. 134 00:08:58,960 --> 00:09:04,240 We could continue doing this until we get our best results. 135 00:09:04,240 --> 00:09:09,680 But then this isn't very efficient since we just trying out different values. 136 00:09:09,700 --> 00:09:16,090 Now, the way we could do this more efficiently is by working with our C plots where we'll be able to 137 00:09:16,120 --> 00:09:20,050 choose a threshold in a more efficient manner using the plots. 138 00:09:20,050 --> 00:09:25,960 And so we'll be able to reduce the number of false negatives or the number of false positives without 139 00:09:25,960 --> 00:09:29,950 having to try out all this trash values manually.