Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated: 1 00:00:00,300 --> 00:00:02,690 Hello and welcome to this new tutorial. 2 00:00:02,730 --> 00:00:09,280 So here we are ready for the big first step of the training that is updating the weight of the neural 3 00:00:09,300 --> 00:00:11,310 network of the discriminator. 4 00:00:11,310 --> 00:00:15,300 This big first step we're going to tackle it in three subsets. 5 00:00:15,330 --> 00:00:20,220 The first step is to train the discriminator with a real image of the data set. 6 00:00:20,220 --> 00:00:26,010 The second step is to train the discriminator with this time a fake image generated by the generator. 7 00:00:26,220 --> 00:00:32,880 And finally will back propagate the total error which will be the sum of the errors of these two previous 8 00:00:32,880 --> 00:00:33,630 trainings. 9 00:00:33,840 --> 00:00:35,660 So now I have a question for you. 10 00:00:35,850 --> 00:00:43,170 Why do we have to do a training of the discriminator with both a real image of the dataset and a fake 11 00:00:43,170 --> 00:00:45,480 image generated by the generator. 12 00:00:45,750 --> 00:00:53,850 And the answer is that simply because we want to train the discriminator to see and understand what's 13 00:00:53,850 --> 00:01:00,180 real and what's fake and therefore to make the discriminator understand that well we need to give him 14 00:01:00,450 --> 00:01:06,720 the two different ground tricks we need to give him the ground truth of what's real and the ground truth 15 00:01:06,840 --> 00:01:08,200 of what's fake. 16 00:01:08,400 --> 00:01:13,230 So the ground truth of what's real is of course the real image and the ground truth of what's fake is 17 00:01:13,230 --> 00:01:16,520 of course the fake image generated by the generator. 18 00:01:16,530 --> 00:01:24,330 So if you understand that well it will be very easy these two subsets will appear very natural to you. 19 00:01:24,480 --> 00:01:28,540 And so now if you're ready let's do these three steps. 20 00:01:28,740 --> 00:01:33,990 But before we start with the first obstacle is training the discriminator with a real image of the data 21 00:01:33,990 --> 00:01:36,460 set to train it to understand what's real. 22 00:01:36,630 --> 00:01:43,650 Well we need to initialize the gradient of the discriminator with respect to the weight to zero and 23 00:01:43,650 --> 00:01:50,850 to do this it's very simple we take the neural network of the discriminator which we called Nedney then 24 00:01:50,850 --> 00:01:59,380 we added that and then we use the zero underscore grad function and that will automatically initialize 25 00:01:59,380 --> 00:02:02,200 to zero the gradients with respect to the weights. 26 00:02:02,210 --> 00:02:09,300 So that's done and now we can move on to the first subset that is training the discriminator with a 27 00:02:09,300 --> 00:02:11,340 real image of the day set. 28 00:02:11,400 --> 00:02:15,300 All right so the first thing we need to do is get the real images. 29 00:02:15,300 --> 00:02:17,250 Why do I say real images. 30 00:02:17,250 --> 00:02:23,220 That's because we're going to get in fact a mini batch of real images and that's because a neural network 31 00:02:23,370 --> 00:02:30,600 actually accept as inputs some many batches of single inputs like single images and therefore we have 32 00:02:30,600 --> 00:02:36,690 to work with many batches but that's perfect because we already iterated through some many batches that 33 00:02:36,690 --> 00:02:38,630 we got with a loader. 34 00:02:38,700 --> 00:02:40,530 So we already have these mini batches. 35 00:02:40,560 --> 00:02:47,720 And therefore as you will see right now it will be very easy to get these in any batch of real images. 36 00:02:48,180 --> 00:02:54,690 So this input many batch I'm going to call it real and it's actually going to be the first element of 37 00:02:54,990 --> 00:02:56,740 our mini batch data. 38 00:02:56,820 --> 00:03:02,080 Right now we're dealing with a specific mini batch which is data and data is composed of two elements. 