1 00:00:00,000 --> 00:00:03,000 So guys, now we are going to continue our discussion with respect to this particular project. 2 00:00:03,000 --> 00:00:07,000 Now already in our previous video we have actually created three pickle file. 3 00:00:07,000 --> 00:00:12,000 One is the scalar pickle file label encoder gender pickle file and one hot encoder for geography pickle 4 00:00:12,000 --> 00:00:13,000 file. 5 00:00:13,000 --> 00:00:16,000 Right now we will go ahead and train an artificial neural network. 6 00:00:16,000 --> 00:00:20,000 And that is what we are going to see how we are going to train this and all. 7 00:00:20,000 --> 00:00:22,000 We'll be discussing about that okay. 8 00:00:23,000 --> 00:00:27,000 So first of all we will go ahead and import TensorFlow okay. 9 00:00:27,000 --> 00:00:30,000 Since we are going to just use TensorFlow as tf okay. 10 00:00:30,000 --> 00:00:33,000 So tf is an alias which we will be specifically using. 11 00:00:33,000 --> 00:00:40,000 Now, uh, before I go ahead, let me just give you a brief idea, like how does a neural network basically 12 00:00:40,000 --> 00:00:42,000 gets created with the help of TensorFlow Keras? 13 00:00:42,000 --> 00:00:43,000 Okay. 14 00:00:43,000 --> 00:00:52,000 So as I said, uh, with respect to Ann, right now in Ann you have a input layer. 15 00:00:52,000 --> 00:00:54,000 Then you have a hidden layer. 16 00:00:54,000 --> 00:01:00,000 You can have multiple layers like hidden layer, like you can have HL one XL two, any number of hidden 17 00:01:00,000 --> 00:01:01,000 layers. 18 00:01:01,000 --> 00:01:01,000 Right. 19 00:01:01,000 --> 00:01:07,000 But just to give you a basic idea, basic Ann looks like it will be having one input layer. 20 00:01:07,000 --> 00:01:10,000 It can have one hidden layer, two hidden layers. 21 00:01:10,000 --> 00:01:12,000 And finally it has a output layer. 22 00:01:12,000 --> 00:01:12,000 Okay. 23 00:01:13,000 --> 00:01:22,000 When we probably take this entire neural network like a specific Ann, let's say in this input I have 24 00:01:22,000 --> 00:01:27,000 like this hidden layer, I have some three nodes, one more hidden layer, I have two nodes. 25 00:01:27,000 --> 00:01:28,000 And finally I have one output node. 26 00:01:29,000 --> 00:01:29,000 Right. 27 00:01:29,000 --> 00:01:35,000 This neural network here we are going to perform the forward propagation and backward propagation. 28 00:01:35,000 --> 00:01:43,000 So in TensorFlow we basically call it as a sequential sequential model. 29 00:01:43,000 --> 00:01:43,000 Okay. 30 00:01:43,000 --> 00:01:45,000 What do we call an As? 31 00:01:45,000 --> 00:01:48,000 It is nothing, but it is a sequential model, right. 32 00:01:48,000 --> 00:01:56,000 And this sequential model here you will be able to see that we will be doing forward propagation, we 33 00:01:56,000 --> 00:01:57,000 will be doing backward propagation. 34 00:01:57,000 --> 00:02:00,000 And all these nodes will be interconnected. 35 00:02:00,000 --> 00:02:00,000 Right. 36 00:02:00,000 --> 00:02:03,000 This input will be interconnected to this. 37 00:02:03,000 --> 00:02:05,000 Then this will be interconnected to this. 38 00:02:06,000 --> 00:02:08,000 And similarly this will be interconnected to this. 39 00:02:09,000 --> 00:02:11,000 This will be interconnected to this. 40 00:02:11,000 --> 00:02:11,000 Right. 41 00:02:12,000 --> 00:02:13,000 And this will also be interconnected. 42 00:02:13,000 --> 00:02:16,000 And finally you'll be interconnected to the output. 43 00:02:16,000 --> 00:02:17,000 So this is my input. 44 00:02:17,000 --> 00:02:18,000 This is my hidden layer one. 45 00:02:18,000 --> 00:02:20,000 This is my hidden layer two. 46 00:02:20,000 --> 00:02:24,000 All this interconnected lines that you have are nothing. 47 00:02:24,000 --> 00:02:31,000 But these are weights W1W2W3 like W4W5W6. 48 00:02:31,000 --> 00:02:32,000 Like this right. 49 00:02:32,000 --> 00:02:37,000 Now in order to give you a brief idea since I have, let's consider this right. 50 00:02:37,000 --> 00:02:37,000 I have two inputs. 51 00:02:37,000 --> 00:02:42,000 I have three, three, uh, nodes, hidden nodes in this hidden layer. 52 00:02:42,000 --> 00:02:42,000 Right. 53 00:02:42,000 --> 00:02:47,000 So here the number of weights that will be it will be nothing but two cross three weights. 54 00:02:47,000 --> 00:02:48,000 Right. 55 00:02:48,000 --> 00:02:50,000 And two cross three is nothing but six weights. 56 00:02:50,000 --> 00:02:51,000 Why two cross three? 57 00:02:51,000 --> 00:02:52,000 It's very simple. 58 00:02:53,000 --> 00:02:58,000 Two inputs, and it is getting connected to three weights at three hidden, uh, neurons in the hidden 59 00:02:58,000 --> 00:02:58,000 layer. 60 00:02:58,000 --> 00:03:02,000 Right along with this here also you will be having a bias B one. 61 00:03:03,000 --> 00:03:05,000 You'll be adding bias b2 and b3. 62 00:03:05,000 --> 00:03:11,000 So along with this six weights they are three bias in this first hidden layer okay. 63 00:03:12,000 --> 00:03:17,000 In the second hidden layer, uh, again here you have three hidden layer, uh, three hidden neurons. 64 00:03:17,000 --> 00:03:19,000 This is getting connected to two hidden neurons. 65 00:03:19,000 --> 00:03:23,000 So over here, you'll be able to see that the weight matrix will be three cross two. 66 00:03:23,000 --> 00:03:23,000 Right. 67 00:03:23,000 --> 00:03:27,000 So again you will be having one more additional six weights. 68 00:03:27,000 --> 00:03:27,000 Right. 69 00:03:28,000 --> 00:03:31,000 And uh here also bias will be added. 70 00:03:31,000 --> 00:03:33,000 So this will be my before this will be my be five. 71 00:03:33,000 --> 00:03:36,000 So here six weights and two bias will be there. 72 00:03:36,000 --> 00:03:37,000 Okay. 73 00:03:38,000 --> 00:03:40,000 So this is with respect to the hidden layer one. 74 00:03:41,000 --> 00:03:42,000 Hidden layer one. 75 00:03:43,000 --> 00:03:44,000 Sorry, hidden layer two. 76 00:03:44,000 --> 00:03:46,000 And this is with respect to hidden layer one. 77 00:03:46,000 --> 00:03:51,000 And finally you have this output layer here also your weights will be connected. 78 00:03:51,000 --> 00:03:53,000 So here some weights will be there. 79 00:03:53,000 --> 00:03:56,000 So here also how many weights will be there. 80 00:03:56,000 --> 00:03:58,000 Two cross one which is two weights. 81 00:03:59,000 --> 00:04:00,000 Right. 82 00:04:00,000 --> 00:04:04,000 And uh here also one bias will get added. 83 00:04:04,000 --> 00:04:06,000 So one bias. 84 00:04:06,000 --> 00:04:11,000 And after this you can probably apply a sigmoid activation function if it is a classification. 85 00:04:12,000 --> 00:04:15,000 If it is a multi-class classification, you can apply softmax right. 86 00:04:15,000 --> 00:04:17,000 So total number of parameters that you see. 87 00:04:17,000 --> 00:04:18,000 Right. 88 00:04:18,000 --> 00:04:22,000 If I say how many trainable parameters are there. 89 00:04:23,000 --> 00:04:30,000 Trainable parameters are there in this neural network it is very simple. 90 00:04:30,000 --> 00:04:36,000 You just say, hey, six plus three, this is nine, then nine plus six, 15, 15 plus two, 17, 17 91 00:04:36,000 --> 00:04:39,000 plus two, 19, 19 plus one, 20. 92 00:04:39,000 --> 00:04:46,000 So with the help of 20 trainable parameters, we will be doing the forward propagation and backward 93 00:04:46,000 --> 00:04:46,000 propagation. 94 00:04:47,000 --> 00:04:54,000 Now similarly this input is basically defined by the number of inputs that we have in our CSV file or 95 00:04:54,000 --> 00:04:55,000 in our data set. 96 00:04:55,000 --> 00:04:55,000 Right. 