WEBVTT 00:01.010 --> 00:01.840 All right, everyone. 00:01.870 --> 00:05.800 So let's continue our discussion on spam detection with CNN. 00:06.700 --> 00:12.370 So we are reaching their last video, tell this padding sequence on our input data. 00:13.150 --> 00:20.380 And now we can grab this one variable, sort of very second value of this Sape of this data underscore 00:20.430 --> 00:22.830 train if you're trying to paint it. 00:23.880 --> 00:29.970 It will be one, it is nine, because later we are going to use it for the I mean, while doing this 00:29.970 --> 00:35.260 padding sequence on the testing, like I say, so seemed like a training data set. 00:35.760 --> 00:37.970 We are going to use it for the testing that does it. 00:38.010 --> 00:40.860 But in case of testing that has it, we don't know the maximum length. 00:41.580 --> 00:47.580 So we have just supplied the state and these netting what, the Lentol individual records. 00:48.180 --> 00:49.200 Now let me run it. 00:50.850 --> 00:58.640 And you can see we have our land 189 for tasting individual kozo soil and for training on so one eighty 00:58.650 --> 00:58.860 nine. 00:59.520 --> 01:06.260 Now, how many toadflax we have available so we can grab this information from over? 01:06.290 --> 01:08.160 Or is it data from objects. 01:08.820 --> 01:10.680 Let me display here. 01:10.860 --> 01:12.500 If not shape. 01:13.580 --> 01:14.860 So it will be five five. 01:14.860 --> 01:15.540 Seventy two. 01:15.660 --> 01:21.260 And while splitting this dataset we kept like a crediting percentage of data set. 01:21.270 --> 01:22.710 We'll go into testing buckets. 01:22.980 --> 01:25.560 So out of five five seven two data set. 01:26.070 --> 01:33.420 Now we have a eighteen hundred thirty nine course in our testing buckets and three seven three three 01:33.420 --> 01:35.490 records inside the training buckets. 01:35.850 --> 01:40.590 And this training buckets we are going to use for building our modern right. 01:40.950 --> 01:42.930 So now data is clear. 01:43.380 --> 01:48.520 Next is we are going to build our first convolutional neural network model. 01:48.930 --> 01:54.650 But before applying those convolution layer, first thing, we are going to apply Embling layer. 01:55.140 --> 01:57.090 And how we are going to build this model. 01:57.510 --> 02:00.150 So let me give the reference like. 02:01.120 --> 02:03.200 Will create a model object from here. 02:03.880 --> 02:09.730 So that will be a lot of of locate us and a model and how we are going to hide the layers inside. 02:10.420 --> 02:12.840 So that will be a part of tensor flow. 02:12.990 --> 02:14.170 Get us an alias. 02:14.540 --> 02:17.520 Aldo's preprocessing velho earlier only QST. 02:18.370 --> 02:18.720 All right. 02:20.810 --> 02:26.950 So this way of creating model inside the cars that will be called as a functional V operating model. 02:27.290 --> 02:29.290 That will be another V also creating models. 02:29.320 --> 02:31.880 So we will see in the upcoming projects. 02:32.630 --> 02:37.280 So first, let's create the input and. 02:38.370 --> 02:45.030 Input will be key because every single record has been represented by this one eighty nine number. 02:45.090 --> 02:46.200 That is nothing but three. 02:46.650 --> 02:49.190 So that is that will become over input. 02:49.860 --> 02:51.030 Next, we are going to. 02:52.000 --> 02:53.200 And when I'm ready. 02:53.320 --> 02:53.590 Clear. 02:53.840 --> 02:55.340 So this is the value of writing. 02:55.730 --> 03:01.910 Like when you add nuclear, you are just going to use the earlier Lear, like, let's say AISI will 03:01.910 --> 03:03.390 become a bit earlier Lear. 03:04.100 --> 03:05.770 So what is the diamonds and all this? 03:06.450 --> 03:07.020 I'm reading. 03:07.430 --> 03:10.040 What kind of output we are expecting. 03:10.490 --> 03:13.390 So in case of I'm reading this V. 03:13.790 --> 03:17.160 Let me tell you what is V very well defined. 03:17.510 --> 03:18.650 So vs nothing but. 03:21.140 --> 03:27.050 Total unique voice, because here, whatever the sequence we have created, that has been given one 03:27.050 --> 03:27.470 number. 03:27.870 --> 03:35.390 But now when we are building this building model, we have to assign specifically those particular index 03:35.620 --> 03:36.180 politician. 03:36.770 --> 03:43.880 So we are going to add one amazing layer where every single word will be represented. 03:44.570 --> 03:51.500 So replacement, that is just because the next thing here starts from the one not with a zero output 03:51.500 --> 03:56.270 will be these, because whatever the dimensional output we will get. 03:57.020 --> 04:02.210 We want to represent over every single token to 20 dimensional space. 04:02.570 --> 04:06.630 Let's write this amending letter will take care of all those things. 