39 00:03:02,190 --> 00:03:07,260 The first elements are the real images themselves and the second element are the labels. 40 00:03:07,260 --> 00:03:09,430 But we don't really care about the labels right now. 41 00:03:09,480 --> 00:03:15,710 So I'm just getting the first elements and to do that to get the first element of technique is to add 42 00:03:15,700 --> 00:03:22,920 a come here and then an underscore to specify We actually don't care about the second element and then 43 00:03:23,400 --> 00:03:25,250 equals data. 44 00:03:25,260 --> 00:03:26,610 All right so perfect. 45 00:03:26,610 --> 00:03:34,260 We have our input but it is not yet an accepted input of a neural network in by torche by torche neural 46 00:03:34,260 --> 00:03:42,060 networks only accept the input in Torch variables and I remind that a torch very well is a highly advanced 47 00:03:42,060 --> 00:03:46,350 variable that contains both a sensor and a gradient. 48 00:03:46,350 --> 00:03:50,720 Right now we have to sensor because real is actually a sensor of images. 49 00:03:50,850 --> 00:03:57,480 But we need to wrap it into a torch very well to associate it with a gradient and to do this we're going 50 00:03:57,480 --> 00:04:02,850 to introduce a new varry Walters we're going to call input because this will be the input of the neural 51 00:04:02,850 --> 00:04:10,020 network and this input is going to be an object of the variable class which will take his arguments 52 00:04:10,250 --> 00:04:18,330 are real input images in them in batch and now are input images are not only in too many Bache but also 53 00:04:18,390 --> 00:04:19,750 in a torch variable. 54 00:04:19,770 --> 00:04:25,140 So now we're allowed to feed the neural network with this input. 55 00:04:25,200 --> 00:04:27,630 But before that we will get the target. 56 00:04:27,870 --> 00:04:33,130 And now the line of good I'm going to type is very very important at this stage. 57 00:04:33,340 --> 00:04:36,520 You're going to try to guess what the target is going to be. 58 00:04:36,540 --> 00:04:42,070 We're going to have to target for the different training this training and the following training. 59 00:04:42,180 --> 00:04:45,730 And so try to guess what exactly these targets are going to be. 60 00:04:45,840 --> 00:04:49,260 So I'm going to introduce here a new target. 61 00:04:49,290 --> 00:04:52,480 So now according to you what is this target going to be. 62 00:04:52,830 --> 00:04:59,830 Well since we are training the discriminator with a real image of the data set to train him to under 63 00:05:00,070 --> 00:05:03,720 one and see what's real what is a real image. 64 00:05:03,890 --> 00:05:10,790 Well for each of the real image of the million batch Well we need to set the target to one white one. 65 00:05:11,000 --> 00:05:15,230 That's because remember zero corresponds to rejection. 66 00:05:15,380 --> 00:05:20,720 The image is rejected by the discriminator and one corresponds to acceptance. 67 00:05:20,750 --> 00:05:26,570 The image is accepted by the discriminator and therefore we need to set the target to one because we 68 00:05:26,570 --> 00:05:33,080 need to specify to the discriminator that the ground truth is actually one to ground truth is the image 69 00:05:33,170 --> 00:05:33,830 is real. 70 00:05:33,890 --> 00:05:35,630 So the image gets a 1. 71 00:05:35,930 --> 00:05:41,390 And that's why right now we're going to create a torch sensor that is going to have the size of the 72 00:05:41,390 --> 00:05:42,090 mini match. 73 00:05:42,260 --> 00:05:48,790 And that is going to be composed of only once we will have a one for each of the input image of the 74 00:05:48,790 --> 00:05:49,710 mini batch. 75 00:05:49,730 --> 00:05:50,270 So there we go. 76 00:05:50,270 --> 00:05:51,200 Let's do this. 77 00:05:51,200 --> 00:05:58,510 We need to take the torch library and then very simply we have a great function that is called. 