97 00:04:56,000 --> 00:05:01,000 In our data set, how many inputs we have in data set, how many inputs we have. 98 00:05:01,000 --> 00:05:01,000 If we have two. 99 00:05:01,000 --> 00:05:02,000 So it will be two. 100 00:05:02,000 --> 00:05:04,000 If we have three, it will be three nodes, right. 101 00:05:04,000 --> 00:05:07,000 If we have four, 11, ten any number of nodes. 102 00:05:07,000 --> 00:05:13,000 So in this way only we will be going and created our Ann which is a sequential network. 103 00:05:13,000 --> 00:05:17,000 Sequential network basically means it will be interconnected, right. 104 00:05:17,000 --> 00:05:22,000 Now there is also one more important point if we really want to just create this hidden neurons. 105 00:05:22,000 --> 00:05:27,000 So in Keras there is a library which is called as dense okay. 106 00:05:27,000 --> 00:05:29,000 There is a class which is called as dense okay. 107 00:05:29,000 --> 00:05:34,000 So let me again repeat what are the key parameters that we will be specifically using. 108 00:05:34,000 --> 00:05:42,000 First, when we start creating our Ann we have to initialize a sequential network okay. 109 00:05:42,000 --> 00:05:44,000 See sequential network. 110 00:05:44,000 --> 00:05:48,000 The second thing is that whenever we need to create a hidden neuron. 111 00:05:48,000 --> 00:05:48,000 Right. 112 00:05:48,000 --> 00:05:50,000 Hidden neuron I'm not talking about hidden layer. 113 00:05:50,000 --> 00:05:55,000 If you need to create a hidden neuron we will specifically using dense okay. 114 00:05:55,000 --> 00:05:58,000 So if I say hey dense I want 64 nodes. 115 00:05:58,000 --> 00:06:03,000 So in that particular hidden layer I will be having 64 hidden neurons. 116 00:06:03,000 --> 00:06:04,000 Okay something like this. 117 00:06:05,000 --> 00:06:07,000 So we will be using dense okay. 118 00:06:07,000 --> 00:06:13,000 Along with this, uh, you know that in every node we apply an activation function, right. 119 00:06:13,000 --> 00:06:18,000 So here we will also be using activation function. 120 00:06:18,000 --> 00:06:21,000 The activation function can be sigmoid. 121 00:06:21,000 --> 00:06:24,000 It can be tan h right. 122 00:06:24,000 --> 00:06:26,000 It can be ReLU. 123 00:06:26,000 --> 00:06:29,000 It can be leaky ReLU right. 124 00:06:29,000 --> 00:06:30,000 It can be many. 125 00:06:30,000 --> 00:06:32,000 Any number of others, right? 126 00:06:33,000 --> 00:06:37,000 But always it is a good idea that in the hidden layer you try to use ReLU. 127 00:06:37,000 --> 00:06:42,000 In the output layer you can use sigmoid or softmax, which I've actually discussed about it. 128 00:06:42,000 --> 00:06:44,000 Now this is my third important parameter. 129 00:06:44,000 --> 00:06:49,000 The fourth parameter that we will be using is something called as optimizer. 130 00:06:50,000 --> 00:06:54,000 Now this optimizer can be any kind of optimizer that we have. 131 00:06:54,000 --> 00:06:56,000 I will show you how you can initialize this particular optimizer. 132 00:06:56,000 --> 00:07:06,000 Also, optimizer is uh, useful in back propagation because these are responsible in updating the weights. 133 00:07:07,000 --> 00:07:07,000 Okay. 134 00:07:07,000 --> 00:07:09,000 Updating the weights. 135 00:07:10,000 --> 00:07:11,000 Very important. 136 00:07:11,000 --> 00:07:12,000 Okay. 137 00:07:12,000 --> 00:07:17,000 And the fifth one that we are specifically going to use is nothing but our loss function. 138 00:07:17,000 --> 00:07:23,000 And our main aim is that what loss function will be using and based on this loss function will be having 139 00:07:23,000 --> 00:07:25,000 a gradient descent and all will try to reduce this. 140 00:07:25,000 --> 00:07:26,000 Okay. 141 00:07:26,000 --> 00:07:29,000 And the sixth and the final thing is something called as metrics. 142 00:07:29,000 --> 00:07:34,000 Usually in the classification problem your metrics is like accuracy. 143 00:07:34,000 --> 00:07:41,000 In our regression problem you have MSE mean squared error, mean absolute error and all are there okay. 144 00:07:42,000 --> 00:07:48,000 So all these things will try to implement and my training information also what I will do my entire 145 00:07:48,000 --> 00:07:54,000 training information I will store the logs in some folder okay. 146 00:07:55,000 --> 00:07:58,000 So that I will be able to use tensor board. 147 00:07:58,000 --> 00:08:06,000 And the main aim of the tensor board library is to display this logs in a way that we will be able to 148 00:08:06,000 --> 00:08:07,000 understand. 149 00:08:07,000 --> 00:08:10,000 Some visualization graphs will be used in order to display it. 150 00:08:10,000 --> 00:08:11,000 Okay. 151 00:08:11,000 --> 00:08:14,000 So these are the steps that we are going to first of all focus. 152 00:08:14,000 --> 00:08:18,000 We are going to just go ahead and create a sequential neural network. 153 00:08:18,000 --> 00:08:22,000 Whenever we really want to create any hidden neurons, this hidden neuron, this hidden neuron or even 154 00:08:22,000 --> 00:08:26,000 even in the input I will be specifically using dense. 155 00:08:26,000 --> 00:08:26,000 Okay. 156 00:08:26,000 --> 00:08:29,000 So all these things we are going to go ahead and uh discuss. 157 00:08:29,000 --> 00:08:35,000 So let me quickly, uh, open this now, I will go back to my code. 158 00:08:37,000 --> 00:08:37,000 Okay. 159 00:08:37,000 --> 00:08:43,000 And, uh, here you will be able to see that I will just go ahead and import TensorFlow okay. 160 00:08:44,000 --> 00:08:49,000 So I'm going to import TensorFlow as tf I told you I'm going to use sequential model. 161 00:08:49,000 --> 00:08:50,000 So how do I import it. 162 00:08:50,000 --> 00:08:59,000 So I will go ahead and say hey from TensorFlow from TensorFlow dot Keras dot models. 163 00:08:59,000 --> 00:09:03,000 We are going to import this sequential model. 164 00:09:05,000 --> 00:09:10,000 Sequential okay from TensorFlow Keras dot model okay. 165 00:09:11,000 --> 00:09:19,000 Similarly I will go ahead and write from TensorFlow dot keras dot. 166 00:09:19,000 --> 00:09:24,000 As I said you whenever I need to create a node it is nothing but using dense. 167 00:09:24,000 --> 00:09:29,000 Okay, dense will be that library that we will be using and this entire thing will be present inside 168 00:09:29,000 --> 00:09:30,000 layers. 169 00:09:30,000 --> 00:09:30,000 Okay. 170 00:09:30,000 --> 00:09:33,000 So I'm just going to import dense. 171 00:09:34,000 --> 00:09:35,000 So this is one more library. 172 00:09:35,000 --> 00:09:41,000 Along with this I will be showing you one very important property we'll discuss about which is called 173 00:09:41,000 --> 00:09:45,000 as TensorFlow Keras dot callbacks. 174 00:09:45,000 --> 00:09:46,000 Okay. 175 00:09:46,000 --> 00:09:50,000 And we're just going to import something called as early stopping. 176 00:09:50,000 --> 00:09:53,000 We'll discuss about this early stopping what exactly it is okay. 177 00:09:53,000 --> 00:09:57,000 And finally we have this tensor board okay. 178 00:09:58,000 --> 00:10:01,000 Now tensor board uh we have implemented this. 179 00:10:01,000 --> 00:10:05,000 And finally I'm also going to import date time because I'm going to use this date time also. 180 00:10:05,000 --> 00:10:09,000 So I have executed or imported all these libraries which we are going to use. 181 00:10:09,000 --> 00:10:14,000 One is sequential, one is dense, one is early stopping and one is TensorBoard. 