WEBVTT 00:01.010 --> 00:02.240 Hey, everyone, welcome back. 00:02.930 --> 00:06.890 So in this video, we will see how we can train our own and model. 00:07.610 --> 00:13.220 And mostly we are going to use for our own training and with our custom data set. 00:13.520 --> 00:18.940 We are going to use this Jansing library so first time and put it dispenser filled with Taldo. 00:19.070 --> 00:24.170 We are not going to use this tense of law, but let me put it in this by default. 00:24.250 --> 00:24.900 Mundos. 00:26.290 --> 00:32.480 As a part of this club environment and defiantly, you will get this sense of low two point zero percent 00:32.920 --> 00:34.010 now for preprocessing. 00:34.120 --> 00:35.980 We are going to use this hand LDK library. 00:35.980 --> 00:39.680 So I'm just going to import this and indicate and return LDK. 00:40.090 --> 00:41.980 That is a one more package. 00:41.980 --> 00:42.940 I am going to download it. 00:43.000 --> 00:44.170 That is nothing but fun. 00:44.500 --> 00:47.930 So that is basically for the organization purpose. 00:47.960 --> 00:49.210 And we are going to use this. 00:50.640 --> 00:51.000 Martin. 00:51.660 --> 00:52.560 All right now. 00:52.670 --> 00:54.620 Important parties about the Jency. 00:55.170 --> 01:01.410 So from Jansing models, we are going to import this right to work and give to function. 01:01.830 --> 01:03.060 So let me embody. 01:03.870 --> 01:05.630 So, Deserto, some of the basics input. 01:06.210 --> 01:11.180 Now, first important us, we are going to start one is a data people assessing. 01:11.760 --> 01:14.340 And for that one database, we are going to use it. 01:14.540 --> 01:17.070 That is available on this Kagle website. 01:17.130 --> 01:20.120 So let us follow this Kaggle website. 01:20.400 --> 01:22.620 And it will open a new tab. 01:23.670 --> 01:31.140 Now, to get any Natus say inside the gaggle, you need a calm before that, you just cannot download 01:31.160 --> 01:31.300 it. 01:31.710 --> 01:38.370 So we are going to use this voice news dataset on ready from two thousand eight to today. 01:39.150 --> 01:45.200 So there is a huge amount of data sets that are available for our vote embedding training model, and 01:45.210 --> 01:47.160 that will be close to around 70 M.B. 01:47.610 --> 01:52.090 So our first task is to get this particular data set inside the collar. 01:52.350 --> 01:52.860 And what I mean. 01:53.480 --> 01:53.850 All right. 01:54.240 --> 02:02.490 And to get it inside the clip file, we are going to use this Kagle API and that is with the help of 02:02.490 --> 02:04.260 this Kagle package module. 02:04.290 --> 02:10.450 So let me put this Google package module with this package manager and tell this installation will complete. 02:10.860 --> 02:19.890 First is we how to grab the token so you can go to your account, Maicon, and just scroll down and 02:19.890 --> 02:25.100 you'll be able to see you'll be you will get this API key. 02:25.140 --> 02:32.070 So just create new EPA token just to download this net asset into our column environment. 02:32.880 --> 02:40.470 And I'm just going to say let me say to the download folder right now. 02:41.220 --> 02:44.070 Next thing is we are going to create one, not Google. 02:45.190 --> 02:45.820 Directory. 02:47.110 --> 02:48.900 And our Kegl dot is on fire. 02:48.940 --> 02:53.080 We are just going to cooperate and into this dark Kagle directly. 02:53.680 --> 02:57.160 But before that, from our local environment, we have to upload it. 02:58.150 --> 02:58.890 So applaud. 02:59.270 --> 03:04.090 Go to download and our piggin, not just one site is available. 03:04.450 --> 03:07.060 Now, if you open this Google, not just on fire. 03:08.080 --> 03:12.520 That is only a user named Bill Gavin and part from user name. 03:12.530 --> 03:14.380 There is a one API key. 03:15.300 --> 03:15.770 All right. 03:16.210 --> 03:18.260 So let's just copy now. 03:18.610 --> 03:22.830 And after copying, let's just disable this particular API key. 03:23.110 --> 03:25.240 So we are just changing the permission. 03:26.140 --> 03:32.280 Let me call OPSEC and now we are going to download this particular dataset. 03:32.740 --> 03:34.230 Avoid news on Reddit. 03:34.810 --> 03:36.100 So let me go back. 03:36.420 --> 03:36.830 Yes. 03:37.150 --> 03:38.770 So if you click on download. 03:39.940 --> 03:44.460 You will be able to download in your local machine on, so if you want to create this whole model on 03:44.460 --> 03:51.810 a local machine and if you want to download on a Kagle magically into a club environment, you can use 03:51.810 --> 03:53.710 this coming Kaggle data set now. 03:54.750 --> 04:01.130 So it is close around seventy eight M.B and it will take little time to download. 