78 00:05:58,730 --> 00:06:01,880 And that will create this sensor only once. 79 00:06:01,880 --> 00:06:08,030 And as you might guess what we need input in this one function is actually the size of the sensor that 80 00:06:08,030 --> 00:06:15,930 is how many ones do we want and we want as many ones as they are real images in the real input mini 81 00:06:15,980 --> 00:06:16,600 batch. 82 00:06:16,850 --> 00:06:20,140 So how can we get this size of the input batch. 83 00:06:20,360 --> 00:06:27,590 Well we just need to take our input many batch and then add that and then add size with some parenthesis 84 00:06:28,040 --> 00:06:33,740 input that size contains the size of the mini batch that is contains the number of real images of the 85 00:06:33,740 --> 00:06:38,560 input many batch and therefore also the number of ones the target should contain. 86 00:06:38,810 --> 00:06:41,160 But as you notice I said contains. 87 00:06:41,240 --> 00:06:49,230 And to get the actual number we need to take the first index of this element in that size which is zero. 88 00:06:49,400 --> 00:06:54,810 So input that size of index 0 will return you the size of the mean bitch. 89 00:06:55,190 --> 00:06:55,840 Perfect. 90 00:06:55,890 --> 00:06:59,270 And now question Are we allowed to move onto the next step. 91 00:06:59,270 --> 00:07:00,280 No we're not. 92 00:07:00,290 --> 00:07:04,650 The reason is we have to wrap the target in a torch viable. 93 00:07:04,660 --> 00:07:09,920 Indeed we're going to compute some gradients of the target as well and therefore we need to attach this 94 00:07:09,920 --> 00:07:13,940 target sensor to a gradient inside a torch variable. 95 00:07:14,000 --> 00:07:22,400 So I'm going to take my variable class again and I'm going to put everything inside some parenthesis 96 00:07:22,730 --> 00:07:29,130 so that target becomes an object of the variable class taking his argument this torch and sort of once 97 00:07:29,940 --> 00:07:30,570 right. 98 00:07:30,600 --> 00:07:33,210 And now we have the inputs and the target. 99 00:07:33,260 --> 00:07:34,530 So we know what to do next. 100 00:07:34,580 --> 00:07:36,490 We need to get the outputs. 101 00:07:36,710 --> 00:07:40,310 So let's do this let's get the output well to get the output. 102 00:07:40,340 --> 00:07:42,610 Actually pretty fun and very simple. 103 00:07:42,650 --> 00:07:49,880 We'll first introduce a new variable for the output output and we are going to call our neural network 104 00:07:50,410 --> 00:07:57,830 of the discriminator so needy and we are going to feed this neural network with of course the inputs 105 00:07:58,340 --> 00:08:02,480 the input which is a torch Vrable of a mini batch of real images. 106 00:08:02,480 --> 00:08:07,260 And so inside here I just need to input Well input. 107 00:08:07,310 --> 00:08:08,280 All right. 108 00:08:08,520 --> 00:08:16,850 So through the main metal module that forward propagates the real input images of the mini match inside 109 00:08:16,940 --> 00:08:22,640 the neural network of the discriminator to get for each of these real inputs images the prediction of 110 00:08:22,640 --> 00:08:26,170 the discriminator whether they should be accepted or not. 111 00:08:26,300 --> 00:08:32,330 So I remind that for each of these real images the output that is a prediction is a number between 0 112 00:08:32,330 --> 00:08:38,360 and 1 and a number close to zero means that the discriminator will reject the image and a number close 113 00:08:38,360 --> 00:08:42,130 to 1 means that the discriminator will accept the image. 114 00:08:42,170 --> 00:08:46,160 So that's the discriminating number and it's a number between 0 and 1. 115 00:08:46,160 --> 00:08:47,200 All right perfect. 116 00:08:47,210 --> 00:08:51,540 And now that we have target full of ones and the outputs. 