182 00:10:14,000 --> 00:10:17,000 Okay, now let's go ahead and build our model. 183 00:10:18,000 --> 00:10:23,000 Build our an model okay. 184 00:10:25,000 --> 00:10:27,000 Now to build the an model. 185 00:10:27,000 --> 00:10:29,000 What all things we are specifically going to do. 186 00:10:29,000 --> 00:10:33,000 As I said I'm going to create first of all a sequential model. 187 00:10:33,000 --> 00:10:37,000 So I will go ahead and initialize and let me just go ahead and initialize sequential. 188 00:10:38,000 --> 00:10:38,000 Okay. 189 00:10:39,000 --> 00:10:43,000 Now inside the sequential it is important. 190 00:10:43,000 --> 00:10:44,000 Please try to understand this okay. 191 00:10:46,000 --> 00:10:51,000 Inside the sequential I am going to use some layers dense layers and all. 192 00:10:51,000 --> 00:10:52,000 Okay. 193 00:10:52,000 --> 00:10:57,000 So here uh uh if I go back to my diagram. 194 00:10:57,000 --> 00:10:59,000 Okay, let's go back to my diagram over here. 195 00:11:00,000 --> 00:11:08,000 First of all, let's go ahead and understand how many number of inputs that I have okay. 196 00:11:09,000 --> 00:11:14,000 So first this is my entire sequential model okay. 197 00:11:17,000 --> 00:11:19,000 This is my entire sequential model. 198 00:11:20,000 --> 00:11:24,000 Now inside the sequential model, how many inputs I need to probably consider. 199 00:11:24,000 --> 00:11:32,000 So if I go ahead and see my X train dot shape, and from this, if I just try to find out how many columns 200 00:11:32,000 --> 00:11:38,000 are there, that will actually help us to decide how many inputs will be there, right? 201 00:11:38,000 --> 00:11:39,000 How many inputs will be there? 202 00:11:39,000 --> 00:11:42,000 So I have to probably use this entire inputs. 203 00:11:42,000 --> 00:11:42,000 Okay. 204 00:11:42,000 --> 00:11:43,000 Over here. 205 00:11:43,000 --> 00:11:50,000 And let's say that I will be using the sequential and I'll say, hey, initially I'll take this 11 inputs 206 00:11:50,000 --> 00:11:53,000 and then I will go ahead and create my dense. 207 00:11:53,000 --> 00:11:56,000 I'll use this dense library and create my hidden layer. 208 00:11:56,000 --> 00:12:00,000 Let's say that I'm going to use 64 nodes, any number of nodes in this hidden layer. 209 00:12:00,000 --> 00:12:01,000 Right. 210 00:12:01,000 --> 00:12:07,000 And when I am creating this dense, I will also give my input shape, like how many elements are coming 211 00:12:07,000 --> 00:12:08,000 from this input data. 212 00:12:08,000 --> 00:12:13,000 So let's say from this I, I'm getting this x train dot shape of one. 213 00:12:13,000 --> 00:12:19,000 If I go ahead and write like this, it says that what is the input that is probably coming from my X 214 00:12:19,000 --> 00:12:19,000 train data. 215 00:12:19,000 --> 00:12:21,000 That input will get connected over here. 216 00:12:22,000 --> 00:12:22,000 Okay. 217 00:12:22,000 --> 00:12:26,000 So we are initializing our input over here along with the dense layer. 218 00:12:26,000 --> 00:12:28,000 So this is what we are going to first of all do okay. 219 00:12:28,000 --> 00:12:35,000 So quickly I'll go back over here and now see focus on this okay I need to give my input and my hidden 220 00:12:35,000 --> 00:12:35,000 layer one. 221 00:12:35,000 --> 00:12:40,000 First I'll use this dense, let's say in my hidden layer what I'm going to say hey, we are just going 222 00:12:40,000 --> 00:12:42,000 to use 64 neurons. 223 00:12:42,000 --> 00:12:43,000 I'm just considering. 224 00:12:43,000 --> 00:12:45,000 Hey let's go ahead and use 64 neurons. 225 00:12:45,000 --> 00:12:48,000 Now in all this neuron I also need to apply an activation function. 226 00:12:48,000 --> 00:12:53,000 So I will go ahead and apply an activation function okay. 227 00:12:53,000 --> 00:12:58,000 Now once I apply the activation function I will just go ahead and apply ReLU activation function. 228 00:12:58,000 --> 00:12:59,000 Okay. 229 00:12:59,000 --> 00:13:04,000 It is up to you whatever activation functions that you really want to apply now to this hidden layer, 230 00:13:04,000 --> 00:13:06,000 what will be the input that will be coming. 231 00:13:06,000 --> 00:13:11,000 So here you will be having this particular parameter which is called as input underscore shape. 232 00:13:11,000 --> 00:13:15,000 And this is only for the first hidden layer that you really need to write, because it is getting connected 233 00:13:15,000 --> 00:13:18,000 to some inputs over there, right in the second hidden layer and all. 234 00:13:18,000 --> 00:13:23,000 We don't need to write this input underscore shape, but always for the first hidden layer we need to 235 00:13:23,000 --> 00:13:23,000 write this. 236 00:13:23,000 --> 00:13:26,000 So my input underscore shape will be nothing. 237 00:13:26,000 --> 00:13:30,000 But it will uh, again just go ahead and write something like this. 238 00:13:30,000 --> 00:13:31,000 Right. 239 00:13:31,000 --> 00:13:37,000 So it will be x underscore train dot shape. 240 00:13:37,000 --> 00:13:43,000 Because this shape of one, if I probably consider it is just going to take that particular value. 241 00:13:43,000 --> 00:13:47,000 And it is going to consider that that many number of inputs are there. 242 00:13:47,000 --> 00:13:57,000 So if I just go ahead and execute this over here, what is my X underscore train dot shape of one okay 243 00:13:57,000 --> 00:13:59,000 I'll be okay. 244 00:13:59,000 --> 00:13:59,000 Tuple. 245 00:13:59,000 --> 00:14:02,000 Uh so here it says okay I have to use this. 246 00:14:03,000 --> 00:14:06,000 Uh I have to use this index okay. 247 00:14:07,000 --> 00:14:09,000 So shape of one. 248 00:14:11,000 --> 00:14:12,000 Okay. 249 00:14:12,000 --> 00:14:15,000 So here you can see that there are two L elements okay. 250 00:14:15,000 --> 00:14:18,000 In my X underscore train dot shape of one. 251 00:14:18,000 --> 00:14:24,000 Right now if I'm giving input underscore shape of this value comma blank. 252 00:14:24,000 --> 00:14:28,000 If I just give comma blank like this in a tuple. 253 00:14:29,000 --> 00:14:34,000 So I'm just saying that hey it is a single dimension and it has 12 inputs okay. 254 00:14:34,000 --> 00:14:37,000 And that is how I'm also giving over here as my input shape. 255 00:14:37,000 --> 00:14:38,000 Clear. 256 00:14:38,000 --> 00:14:39,000 Very much simple. 257 00:14:39,000 --> 00:14:41,000 Now this is my first hidden layer. 258 00:14:41,000 --> 00:14:43,000 So I'll write down the comment over here. 259 00:14:43,000 --> 00:14:43,000 First hidden layer. 260 00:14:44,000 --> 00:14:48,000 Uh and it is in uh connected with input layer. 261 00:14:48,000 --> 00:14:49,000 Okay. 262 00:14:49,000 --> 00:14:51,000 Connected with input layer. 263 00:14:51,000 --> 00:14:53,000 This is perfect okay. 264 00:14:53,000 --> 00:15:00,000 Now in my second layer again let's go ahead and use the dense I will use 32 neurons over here. 265 00:15:00,000 --> 00:15:03,000 And I have to just go ahead and apply my activation function. 266 00:15:03,000 --> 00:15:08,000 Now after I apply my activation function, I don't even have to give any input shape over here, because 267 00:15:08,000 --> 00:15:13,000 now we know since we have used sequential, it knows that it will be connected, right? 268 00:15:13,000 --> 00:15:15,000 All these layers will be specifically connected. 269 00:15:15,000 --> 00:15:15,000 Okay. 270 00:15:15,000 --> 00:15:21,000 Now finally you will be also see that I'll go ahead and apply one dense one okay. 271 00:15:21,000 --> 00:15:25,000 And then I will say hey go ahead and apply another activation function. 