04:07.400 --> 04:10.830 Next, let's have a first convolution layer. 04:11.210 --> 04:14.020 And this type of pre layers we are going to add. 04:14.590 --> 04:19.760 So in the first convolution layer, there are 32 convolution feature. 04:20.240 --> 04:21.940 We will try to extract it. 04:22.550 --> 04:29.580 And each of the feature having a total three values seem like a max pulling. 04:29.660 --> 04:35.420 So out of three values, we will try to take maximum in the next convolution layer. 04:35.540 --> 04:39.750 Sixty four feature each of the features of same size like a tree. 04:40.280 --> 04:45.110 We'll try to extract it in the third convolution lea it seemed like earlier. 04:45.200 --> 04:46.760 So that will be a one. 04:46.970 --> 04:47.660 Twenty eight. 04:48.350 --> 04:51.260 And then we'll just create a dense layer. 04:51.680 --> 04:53.840 So dense layer having output only one. 04:54.210 --> 04:58.130 And here you can see our activism will be a sigmoid. 04:58.460 --> 05:01.700 So in all our earlier cars that we send, you need will be available. 05:02.150 --> 05:07.490 So all those negative value will be suppressed and positive value will be passed on as it is. 05:08.010 --> 05:08.230 Here. 05:08.310 --> 05:09.200 We'll use the sigmoid. 05:09.280 --> 05:12.450 Because we are dealing here with the classification problem. 05:13.310 --> 05:15.590 And that is also binary classification problem. 05:15.920 --> 05:18.020 So either it will be a zero or it will be at one. 05:18.440 --> 05:21.830 So Hollowell layers got defined. 05:21.980 --> 05:23.830 Next is we need to create a model. 05:23.880 --> 05:25.850 So model requires one input. 05:26.180 --> 05:28.090 And all of a layer. 05:28.190 --> 05:30.440 So at the end of a layer will be X. 05:30.860 --> 05:34.130 X is a function of this particular X. 05:34.310 --> 05:36.890 So we layers we are just keep adding. 05:36.920 --> 05:40.750 So this is the one of the V to build a model inside the US. 05:42.240 --> 05:48.870 All right, so now you can see model guard defined next is we need to compile this, Martin. 05:50.310 --> 05:53.490 So for compilers, indirectly, arguments we need to supply. 05:53.940 --> 05:58.650 So when is Optimizer that how we are going to optimize your last function? 05:59.200 --> 06:05.070 Our last function will be a binary cross, and because here we are dealing with the binary classification 06:05.070 --> 06:09.630 problem and how you're going to measure that, how good your model is. 06:09.630 --> 06:14.160 So for that, we are going to use this accuracy misery criteria. 06:14.640 --> 06:16.890 So that will be a modern campaign. 06:18.230 --> 06:21.080 And adding, we need to fit the model. 06:21.320 --> 06:27.230 So here, the interesting or most moving parts, I would say will start like fitting your data. 06:27.440 --> 06:31.970 Let me run it and all of an output will be tracked inside. 06:31.970 --> 06:33.080 This are variable. 06:33.500 --> 06:35.270 So only for the fire pork. 06:35.330 --> 06:36.120 I'm going to run it. 06:36.560 --> 06:39.860 And as a validation data, I am just going to supply this testing. 06:39.970 --> 06:40.270 Does it. 06:41.890 --> 06:44.230 And input data set will be training data set. 06:44.300 --> 06:46.220 So that will be data on the screen. 06:46.450 --> 06:47.380 In a wide discovery. 06:48.340 --> 06:50.250 And you can see we just fine apple. 06:50.560 --> 06:54.250 We got the validation accuracy cluster on ninety eight percent. 06:54.250 --> 06:55.820 Is that that's a very good use. 06:56.380 --> 06:59.470 And for any accuracy, it will be ninety nine point ninety two. 07:00.070 --> 07:04.020 Let's try to display this thing in terms of check. 07:04.600 --> 07:08.080 So here we are trying to display from this automatable. 07:08.290 --> 07:14.020 So our history from there, we are going to get the loss and a values and lost. 07:14.230 --> 07:19.110 So with respect to every single iteration now in our case, the accretion will be just fine. 07:20.110 --> 07:24.400 So at each present and what is the loss and what is a relevation loss? 07:24.700 --> 07:25.750 It will display. 07:27.280 --> 07:32.800 So although the attrition level information will be stored in this are variable, and you can see with 07:32.800 --> 07:38.170 respect to each iteration, the losses and reducing seem like a loss. 07:38.200 --> 07:40.010 We can even display accuracy. 07:40.600 --> 07:42.070 So let me run this one. 07:43.510 --> 07:46.990 And you can see from zero to one, two, two. 07:47.440 --> 07:50.470 Could I see invalidation as it is, keep on increasing. 07:51.370 --> 07:51.760 All right. 07:51.790 --> 07:56.650 So that is all about the how to implement this family, the extent of it convolution. 07:57.850 --> 07:59.230 So see you the next video.