117 00:08:51,590 --> 00:08:53,110 Well guess what we're going to get. 118 00:08:53,270 --> 00:08:55,460 Well of course we're going to get the error. 119 00:08:55,640 --> 00:09:02,570 The first error coming from this first training of the discriminator with the real image ground troops. 120 00:09:03,050 --> 00:09:07,990 And so this specific first arrow we're going to call it e r r d. 121 00:09:08,180 --> 00:09:13,040 Because we're also going to have an area for the generator but much later that is in the second big 122 00:09:13,040 --> 00:09:14,440 step of the training. 123 00:09:14,440 --> 00:09:15,590 E r r d. 124 00:09:15,620 --> 00:09:20,020 And since this area corresponds to the real ground truth. 125 00:09:20,210 --> 00:09:24,440 Well I'm going to add here and underscore and real. 126 00:09:24,440 --> 00:09:25,320 All right. 127 00:09:25,400 --> 00:09:28,780 We are the real and now to get this error. 128 00:09:28,820 --> 00:09:32,520 Well what should I take I should take my criterion. 129 00:09:32,630 --> 00:09:36,110 We have a new object that will compute the last error for us. 130 00:09:36,110 --> 00:09:39,820 It will compute the last area between the output and the target. 131 00:09:39,890 --> 00:09:40,940 So that's perfect. 132 00:09:41,000 --> 00:09:44,490 And as you might guess inside this concern we need two inputs. 133 00:09:44,570 --> 00:09:48,700 First the output and second the target. 134 00:09:48,830 --> 00:09:49,670 And there we go. 135 00:09:49,700 --> 00:09:56,030 We have our first last error of the discriminator the one corresponding to the training of the discriminator 136 00:09:56,300 --> 00:10:02,950 with the real images to train it to understand to recognize what's real real images perfect. 137 00:10:02,950 --> 00:10:08,790 And now we're ready to move on to the second step of the training of the discriminator. 138 00:10:08,850 --> 00:10:13,890 It is the training with this time a fake image generated by the generator. 139 00:10:13,920 --> 00:10:19,590 So this time we're training the discriminator to see and understand what's fake. 140 00:10:19,590 --> 00:10:21,960 That is to recognize fake images. 141 00:10:22,170 --> 00:10:24,680 So we're going to do the same process as we did here. 142 00:10:24,680 --> 00:10:31,020 We're going to get first the input then the target then the output then this will generate a loss which 143 00:10:31,020 --> 00:10:37,170 will call our already underscore fake and we'll be done with this second step. 144 00:10:37,260 --> 00:10:44,030 And then finally we'll get the all error as to some of these two errors are already real and are fake. 145 00:10:44,250 --> 00:10:48,370 And then we'll do the big back propagation of this total error. 146 00:10:48,480 --> 00:10:51,470 Back inside the neural network of the discriminator. 147 00:10:51,480 --> 00:10:56,940 So let's do this let's tackle this second training with the ground truth of the vague images and let's 148 00:10:56,940 --> 00:10:59,670 start right now by getting the inputs. 149 00:10:59,670 --> 00:11:02,370 So it's actually not that direct according to you. 150 00:11:02,460 --> 00:11:09,070 How are we going to get the input which this time should be a mini batch of fake images. 151 00:11:09,300 --> 00:11:15,510 Well if you remember what we did here when defining the architecture of the generator. 152 00:11:15,840 --> 00:11:23,970 Well remember that the first inverted convolution takes as input a random vector of size 100. 153 00:11:24,240 --> 00:11:29,410 And that I remind is because the generator is like an inverted CNN. 154 00:11:29,490 --> 00:11:34,890 And since CNN takes as input some images and returns a flattened vector of one dimension. 155 00:11:35,010 --> 00:11:40,860 Well this inverted CNN of the generator will do exactly the opposite it will take as input a vector 156 00:11:41,040 --> 00:11:47,220 one dimension and we'll return the images or return some fake images and since we specify here that 157 00:11:47,220 --> 00:11:49,800 the input vector is of size 100. 