272 00:15:25,000 --> 00:15:32,000 And the reason I'm saying that another activation function, this is my hidden layer two HL two okay. 273 00:15:32,000 --> 00:15:36,000 And here you will be able to see that this is nothing. 274 00:15:36,000 --> 00:15:37,000 But this is my output layer. 275 00:15:38,000 --> 00:15:39,000 Output layer. 276 00:15:39,000 --> 00:15:40,000 How many outputs I need to find out? 277 00:15:40,000 --> 00:15:41,000 Only one. 278 00:15:41,000 --> 00:15:41,000 Right. 279 00:15:41,000 --> 00:15:42,000 It is a binary output. 280 00:15:42,000 --> 00:15:43,000 Right. 281 00:15:43,000 --> 00:15:47,000 So I'll say hey let's go ahead and use this sigmoid over here Okay. 282 00:15:47,000 --> 00:15:51,000 So once I execute this so this actually becomes my entire model. 283 00:15:52,000 --> 00:15:52,000 Okay. 284 00:15:52,000 --> 00:15:54,000 My entire model over here okay. 285 00:15:54,000 --> 00:15:57,000 So if I go ahead and execute this this is my entire model. 286 00:15:58,000 --> 00:16:04,000 Very simple I guess right now just imagine how many parameters will be there. 287 00:16:04,000 --> 00:16:08,000 So what I've actually done over here, let me just go ahead and talk about it okay. 288 00:16:09,000 --> 00:16:13,000 So here I've created a input layer. 289 00:16:13,000 --> 00:16:15,000 So this is my input layer. 290 00:16:15,000 --> 00:16:21,000 This input layer has uh inputs of 12 inputs. 291 00:16:21,000 --> 00:16:22,000 Right. 292 00:16:22,000 --> 00:16:23,000 12 inputs 12 comma some. 293 00:16:23,000 --> 00:16:24,000 Let's consider 12. 294 00:16:24,000 --> 00:16:27,000 Then I have a hidden layer. 295 00:16:27,000 --> 00:16:33,000 This hidden layer is basically happening having 64 hidden neurons. 296 00:16:33,000 --> 00:16:36,000 Right 64 hidden neurons. 297 00:16:36,000 --> 00:16:38,000 Then I have one more hidden layer. 298 00:16:40,000 --> 00:16:43,000 Which has 32 hidden neurons. 299 00:16:43,000 --> 00:16:48,000 And finally, this is my output layer, which is having one output, right? 300 00:16:48,000 --> 00:16:49,000 One neuron. 301 00:16:49,000 --> 00:16:50,000 And this all will be interconnected. 302 00:16:50,000 --> 00:16:52,000 And this is what we have actually created. 303 00:16:52,000 --> 00:16:52,000 Right. 304 00:16:52,000 --> 00:16:56,000 And now we have to probably do the forward and backward propagation. 305 00:16:58,000 --> 00:17:00,000 Now let me just go back to the code. 306 00:17:00,000 --> 00:17:02,000 Now let's go ahead and write model dot summary. 307 00:17:02,000 --> 00:17:06,000 And with the help of model dot summary you'll be able to see that I'll be able to see how many total 308 00:17:06,000 --> 00:17:07,000 number of parameters are there. 309 00:17:07,000 --> 00:17:11,000 So they are around 2945 parameters. 310 00:17:11,000 --> 00:17:12,000 Okay. 311 00:17:12,000 --> 00:17:17,000 So here uh, you can probably see in my dense, uh, initially you have the 64. 312 00:17:18,000 --> 00:17:21,000 Um, then you have this 32, then you have this one. 313 00:17:21,000 --> 00:17:21,000 Okay. 314 00:17:22,000 --> 00:17:22,000 Parameters. 315 00:17:22,000 --> 00:17:26,000 This is nothing, but it is a combination of all the weights and bias. 316 00:17:26,000 --> 00:17:30,000 And that is the reason the total number of parameters will be two nine, four, five okay. 317 00:17:30,000 --> 00:17:32,000 So this is the entire model. 318 00:17:32,000 --> 00:17:35,000 And I'm actually going to do the forward and backward propagation with respect to this. 319 00:17:35,000 --> 00:17:39,000 Now in order to do the forward and backward propagation we need to compile this model. 320 00:17:39,000 --> 00:17:43,000 So let me just go ahead and write compile the model okay. 321 00:17:43,000 --> 00:17:47,000 Now with respect to the compile the model I'll just go ahead and write model dot compile. 322 00:17:48,000 --> 00:17:50,000 And here I'm just going to use some parameters. 323 00:17:50,000 --> 00:17:52,000 One is optimizer. 324 00:17:52,000 --> 00:17:53,000 Okay. 325 00:17:53,000 --> 00:17:57,000 Right now let's keep it blank I will initialize some optimizer over here. 326 00:17:57,000 --> 00:17:59,000 Then you have loss. 327 00:17:59,000 --> 00:18:04,000 And this loss uh I'm just going to use binary cross entropy. 328 00:18:04,000 --> 00:18:07,000 If it is a multi classification you can use sparse cross entropy okay. 329 00:18:07,000 --> 00:18:11,000 So there is a different different values with respect to multi classification. 330 00:18:11,000 --> 00:18:15,000 You will be seeing a parameter which is called as sparse cross entropy. 331 00:18:15,000 --> 00:18:17,000 Again you can go ahead and check with respect to Keras API. 332 00:18:18,000 --> 00:18:21,000 And here we will also go ahead and use one matrix parameter. 333 00:18:21,000 --> 00:18:24,000 And this will be nothing but it will be accuracy. 334 00:18:25,000 --> 00:18:28,000 It is given in the form of list because you can add multiple values also. 335 00:18:29,000 --> 00:18:33,000 Now uh, quickly, one more thing that I really want to show that over here. 336 00:18:33,000 --> 00:18:40,000 Uh, is this the only way to probably put any values, or can I probably go ahead and use some other 337 00:18:40,000 --> 00:18:42,000 way of applying optimizers and all? 338 00:18:42,000 --> 00:18:45,000 Okay, so, uh, let me just show you one more way. 339 00:18:45,000 --> 00:18:46,000 Okay. 340 00:18:46,000 --> 00:18:53,000 So here I'll say, uh, Opt optimizer is nothing but TensorFlow dot. 341 00:18:53,000 --> 00:18:53,000 Okay. 342 00:18:53,000 --> 00:18:56,000 There is a function which is called as optimizers. 343 00:18:57,000 --> 00:19:00,000 TensorFlow optimizers. 344 00:19:00,000 --> 00:19:01,000 Okay. 345 00:19:01,000 --> 00:19:01,000 Sorry. 346 00:19:01,000 --> 00:19:02,000 TensorFlow. 347 00:19:02,000 --> 00:19:05,000 Keras dot Keras dot optimizer. 348 00:19:06,000 --> 00:19:06,000 Okay. 349 00:19:06,000 --> 00:19:10,000 And here we have something called as Adam optimizer. 350 00:19:12,000 --> 00:19:14,000 I think so it should be working fine. 351 00:19:14,000 --> 00:19:22,000 And let's use my learning underscore rate is equal to 0.01. 352 00:19:23,000 --> 00:19:23,000 Okay. 353 00:19:24,000 --> 00:19:24,000 or okay. 354 00:19:24,000 --> 00:19:26,000 TensorFlow is not defined. 355 00:19:26,000 --> 00:19:28,000 So that is the reason you could not see it okay. 356 00:19:29,000 --> 00:19:30,000 Okay. 357 00:19:30,000 --> 00:19:36,000 So now what we have actually done is that see over here when we go ahead and give one optimizer, let's 358 00:19:36,000 --> 00:19:39,000 say that I want to go ahead and give one optimizer over here. 359 00:19:39,000 --> 00:19:41,000 And the optimizer name is Adam okay. 360 00:19:41,000 --> 00:19:45,000 This is the keyword that we can specifically use again in Keras API documentation. 361 00:19:45,000 --> 00:19:48,000 If you go and see I'm actually applying Adam over here. 362 00:19:48,000 --> 00:19:49,000 Okay. 363 00:19:49,000 --> 00:19:52,000 But this Adam has a fixed learning rate okay. 364 00:19:52,000 --> 00:19:53,000 Over here it has a learning rate. 365 00:19:53,000 --> 00:19:56,000 But you you need to provide your own learning rate. 366 00:19:56,000 --> 00:20:01,000 So what you can do you can just import TensorFlow and you can write hey TensorFlow keras dot optimizer 367 00:20:01,000 --> 00:20:03,000 Adam and not any Adam. 368 00:20:03,000 --> 00:20:05,000 See there are a lot of other optimizers. 369 00:20:05,000 --> 00:20:06,000 Also, let me just show you. 370 00:20:06,000 --> 00:20:09,000 So here I'll go ahead and write optimizers. 