04:02.640 --> 04:04.790 All right, so the data got downloaded. 04:05.720 --> 04:07.930 We can even see whether data is available yet. 04:08.090 --> 04:11.870 Yes, data is available via news on Reddit GFI. 04:12.440 --> 04:19.840 Now, let's just unzip it and we will keep it inside this contain vital news on a ready for you. 04:20.270 --> 04:21.100 Let me finish it. 04:21.380 --> 04:23.770 So our hopes. 04:24.180 --> 04:27.770 Let's go to contain and inside the contain. 04:28.130 --> 04:32.840 You can see the file got chipped in two nights since we filed for me. 04:33.320 --> 04:34.340 Now, next, this. 04:34.430 --> 04:35.270 We are going to. 04:36.690 --> 04:44.130 Read this NAACP file into a data free object for reading, we are going to use this bindis library. 04:44.570 --> 04:46.450 Let this play forcefield, Akos. 04:47.750 --> 04:50.380 All right, so it's a race that has a nail. 04:50.430 --> 04:51.040 Goodness. 04:51.890 --> 04:57.110 And mainly we are going to use this title column for our war and reading Modern. 04:57.830 --> 05:00.920 How many total number of records are available? 05:00.950 --> 05:03.470 We can grab it from ship. 05:03.950 --> 05:10.490 So we have a file X nine thousand two hundred and thirty six records are available. 05:10.520 --> 05:11.930 That's quite good enough. 05:12.350 --> 05:15.830 And this is the only column which contains all those title. 05:16.580 --> 05:21.740 So we'll just grab this title column into another variable news and this called title. 05:22.060 --> 05:22.490 All right. 05:22.970 --> 05:24.770 If you just try to display. 05:25.920 --> 05:27.960 Let's say first Fluticasone leaving today. 05:28.910 --> 05:31.980 So scores killed in a Pakistan class. 05:32.330 --> 05:32.480 Yes. 05:32.580 --> 05:33.840 So this is the first record. 05:34.380 --> 05:37.650 Now we are going to first tokenized this one. 05:38.130 --> 05:42.880 So for tokenization, just like earlier or so, we use this analytical library. 05:43.140 --> 05:49.670 So from an indicator not PAVOL and the called tokenization, we are going to do and it will do the tokenization 05:49.680 --> 05:51.870 on every single record. 05:52.290 --> 05:54.660 So we'll get a director like new on this. 05:54.740 --> 05:55.200 Correct. 05:55.470 --> 05:58.350 Now it has a good amount of datasets that are available. 05:58.710 --> 06:01.260 So it will take a little bit amount of time. 06:03.620 --> 06:06.330 All right, so now you can see tokenization, Fenice. 06:07.050 --> 06:08.660 Let's display first few records. 06:08.950 --> 06:09.720 So a new. 06:10.850 --> 06:11.640 That's correct. 06:12.380 --> 06:13.340 Let's just deal. 06:14.680 --> 06:16.770 So very first record will be released. 06:17.080 --> 06:20.260 And having our individual values will be a different voice. 06:20.270 --> 06:23.320 What I would say in a very NLB language. 06:23.440 --> 06:24.640 It will be kind of took. 06:24.680 --> 06:28.660 So scores killed in Pakistan and classes. 06:28.930 --> 06:31.510 So each individual word is considered as a token. 06:31.960 --> 06:35.870 The next is we're going to build the actual vote work model. 06:36.310 --> 06:40.420 And this will direct model functionalities available in this Jency library. 06:40.920 --> 06:45.210 Right now, it is expecting this to argument ventilatory arguments. 06:45.240 --> 06:47.520 So first one is for the text. 06:48.190 --> 06:53.590 So tax will be a new underscore quarterback model or I would say text. 06:53.650 --> 06:57.400 We will supply one of the two argument will be minimum gone. 06:57.760 --> 07:03.040 So it will consider only those where we're having a minimum count will be one. 07:03.070 --> 07:09.070 So obviously every single word or unique way it will take into consideration and the size of vector 07:09.070 --> 07:09.770 will be 32. 07:09.850 --> 07:17.110 Because after embedding whatever the output we get inside the model, that will be OK size thirty two 07:17.830 --> 07:19.610 numbers in terms of timings. 07:19.630 --> 07:26.370 And so he tries to say that every single word or a token will be represented. 07:26.570 --> 07:26.760 Wait. 07:27.260 --> 07:30.370 Thirty two numbers are attracted to damage that space. 07:30.850 --> 07:34.270 So let me execute it and it will take good amount of time. 07:34.570 --> 07:39.880 So I'm just winding up in this video at this particular time. 07:40.270 --> 07:46.300 And in the next video, we will see after the model will be created how we can do a prediction based 07:46.300 --> 07:47.680 on this particular model. 07:47.990 --> 07:49.570 So see you in the next video.