158 00:11:49,980 --> 00:11:55,450 Well right now we are exactly going to create a vector of size 100. 159 00:11:55,620 --> 00:12:01,950 This will be a random vector and that will represent some noise and then we'll feed the neural network 160 00:12:01,950 --> 00:12:07,680 of the generator with this random vector and it will return some fake images. 161 00:12:07,770 --> 00:12:13,710 Of course at the beginning it will return some images that look like nothing but over the books we will 162 00:12:13,710 --> 00:12:19,920 update the weights so that the images look like something that is look like some real images. 163 00:12:20,160 --> 00:12:22,620 But before doing that that's actually what we'll be doing. 164 00:12:22,620 --> 00:12:28,440 The second step before doing that we need to train the discriminator to recognize what's fake. 165 00:12:28,710 --> 00:12:33,120 So let's do this let's make this random vector of size 100. 166 00:12:33,270 --> 00:12:39,420 And as we just said we're going to call it noise it actually represents some noise being a random input 167 00:12:39,420 --> 00:12:40,050 vector. 168 00:12:40,170 --> 00:12:46,920 So to create with by toward a vector of random values of a specific size Well it's actually very simple 169 00:12:46,930 --> 00:12:48,570 We have a function for this. 170 00:12:48,610 --> 00:12:55,050 This function we get it from the torch library of course and the name of this function is round and 171 00:12:55,650 --> 00:12:56,220 runs. 172 00:12:56,310 --> 00:13:01,190 And now inside this runs and function we need to input several arguments. 173 00:13:01,290 --> 00:13:06,830 The first one is going to be the batch size which I remind is 64. 174 00:13:06,870 --> 00:13:09,600 So that's the first argument we to input here. 175 00:13:09,720 --> 00:13:15,870 And therefore I'm going to copy this batch size which I got for my one function. 176 00:13:15,870 --> 00:13:16,610 There we go. 177 00:13:16,620 --> 00:13:19,630 Copy and paste. 178 00:13:19,680 --> 00:13:25,320 So we need to input this first argument of the batch size because we're not only going to create one 179 00:13:25,500 --> 00:13:32,250 random vector of size 100 we're going to create of course a mini batch of random vectors of size 100. 180 00:13:32,520 --> 00:13:36,630 So that then we can get a mini batch of fake images. 181 00:13:36,840 --> 00:13:43,470 So input size corresponds to the size of the batch and then the second argument will be well the number 182 00:13:43,470 --> 00:13:50,190 of elements we want in this vector and that is 100 because we specify in the architecture of the generator 183 00:13:50,520 --> 00:13:57,840 that the input vector should be of size 100 and then we're going to add to arguments which are going 184 00:13:57,840 --> 00:14:06,300 to be one and one and that is just to give to these random vectors some vague dimensions that will correspond 185 00:14:06,300 --> 00:14:12,150 to a future map that is in fact instead of having 100 values in the vector. 186 00:14:12,210 --> 00:14:19,260 It's like we will have 100 feature maps of size one by one meaning each of the 100 feature maps will 187 00:14:19,260 --> 00:14:21,690 be a matrix of size one by one. 188 00:14:21,840 --> 00:14:22,360 Great. 189 00:14:22,380 --> 00:14:23,960 So now we get our noise. 190 00:14:24,030 --> 00:14:26,240 So are we ready to move on to the next step. 191 00:14:26,340 --> 00:14:28,510 Well no no we're not. 192 00:14:28,530 --> 00:14:36,150 Because always for the same reason we have to wrap this mini batch of random vectors inside a variable. 193 00:14:36,150 --> 00:14:36,960 Why is that. 194 00:14:36,960 --> 00:14:42,750 That's because this noise here is going to be the input of the new one that work of the generator and 195 00:14:42,750 --> 00:14:46,750 neural networks and pite which only accept torch variables. 