371 00:20:11,000 --> 00:20:13,000 Dot Adam. 372 00:20:13,000 --> 00:20:16,000 So instead of using Adam you have adadelta add a factor. 373 00:20:16,000 --> 00:20:17,000 Adagrad Adam, Adam w Adam. 374 00:20:17,000 --> 00:20:21,000 Max I think, uh, most of them I have already discussed. 375 00:20:21,000 --> 00:20:24,000 We have discussed Adam, we have discussed adadelta all these things. 376 00:20:24,000 --> 00:20:25,000 Right? 377 00:20:25,000 --> 00:20:25,000 Adagrad. 378 00:20:25,000 --> 00:20:26,000 Also, we have discussed. 379 00:20:26,000 --> 00:20:27,000 Right. 380 00:20:27,000 --> 00:20:30,000 So let's consider right now we'll just go ahead and use Adam. 381 00:20:30,000 --> 00:20:34,000 And here, uh, there will be a parameter which is called as the learning rate, which we can actually 382 00:20:34,000 --> 00:20:35,000 define. 383 00:20:35,000 --> 00:20:35,000 Okay. 384 00:20:36,000 --> 00:20:38,000 So here I have actually defined 0.01. 385 00:20:38,000 --> 00:20:43,000 Similarly, what you can do is that you can also go ahead and write TensorFlow dot Keras. 386 00:20:43,000 --> 00:20:56,000 And you can also set your uh loss okay losses and uh dot different different losses will be also there 387 00:20:56,000 --> 00:20:57,000 I guess. 388 00:20:57,000 --> 00:21:02,000 Let's see uh see over here you have are binary cross entropy. 389 00:21:02,000 --> 00:21:04,000 Then you have categorical cross entropy. 390 00:21:04,000 --> 00:21:06,000 You can use all these specific losses if you want. 391 00:21:06,000 --> 00:21:10,000 Let's say if I want to directly use binary cross entropy I can also use this okay. 392 00:21:10,000 --> 00:21:14,000 And I can store this in one variable like one loss. 393 00:21:14,000 --> 00:21:16,000 And I can use this particular loss over here. 394 00:21:16,000 --> 00:21:19,000 Like see if I use this opt I'll be using this over here. 395 00:21:19,000 --> 00:21:20,000 Copy and paste it. 396 00:21:20,000 --> 00:21:21,000 Right. 397 00:21:21,000 --> 00:21:23,000 So I'm showing you both the ways how you can actually do it directly. 398 00:21:23,000 --> 00:21:28,000 You can also use this particular keyword, or you can also go ahead and initialize this okay. 399 00:21:28,000 --> 00:21:32,000 So if I go ahead and initialize this so I will be able to see that hey this is my loss which looks like 400 00:21:32,000 --> 00:21:33,000 right. 401 00:21:33,000 --> 00:21:35,000 But uh both these ways. 402 00:21:35,000 --> 00:21:38,000 So what I'll do I'll take this optimizer, I'll put it over here. 403 00:21:38,000 --> 00:21:42,000 And similarly I'll use this particular loss and we'll go ahead and do the compilation. 404 00:21:42,000 --> 00:21:42,000 Okay. 405 00:21:43,000 --> 00:21:46,000 Now when we are doing the compilation this is the compiling the model over here. 406 00:21:46,000 --> 00:21:47,000 So okay. 407 00:21:47,000 --> 00:21:50,000 Uh, there is an error metric. 408 00:21:50,000 --> 00:21:52,000 Uh, please. 409 00:21:52,000 --> 00:21:54,000 Uh, okay, let's see what it says. 410 00:21:54,000 --> 00:22:01,000 Including this, this this this this this metric metric metric metrics. 411 00:22:01,000 --> 00:22:03,000 It should be, uh, something like metrics. 412 00:22:03,000 --> 00:22:03,000 Okay. 413 00:22:03,000 --> 00:22:05,000 So this is the keyword that is used. 414 00:22:05,000 --> 00:22:07,000 So uh, again this kind of error will come. 415 00:22:07,000 --> 00:22:10,000 I'm making sure that I don't delete any of the error okay. 416 00:22:11,000 --> 00:22:17,000 Now the next thing what I will do I will go ahead and set up the TensorBoard because we have still not 417 00:22:17,000 --> 00:22:18,000 started the training. 418 00:22:18,000 --> 00:22:19,000 Right? 419 00:22:19,000 --> 00:22:20,000 We have still not started. 420 00:22:20,000 --> 00:22:25,000 The training and training will be set up, you know, as soon as I probably start training. 421 00:22:25,000 --> 00:22:26,000 Uh, sorry. 422 00:22:26,000 --> 00:22:27,000 This. 423 00:22:27,000 --> 00:22:32,000 Once the training gets completed, we will also be able to, uh, capture that logs and we'll be able 424 00:22:32,000 --> 00:22:33,000 to visualize that logs. 425 00:22:33,000 --> 00:22:35,000 So that is the reason we set up this TensorBoard. 426 00:22:35,000 --> 00:22:39,000 So I'll create a directory which is called as log underscore directory. 427 00:22:39,000 --> 00:22:42,000 And here I will say hey make this folder okay. 428 00:22:42,000 --> 00:22:48,000 So first of all log slash fit I will probably make this folder over here okay. 429 00:22:49,000 --> 00:22:54,000 And I'll say hey go ahead and make this and use some date time. 430 00:22:54,000 --> 00:22:58,000 Use some date time format so that you'll be able to see this. 431 00:22:58,000 --> 00:23:03,000 And then I will go ahead and write date time okay Dot now. 432 00:23:03,000 --> 00:23:04,000 Okay. 433 00:23:04,000 --> 00:23:05,000 I'm using this date time. 434 00:23:05,000 --> 00:23:08,000 And you let me use some time format. 435 00:23:08,000 --> 00:23:11,000 So I will go ahead and write str of time. 436 00:23:11,000 --> 00:23:15,000 And I will just use this particular time format to capture my logs. 437 00:23:15,000 --> 00:23:15,000 Okay. 438 00:23:15,000 --> 00:23:18,000 So let me just copy and paste it over here okay. 439 00:23:18,000 --> 00:23:21,000 So this is my time format that I'm actually going to use okay. 440 00:23:22,000 --> 00:23:26,000 It will basically go ahead and create this entire directory in shot. 441 00:23:26,000 --> 00:23:26,000 Okay. 442 00:23:27,000 --> 00:23:32,000 Now, uh, quickly, uh, let's do one more, uh, thing over here. 443 00:23:32,000 --> 00:23:36,000 Is that, uh, I'm also going to, uh, uh, okay. 444 00:23:36,000 --> 00:23:39,000 For right now, let me just go ahead and create this logs directory. 445 00:23:39,000 --> 00:23:45,000 And, uh, it is going to make sure that whenever we train our model, it will go ahead and it will 446 00:23:45,000 --> 00:23:46,000 store all the logs over here. 447 00:23:46,000 --> 00:23:47,000 Okay. 448 00:23:48,000 --> 00:23:49,000 Perfect. 449 00:23:49,000 --> 00:23:50,000 Uh, till here. 450 00:23:50,000 --> 00:23:52,000 I think everybody is, uh, good enough. 451 00:23:52,000 --> 00:23:55,000 Uh, now, uh, let me do one more thing. 452 00:23:55,000 --> 00:24:00,000 Let me just go ahead and create my TensorFlow callback. 453 00:24:00,000 --> 00:24:00,000 Okay. 454 00:24:00,000 --> 00:24:03,000 So I'm going to probably create this particular variable. 455 00:24:03,000 --> 00:24:05,000 And what is the main aim of this. 456 00:24:05,000 --> 00:24:08,000 We will just go ahead and say hey use this TensorBoard. 457 00:24:08,000 --> 00:24:10,000 What we are trying to import. 458 00:24:10,000 --> 00:24:13,000 Now this TensorBoard we need to import it okay. 459 00:24:13,000 --> 00:24:16,000 So in order to import it I will be importing two libraries. 460 00:24:16,000 --> 00:24:18,000 Since I'm going to import two important things. 461 00:24:18,000 --> 00:24:26,000 One is import tensorflow dot Keras dot calls back okay I'll be talking about what is the importance 462 00:24:26,000 --> 00:24:26,000 of this. 463 00:24:26,000 --> 00:24:28,000 Calls back just in some time. 464 00:24:28,000 --> 00:24:34,000 So I'll go ahead and write hey import early stopping along with this I'm just going to import TensorBoard. 465 00:24:34,000 --> 00:24:36,000 I will talk about both, uh, early stopping and TensorBoard. 466 00:24:36,000 --> 00:24:43,000 But right now you just understand that tensor board is just to visualize all your logs that you have 467 00:24:43,000 --> 00:24:44,000 while training your entire model. 468 00:24:44,000 --> 00:24:45,000 Okay. 469 00:24:45,000 --> 00:24:47,000 You can also do it manually with the help of matplotlib. 470 00:24:47,000 --> 00:24:50,000 But I feel tensor board will be a good one. 471 00:24:50,000 --> 00:24:52,000 Okay, so here I have initialized tensor board. 472 00:24:52,000 --> 00:24:56,000 I'm saying hey, let's give this as my log directory over here. 473 00:24:56,000 --> 00:24:57,000 Log directory is equal to. 474 00:24:57,000 --> 00:25:01,000 I'm just going to give my log directory whichever log directory I've actually created. 475 00:25:01,000 --> 00:25:07,000 The next thing that I will be using is some histogram frequency as one. 476 00:25:07,000 --> 00:25:12,000 To just see that some of the histogram diagrams is also created from here. 477 00:25:12,000 --> 00:25:12,000 Okay. 478 00:25:13,000 --> 00:25:16,000 So once this is done I'll go ahead and execute it. 479 00:25:16,000 --> 00:25:17,000 Invalid syntax. 480 00:25:18,000 --> 00:25:19,000 Okay. 481 00:25:21,000 --> 00:25:24,000 Uh from I have to use over here as from. 482 00:25:25,000 --> 00:25:25,000 Okay. 483 00:25:25,000 --> 00:25:26,000 So perfect. 484 00:25:27,000 --> 00:25:29,000 This works absolutely fine. 485 00:25:29,000 --> 00:25:29,000 Okay. 486 00:25:29,000 --> 00:25:36,000 Now we will also go ahead and set up something called as early stopping. 487 00:25:36,000 --> 00:25:38,000 Now why exactly is this early stopping? 488 00:25:38,000 --> 00:25:43,000 See guys, when we are training a neural network we can train it for any number of epochs. 489 00:25:43,000 --> 00:25:43,000 Okay. 490 00:25:44,000 --> 00:25:46,000 We can train it for multiple epochs. 491 00:25:46,000 --> 00:25:47,000 100 epochs. 492 00:25:47,000 --> 00:25:51,000 You know, we can create different different batch size and we can do forward and backward propagation. 493 00:25:51,000 --> 00:25:58,000 But, uh, you know, it may happen in such a way that even we run for 5 to 6 epochs continuously there. 494 00:25:58,000 --> 00:26:01,000 Let's say I'm trying it to run it for 100 epochs. 495 00:26:01,000 --> 00:26:01,000 Okay. 496 00:26:02,000 --> 00:26:10,000 But if we see such a scenario where in 20 epochs, only my model has trained at this batch level, and 497 00:26:10,000 --> 00:26:14,000 after that it is just varying by 1 or 2%, right? 498 00:26:14,000 --> 00:26:19,000 Then it is not necessary that you need to probably train that particular model for 100 epochs, right? 499 00:26:20,000 --> 00:26:24,000 Uh, whenever we train our model, we have to probably continuously monitor our loss value. 500 00:26:24,000 --> 00:26:26,000 If that loss value is decreasing, then it is fine. 501 00:26:26,000 --> 00:26:27,000 We can train it continuously. 502 00:26:27,000 --> 00:26:32,000 But let's say that after a certain number of epochs, that particular loss value is not decreasing. 503 00:26:32,000 --> 00:26:34,000 So what do we do at that point time. 504 00:26:34,000 --> 00:26:39,000 We don't need to probably run the entire 100 epochs since, uh, what we can do, we can say, hey, 505 00:26:39,000 --> 00:26:40,000 go ahead and apply this early. 506 00:26:40,000 --> 00:26:42,000 Stopping where? 507 00:26:42,000 --> 00:26:43,000 Go ahead and see only the loss value. 508 00:26:43,000 --> 00:26:49,000 If it is not decreasing, let's say for the five epochs itself, then you can stop the training there 509 00:26:49,000 --> 00:26:49,000 itself. 510 00:26:49,000 --> 00:26:52,000 So and that is the reason why we use early stopping. 511 00:26:52,000 --> 00:26:56,000 So here I'm just going to create my early stopping callback okay. 512 00:26:56,000 --> 00:26:58,000 So this is just like a callback that we use. 513 00:26:58,000 --> 00:27:02,000 And here uh we will go ahead and write early stopping. 514 00:27:02,000 --> 00:27:07,000 We will go ahead and monitor one very important parameter while we are training, right, which is called 515 00:27:07,000 --> 00:27:10,000 as validation loss val underscore loss. 516 00:27:10,000 --> 00:27:13,000 I will talk about what all keys is also getting created. 517 00:27:13,000 --> 00:27:14,000 Okay. 518 00:27:14,000 --> 00:27:16,000 Um like you'll be able to see multiple keys. 519 00:27:16,000 --> 00:27:18,000 Uh you'll have validation loss. 520 00:27:18,000 --> 00:27:19,000 You'll have training loss and all. 521 00:27:19,000 --> 00:27:22,000 But right now I'm just going to monitor this validation loss. 522 00:27:22,000 --> 00:27:27,000 And let's say I'm saying hey be patient for at least five epochs, okay. 523 00:27:27,000 --> 00:27:29,000 Before taking that particular decision. 524 00:27:29,000 --> 00:27:33,000 If till the five epochs you see there is no improvement, then you can stop it over there. 525 00:27:33,000 --> 00:27:33,000 Okay. 526 00:27:33,000 --> 00:27:38,000 And then one more parameter that I'm actually going to say hey, it should restore the best weights. 527 00:27:38,000 --> 00:27:43,000 So when you are, uh, going through the forward and the backward propagation which at which epoch, 528 00:27:43,000 --> 00:27:46,000 you find out the best weight, you can probably consider that. 529 00:27:46,000 --> 00:27:49,000 And you can reload it when you stop the early stopping. 530 00:27:49,000 --> 00:27:49,000 Okay. 531 00:27:49,000 --> 00:27:53,000 So I'm just going to use all these parameters and do the early stopping right. 532 00:27:53,000 --> 00:27:54,000 Very important. 533 00:27:54,000 --> 00:28:01,000 So this will be my early stopping that I'll use now finally now this is the stage of training the model 534 00:28:01,000 --> 00:28:08,000 okay I've created my step by step what I've actually done I have created my entire model. 535 00:28:08,000 --> 00:28:14,000 I have said, hey, which, uh, optimizes I'm going to use, which loss I'm going to use and I've, 536 00:28:15,000 --> 00:28:20,000 I've assigned that entire optimizer and loss to my model along with my metrics. 537 00:28:20,000 --> 00:28:20,000 Okay. 538 00:28:21,000 --> 00:28:24,000 Then I'm saying, hey, go ahead and set up the TensorBoard. 539 00:28:24,000 --> 00:28:28,000 Um, you have to probably create all the logs while training. 540 00:28:28,000 --> 00:28:30,000 So this is the logs. 541 00:28:30,000 --> 00:28:33,000 And from this particular logs only we will try to visualize some of the data. 542 00:28:33,000 --> 00:28:35,000 Now here I have created TensorFlow callback. 543 00:28:35,000 --> 00:28:37,000 And here I have actually created Earlystopping callback. 544 00:28:37,000 --> 00:28:41,000 And I've told you what is the importance of early stopping callback. 545 00:28:41,000 --> 00:28:47,000 Now for training the model, I will just go ahead and say, hey, history is equal to nothing but model 546 00:28:47,000 --> 00:28:47,000 dot fit. 547 00:28:48,000 --> 00:28:48,000 Okay. 548 00:28:48,000 --> 00:28:56,000 And here, uh, I'm going to give my X train comma Y train okay. 549 00:28:56,000 --> 00:28:57,000 Comma. 550 00:28:57,000 --> 00:29:00,000 Along with this I'm also going to give my validation underscore data. 551 00:29:02,000 --> 00:29:05,000 Validation on the score data will be the combination of x test. 552 00:29:05,000 --> 00:29:07,000 Comma y test. 553 00:29:07,000 --> 00:29:12,000 Understand this y test data will be used for comparison with respect to the accuracy. 554 00:29:12,000 --> 00:29:15,000 Then I'm going to say hey, run it for 100 epochs. 