196 00:14:46,810 --> 00:14:48,090 So let's do that quickly. 197 00:14:48,150 --> 00:14:57,390 Let's get our variable class and put this torche tensor of random vectors inside the variable so that 198 00:14:57,390 --> 00:15:03,090 now noise becomes an object of this variable class containing this tensor. 199 00:15:03,330 --> 00:15:03,760 Right. 200 00:15:03,780 --> 00:15:06,840 And now we are allowed to move on to the next step. 201 00:15:06,840 --> 00:15:08,780 So now according to you what is the next step. 202 00:15:08,880 --> 00:15:12,190 Well obviously the next step is to get what we want. 203 00:15:12,210 --> 00:15:14,090 That is this new ground truth. 204 00:15:14,120 --> 00:15:19,440 We're looking for which of the fake images and the fake images we can now get them because we have the 205 00:15:19,440 --> 00:15:22,830 right inputs of the neural network of the generator. 206 00:15:22,830 --> 00:15:25,970 So let's get them we're going to call them fake. 207 00:15:26,160 --> 00:15:31,690 But keep in mind that this will represent a mini batch of fake images a fake is the name of the match 208 00:15:32,080 --> 00:15:33,760 so fake equals. 209 00:15:33,960 --> 00:15:40,260 Then very simply again we're going to take the neural network of the generator to which we're going 210 00:15:40,260 --> 00:15:40,980 to feed. 211 00:15:41,280 --> 00:15:48,090 Well this noise mini batch containing the random vectors and therefore will get a mini batch of the 212 00:15:48,090 --> 00:15:51,380 same size containing some fake images. 213 00:15:51,540 --> 00:15:52,100 Awesome. 214 00:15:52,170 --> 00:15:58,510 So now it is important to understand that this is the new input compared to this one. 215 00:15:58,530 --> 00:16:01,930 This was the input containing the ground truth of the real images. 216 00:16:02,070 --> 00:16:08,250 And this is a new input containing the ground truth of the fake images and therefore very quickly will 217 00:16:08,250 --> 00:16:09,980 get the new output. 218 00:16:10,020 --> 00:16:17,130 That is the output we'll get after feeding our discriminator with these new inputs fake images. 219 00:16:17,280 --> 00:16:20,420 But before we get this outputs we need to get the target. 220 00:16:20,610 --> 00:16:27,120 And now that's crucial to understand that this time the target is going to be a new kind of target. 221 00:16:27,150 --> 00:16:28,690 And so what is it going to be. 222 00:16:28,890 --> 00:16:34,350 Well this time since we are training the discriminator with some fake images we want to train it to 223 00:16:34,350 --> 00:16:40,560 recognize the fake images we want to train it to recognize what's fake and therefore the target should 224 00:16:40,560 --> 00:16:47,220 be the rejection of the images and the rejection of the images correspond to 0 0 means that the image 225 00:16:47,280 --> 00:16:53,990 is rejected by the generator and therefore the target should be this time a sensor full of zeroes. 226 00:16:54,180 --> 00:17:00,960 And so what I'm going to do I'm going to take this again because it's the same we're going to wrap it 227 00:17:00,960 --> 00:17:07,030 into a viable but instead of using the ones function to get sensor full of ones. 228 00:17:07,200 --> 00:17:14,750 Well I'm going to replace it by guess what zeros we have these very practical functions intuitive to 229 00:17:14,750 --> 00:17:20,790 remember and by torche this time we have the zero's function that will return a sensor full of zeros 230 00:17:21,030 --> 00:17:26,240 and the number of zeroes will be the size of the batch and put that size of index zero. 231 00:17:26,520 --> 00:17:27,180 Perfect. 232 00:17:27,210 --> 00:17:28,830 We have our new target. 233 00:17:28,830 --> 00:17:31,130 And so now let's get the outputs. 234 00:17:31,170 --> 00:17:36,840 So I'm going to introduce the variable for this output which I'm going to call outputs again and to 235 00:17:36,840 --> 00:17:38,970 get my new output. 