555 00:29:15,000 --> 00:29:17,000 Let's say I'm saying go ahead and run it for 100 epochs. 556 00:29:17,000 --> 00:29:21,000 And finally I'm also going to apply my callbacks. 557 00:29:21,000 --> 00:29:26,000 So this callbacks is nothing but two callbacks I'm going to apply. 558 00:29:26,000 --> 00:29:27,000 One is TensorFlow callback. 559 00:29:29,000 --> 00:29:35,000 And the other one is specifically my other callback, which I am going to use, which is nothing but 560 00:29:35,000 --> 00:29:37,000 early stopping callback. 561 00:29:37,000 --> 00:29:42,000 Okay, so this two callback I'm going to basically see. 562 00:29:42,000 --> 00:29:49,000 And then uh, after we train this entire model, uh, we will load our entire TensorBoard that I will 563 00:29:49,000 --> 00:29:51,000 show you how it is basically done. 564 00:29:51,000 --> 00:29:53,000 But over here you will be able to see my logs. 565 00:29:53,000 --> 00:29:55,000 File will start getting created. 566 00:29:55,000 --> 00:29:55,000 Okay. 567 00:29:55,000 --> 00:30:00,000 So let's quickly, uh, go ahead and uh, you know, execute this. 568 00:30:00,000 --> 00:30:02,000 And I hope everyone things should work fine. 569 00:30:02,000 --> 00:30:04,000 So I will just go ahead and hit execution. 570 00:30:04,000 --> 00:30:08,000 Now here you can see my epochs has started right. 571 00:30:08,000 --> 00:30:10,000 Initially I'm getting the accuracy over here. 572 00:30:10,000 --> 00:30:11,000 Validation accuracy is over here. 573 00:30:12,000 --> 00:30:20,000 And uh, you can see I told to run it for 100 epochs, but this is just stopped at seven. 574 00:30:20,000 --> 00:30:21,000 Okay, let me do one thing. 575 00:30:21,000 --> 00:30:24,000 Let me increase this patience level to ten. 576 00:30:24,000 --> 00:30:25,000 Okay? 577 00:30:25,000 --> 00:30:28,000 I'm saying, hey, try to just see till ten epochs. 578 00:30:28,000 --> 00:30:29,000 Now let's go ahead and execute it. 579 00:30:30,000 --> 00:30:32,000 So here you can see accuracy is 85 validation. 580 00:30:32,000 --> 00:30:37,000 This one I think we are getting the best validation loss till here okay. 581 00:30:37,000 --> 00:30:39,000 It will wait for the ten. 582 00:30:39,000 --> 00:30:42,000 Now in the 16th 17th epoch, it has stopped, right? 583 00:30:42,000 --> 00:30:44,000 And this is all the values that I get. 584 00:30:44,000 --> 00:30:45,000 I get my loss. 585 00:30:45,000 --> 00:30:48,000 So till 87% I am able to get it. 586 00:30:48,000 --> 00:30:49,000 Validation loss. 587 00:30:49,000 --> 00:30:51,000 I am able to get this much validation accuracy. 588 00:30:51,000 --> 00:30:54,000 I am able to get 0.85 right. 589 00:30:55,000 --> 00:31:01,000 So these are all the information that, uh, you will be seeing over here that we are able to get it 590 00:31:01,000 --> 00:31:02,000 right, which is amazing. 591 00:31:02,000 --> 00:31:06,000 And here you can see that I am able to train the neural network quickly as possible, right? 592 00:31:06,000 --> 00:31:08,000 Uh, let's do one thing. 593 00:31:08,000 --> 00:31:10,000 Let's save this entire model file. 594 00:31:10,000 --> 00:31:13,000 So I will say, hey, go ahead and save this model dot save. 595 00:31:14,000 --> 00:31:17,000 Um, and I'm going to probably create one h5 file. 596 00:31:17,000 --> 00:31:20,000 So h5 file will be compatible with Keras. 597 00:31:20,000 --> 00:31:22,000 So that is the reason I'm actually creating it. 598 00:31:22,000 --> 00:31:24,000 So here you can also see model dot h5 file. 599 00:31:24,000 --> 00:31:27,000 Okay uh let's do one thing. 600 00:31:27,000 --> 00:31:28,000 Let's load. 601 00:31:29,000 --> 00:31:34,000 load TensorBoard extension. 602 00:31:34,000 --> 00:31:34,000 Okay. 603 00:31:37,000 --> 00:31:38,000 So let's load this. 604 00:31:38,000 --> 00:31:47,000 So for loading I will just go ahead and write load uh load underscore extension. 605 00:31:47,000 --> 00:31:51,000 And here I'm just going to use TensorBoard. 606 00:31:51,000 --> 00:31:59,000 Now this extension also we need to install in our uh so here you can see line magic function uh load 607 00:31:59,000 --> 00:32:03,000 underscore extension not found I think the spelling is wrong over here. 608 00:32:03,000 --> 00:32:04,000 I just need to write like this. 609 00:32:04,000 --> 00:32:07,000 So here you can see I have launched my TensorBoard session. 610 00:32:07,000 --> 00:32:09,000 It looks something like this. 611 00:32:09,000 --> 00:32:15,000 Then what I will do, I will just go ahead and say, hey, take this TensorBoard, my magic function. 612 00:32:15,000 --> 00:32:15,000 Okay. 613 00:32:16,000 --> 00:32:19,000 And along with this, use this log directory. 614 00:32:19,000 --> 00:32:24,000 And inside that log directory I have my logs slash fit folder okay. 615 00:32:25,000 --> 00:32:28,000 So this is how you load this entire tensor board. 616 00:32:28,000 --> 00:32:31,000 Now if you see inside this logs this is nothing. 617 00:32:31,000 --> 00:32:36,000 But it has your entire information with respect to training and with respect to validation. 618 00:32:36,000 --> 00:32:36,000 Okay. 619 00:32:36,000 --> 00:32:41,000 And it is just going to load this and it is showing going to show us in this visualized format. 620 00:32:41,000 --> 00:32:42,000 Let's see this. 621 00:32:42,000 --> 00:32:42,000 Okay. 622 00:32:43,000 --> 00:32:48,000 So here, uh, you will be able to see it I think it is getting loaded. 623 00:32:48,000 --> 00:32:54,000 Let's see whether it is getting loaded or not so quickly. 624 00:32:54,000 --> 00:32:56,000 I think it should get loaded. 625 00:32:56,000 --> 00:32:56,000 Yes. 626 00:32:56,000 --> 00:33:01,000 So it says hey uh um. 627 00:33:01,000 --> 00:33:01,000 Mhm. 628 00:33:01,000 --> 00:33:05,000 So it says hey no dashboards are active for the current data. 629 00:33:05,000 --> 00:33:08,000 The problem may be let's go ahead and launch this. 630 00:33:09,000 --> 00:33:09,000 Uh sorry. 631 00:33:09,000 --> 00:33:11,000 Let's go. 632 00:33:11,000 --> 00:33:12,000 Okay. 633 00:33:12,000 --> 00:33:15,000 See what you have to do is that let me click it over here. 634 00:33:15,000 --> 00:33:19,000 Click over here right launch TensorBoard session okay. 635 00:33:20,000 --> 00:33:24,000 Now in this TensorBoard session just click and launch over here. 636 00:33:24,000 --> 00:33:28,000 You'll be able to see use Current Working directory or select another folder. 637 00:33:28,000 --> 00:33:31,000 I will go ahead and select another folder and let me just go ahead and click this. 638 00:33:31,000 --> 00:33:34,000 And let me just go ahead and inside this particular folder. 639 00:33:34,000 --> 00:33:41,000 And let's go ahead and select this train folder okay I will say hey just go ahead and load this train 640 00:33:41,000 --> 00:33:41,000 folder. 641 00:33:41,000 --> 00:33:43,000 Let's see whether it will load or not. 642 00:33:43,000 --> 00:33:47,000 I think it is not able to find out this log fit directory which is already there. 643 00:33:47,000 --> 00:33:48,000 Okay. 644 00:33:48,000 --> 00:33:49,000 That is the problem. 645 00:33:49,000 --> 00:33:52,000 See okay okay I got it now okay. 646 00:33:52,000 --> 00:33:54,000 The problem is what Right. 647 00:33:54,000 --> 00:33:57,000 It is trying to find out this log slash fit folder. 648 00:33:57,000 --> 00:34:00,000 But this is the folder name that I'm actually getting. 649 00:34:00,000 --> 00:34:04,000 So let me just go ahead and write log fit folder okay. 650 00:34:06,000 --> 00:34:13,000 Let's see I think it should uh, replace the name or I can just give that particular folder name over 651 00:34:13,000 --> 00:34:18,000 here, even though if I execute it here, you'll be able to see that I'll not be able to get it because 652 00:34:18,000 --> 00:34:19,000 this is getting renamed. 