236 00:17:39,150 --> 00:17:45,740 Well I'm going to take the neural network of the discriminator because we're still training discriminator 237 00:17:46,200 --> 00:17:54,180 and I'm going to feed this do all that work with of course the fake images and then for each of the 238 00:17:54,180 --> 00:18:00,750 fake images of the fake mini batch Well you'll get a prediction which is a discriminating number between 239 00:18:00,750 --> 00:18:01,680 0 and 1. 240 00:18:01,830 --> 00:18:07,170 And again if it is close to zero the discriminator will reject the image and if it is close to one it 241 00:18:07,170 --> 00:18:08,820 will accept it. 242 00:18:08,880 --> 00:18:09,950 All right perfect. 243 00:18:09,990 --> 00:18:13,190 But we can do something actually better here. 244 00:18:13,230 --> 00:18:15,210 We can save up some memory. 245 00:18:15,210 --> 00:18:22,220 Remember that fake is a total variable because the output of a PI torche neural network is also a torch 246 00:18:22,240 --> 00:18:27,680 variable and therefore it contains not only the tensor of the predictions the discriminating numbers 247 00:18:27,690 --> 00:18:30,600 between 0 and 1 but also the gradients. 248 00:18:30,810 --> 00:18:36,330 But actually we're not going to use this gradient after back propagating the error back inside the neural 249 00:18:36,330 --> 00:18:39,060 network and when applying stochastic great in the sense. 250 00:18:39,060 --> 00:18:46,200 So what we can do now is actually detach the gradient of this fake torche viable that will save up some 251 00:18:46,200 --> 00:18:48,850 memory and that will speed up the computations. 252 00:18:48,870 --> 00:18:53,820 And trust me we want to do this because the training is going to take quite a while. 253 00:18:53,830 --> 00:18:59,780 So want to savor as much memory as possible and get the fastest computations as possible. 254 00:18:59,810 --> 00:19:05,600 So we're going to detach the gradient here of this toward horrible and to do this we are done here and 255 00:19:05,600 --> 00:19:13,250 then detach and then some parenthesis we absolutely don't care of the gradient of the output with respect 256 00:19:13,250 --> 00:19:14,850 to the weight of the generator. 257 00:19:14,850 --> 00:19:19,060 It will not be part of the considerations in stochastic gradient descent. 258 00:19:19,070 --> 00:19:23,610 All right so now we have the time yet we have the output. 259 00:19:23,720 --> 00:19:29,750 So guess what we're ready to have we are ready to have the new error corresponding to the training of 260 00:19:29,750 --> 00:19:32,120 the discriminator with the fake images. 261 00:19:32,300 --> 00:19:33,920 So let's do this. 262 00:19:33,920 --> 00:19:40,080 And actually it's very simple we just need to copy this line because it's almost going to be the same. 263 00:19:40,090 --> 00:19:44,080 We will just need to change the name of this new error. 264 00:19:44,210 --> 00:19:48,340 So this new error corresponds to the training of the discriminator with the fake image. 265 00:19:48,350 --> 00:19:54,580 So instead of calling it we are the real we will call it R R D fake. 266 00:19:54,800 --> 00:19:59,840 And there we go and then that's the same because we have the same variable names for the outputs and 267 00:19:59,840 --> 00:20:00,620 the target. 268 00:20:00,830 --> 00:20:01,830 Wonderful. 269 00:20:01,850 --> 00:20:08,940 And so now congratulations you are done with the two trainings that we had to do with the discriminator. 270 00:20:09,080 --> 00:20:15,230 We trained the discriminator to recognize real images and fake images. 271 00:20:15,230 --> 00:20:21,770 So two subsets done one more to go and we'll do the last one about back propagating the total error 272 00:20:21,890 --> 00:20:26,000 back into the new one that work of discriminator and the next tutorial. 273 00:20:26,000 --> 00:20:27,860 Until then enjoy computer vision. 30001