653 00:34:19,000 --> 00:34:19,000 Okay. 654 00:34:20,000 --> 00:34:22,000 So let me quickly rename this. 655 00:34:22,000 --> 00:34:25,000 So I will say, hey, reveal in the File Explorer. 656 00:34:25,000 --> 00:34:29,000 Uh uh, inside this logs folder, this is the folder name. 657 00:34:29,000 --> 00:34:30,000 And here is my events. 658 00:34:30,000 --> 00:34:32,000 That is the information that is coming up. 659 00:34:32,000 --> 00:34:33,000 Okay. 660 00:34:33,000 --> 00:34:35,000 So quickly, uh, let me do one thing. 661 00:34:35,000 --> 00:34:37,000 Let me rename it from here. 662 00:34:37,000 --> 00:34:39,000 That would be a little bit faster. 663 00:34:41,000 --> 00:34:46,000 I think it will not allow because the file may be getting used somewhere. 664 00:34:46,000 --> 00:34:48,000 Okay, so let me do one thing. 665 00:34:48,000 --> 00:34:53,000 Let me quickly rename this instead of rename. 666 00:34:53,000 --> 00:34:54,000 Right? 667 00:34:54,000 --> 00:34:58,000 I will just go ahead and take this name entirely okay. 668 00:34:58,000 --> 00:35:06,000 And now I will go back over here and say instead of fit you just go ahead and use this entire name, 669 00:35:06,000 --> 00:35:07,000 okay? 670 00:35:07,000 --> 00:35:13,000 So now I think it should be able to load it because I've given the correct loading directory. 671 00:35:13,000 --> 00:35:15,000 See, I'm not going to edit all these things. 672 00:35:15,000 --> 00:35:20,000 Whatever error that I'm actually facing, I will be showing you so that you will be able to execute 673 00:35:20,000 --> 00:35:20,000 it. 674 00:35:20,000 --> 00:35:21,000 Okay. 675 00:35:21,000 --> 00:35:22,000 That is my main aim. 676 00:35:22,000 --> 00:35:22,000 Okay. 677 00:35:24,000 --> 00:35:28,000 I don't know why we have to probably execute two times or it was getting loaded. 678 00:35:28,000 --> 00:35:29,000 Now here you can see. 679 00:35:29,000 --> 00:35:30,000 beautiful. 680 00:35:30,000 --> 00:35:36,000 You are getting all this diagrams see so here based on epoch accuracy. 681 00:35:36,000 --> 00:35:37,000 Let's go ahead and see this. 682 00:35:37,000 --> 00:35:39,000 Let's say I will just uncheck it. 683 00:35:39,000 --> 00:35:42,000 And here I will just go ahead and write epoch accuracy. 684 00:35:45,000 --> 00:35:48,000 Now with respect to this epoch accuracy you will be able to see. 685 00:35:48,000 --> 00:35:49,000 Let me do one thing. 686 00:35:49,000 --> 00:35:53,000 Let me just uncheck this okay. 687 00:35:53,000 --> 00:35:55,000 So this is my epoch accuracy. 688 00:35:55,000 --> 00:35:56,000 This is my epoch loss. 689 00:35:57,000 --> 00:35:57,000 Okay. 690 00:35:57,000 --> 00:36:02,000 Now see with respect to this epoch loss, uh, you'll be able to see my loss is decreasing. 691 00:36:02,000 --> 00:36:03,000 This is good, right? 692 00:36:04,000 --> 00:36:08,000 Uh, if I close this, uh, over here also, you'll be able to see. 693 00:36:08,000 --> 00:36:09,000 Let's go ahead and use this. 694 00:36:10,000 --> 00:36:11,000 An epoch accuracy. 695 00:36:11,000 --> 00:36:12,000 You'll be able to see this graph. 696 00:36:12,000 --> 00:36:17,000 See, it is increasing from 0.855 to 0.8655. 697 00:36:17,000 --> 00:36:18,000 So this is good. 698 00:36:18,000 --> 00:36:20,000 It is going in this direction right. 699 00:36:20,000 --> 00:36:26,000 And uh you can see what relative step, what value, what accuracy you have actually got. 700 00:36:26,000 --> 00:36:29,000 Every information is basically getting stored over here. 701 00:36:30,000 --> 00:36:34,000 You can also change this value with respect to smoothing. 702 00:36:34,000 --> 00:36:37,000 And here you will be able to see how the accuracy changes. 703 00:36:37,000 --> 00:36:37,000 Right. 704 00:36:38,000 --> 00:36:40,000 See these are some of the parameters. 705 00:36:41,000 --> 00:36:42,000 This is also something very important. 706 00:36:42,000 --> 00:36:44,000 Uh, the dense underscore three. 707 00:36:44,000 --> 00:36:44,000 Right. 708 00:36:45,000 --> 00:36:52,000 This, uh, uh, it just looks like, uh, say, uh, if you have seen this histogram, it just looks 709 00:36:52,000 --> 00:36:54,000 like a gradient descent in opposite manner. 710 00:36:54,000 --> 00:36:54,000 Right. 711 00:36:54,000 --> 00:36:59,000 This is the maximum top pick and this is the minimum loss we are able to get right. 712 00:36:59,000 --> 00:37:00,000 Which looks like this over here. 713 00:37:01,000 --> 00:37:04,000 So here also you can probably observe each and every this. 714 00:37:04,000 --> 00:37:08,000 But the main thing is epoch accuracy and epoch loss which you are able to see it over here. 715 00:37:08,000 --> 00:37:11,000 Not only that, uh, let's do one thing. 716 00:37:11,000 --> 00:37:14,000 Let's, uh, instead of taking train, let's go ahead and see validation. 717 00:37:14,000 --> 00:37:16,000 Now in validation you have this epoch accuracy. 718 00:37:17,000 --> 00:37:22,000 Um, so here again there is some up and down, but at least you are able to get it to 0.855. 719 00:37:23,000 --> 00:37:29,000 Um, then you also have this epoch loss, epoch accuracy versus iteration, epoch loss versus iteration. 720 00:37:29,000 --> 00:37:31,000 So please go ahead and check this out. 721 00:37:31,000 --> 00:37:32,000 Uh, entirely. 722 00:37:32,000 --> 00:37:34,000 I've already shown you how to do it. 723 00:37:34,000 --> 00:37:36,000 Uh, first of all, you need to set this entire logs. 724 00:37:36,000 --> 00:37:38,000 Whatever folder name you are probably getting. 725 00:37:38,000 --> 00:37:40,000 You have to probably use that. 726 00:37:40,000 --> 00:37:40,000 Okay. 727 00:37:41,000 --> 00:37:45,000 Um, but still, you need to think why that folder came in that way. 728 00:37:45,000 --> 00:37:46,000 Okay? 729 00:37:46,000 --> 00:37:46,000 Okay. 730 00:37:46,000 --> 00:37:48,000 I was using this plus, right? 731 00:37:48,000 --> 00:37:54,000 Some date sign over here, uh, automatically, with the help of date sign, you could see that, uh, 732 00:37:55,000 --> 00:37:56,000 instead of writing like this, right? 733 00:37:56,000 --> 00:37:58,000 I could have written something like this. 734 00:37:58,000 --> 00:37:59,000 I'll keep this code like this. 735 00:37:59,000 --> 00:37:59,000 Okay. 736 00:37:59,000 --> 00:38:00,000 It will be slash. 737 00:38:00,000 --> 00:38:03,000 And inside that, any date format will be created. 738 00:38:03,000 --> 00:38:04,000 Okay, perfect. 739 00:38:04,000 --> 00:38:12,000 Uh, till here, I think everything looks good, but, uh, always make sure that if I am using this 740 00:38:12,000 --> 00:38:16,000 folder wise over here, let me just show you again. 741 00:38:17,000 --> 00:38:18,000 Please make this change. 742 00:38:18,000 --> 00:38:23,000 So I'm saying, hey, inside the fit folder, you have to probably create every time date, time stamp. 743 00:38:23,000 --> 00:38:25,000 And then you have to probably go ahead and create your logs. 744 00:38:25,000 --> 00:38:31,000 So in this case, uh, you don't even have to use this, right. 745 00:38:32,000 --> 00:38:33,000 You'll be able to get it perfect. 746 00:38:34,000 --> 00:38:37,000 So, uh, this was about this one. 747 00:38:37,000 --> 00:38:38,000 We have also saved the pickle file. 748 00:38:38,000 --> 00:38:41,000 Uh, we have done almost each and every thing. 749 00:38:41,000 --> 00:38:44,000 Uh, and this is how, uh, an neural network is basically created, right?