WEBVTT 00:00.780 --> 00:01.810 All right, everyone. 00:01.830 --> 00:05.650 So in the last video, we have successfully installed this first text library. 00:06.190 --> 00:10.800 Now we are going to use this first text to do a text classification. 00:10.800 --> 00:16.830 And for that, we are going to use the same review, a restaurant review files that I just made some 00:16.830 --> 00:18.070 modification in that file. 00:18.090 --> 00:25.260 Let me just open it and I'll share this with you, because this text requires data in such a kind of 00:25.290 --> 00:25.970 format. 00:25.980 --> 00:28.770 So I just formatted all those data in this way. 00:29.220 --> 00:33.470 So you can see corresponding to level one, we have a ham data. 00:34.140 --> 00:39.810 Either way, I would say positive reviews and corresponding to zero. 00:39.810 --> 00:41.430 It's a negative review. 00:41.760 --> 00:45.450 This way there are a thousand records are available. 00:46.070 --> 00:46.410 All right. 00:46.890 --> 00:56.970 So next thing is, let me go to the terminal or will the funding will create a new folder and up past 00:57.150 --> 00:57.480 X. 01:01.540 --> 01:07.150 Or I will write it like a tax classification. 01:08.320 --> 01:16.780 All right, and let me first go to our first tax where we are installed as fast tax. 01:16.780 --> 01:20.020 So it will be in a city fast tax. 01:20.620 --> 01:21.290 Let's go. 01:21.580 --> 01:22.180 What is there? 01:22.180 --> 01:22.730 Inside there? 01:23.530 --> 01:25.480 So inside the first tax, there will be one more. 01:26.270 --> 01:32.770 Let me go inside the box and you can see there are a number of files that are available. 01:33.160 --> 01:34.660 We can just simply type. 01:37.430 --> 01:47.750 Foster hopes in this case it didn't work, so you can just simply type notes and you can see now there 01:47.750 --> 01:49.370 are a number of options provided. 01:49.370 --> 01:54.590 That means these are the things you can do with this tax levy so far is like a on. 01:54.920 --> 01:56.600 You can provide a different arguments. 01:56.990 --> 01:59.570 So Carmines will be something like a fast tax. 01:59.570 --> 02:04.370 You can do the supervised learning, you can test your different models. 02:04.370 --> 02:05.720 You can have a prediction. 02:05.720 --> 02:06.950 You can apply this Kebra. 02:07.310 --> 02:11.500 So Majority will try to see in this video and upcoming videos. 02:12.020 --> 02:14.000 So first thing what we are going to do. 02:14.480 --> 02:18.050 Let me go to the tax classification and remember this. 02:18.050 --> 02:25.970 But every time we are just going to fire this pot to fire the first tax before that, let me go back 02:26.480 --> 02:31.550 and let me go to the next top tax classification. 02:31.910 --> 02:36.690 And I'm just going to move this review, start to fight inside the tax classification. 02:37.310 --> 02:37.790 All right. 02:38.030 --> 02:43.070 Let me put tallies here and I have reviewed not the file available with me. 02:44.030 --> 02:50.240 All right, let me put get and it will display this previews, not testify. 02:50.480 --> 02:51.050 All right. 02:51.060 --> 02:52.050 So it's all available. 02:52.850 --> 02:54.470 Next thing, what we are going to do. 02:54.830 --> 02:58.880 Let me clear the screen, put this data into two different buckets. 02:59.090 --> 03:06.080 So something like some 70 percent of your data will put it inside the cleaning basket and remaining 03:06.080 --> 03:07.100 on our testing basket. 03:07.460 --> 03:11.010 So what we will do, we are going to use this hard. 03:11.180 --> 03:11.660 Come on. 03:12.480 --> 03:14.750 Let me make it a little bigger screen. 03:16.640 --> 03:22.850 Head first, the let's say seven hundred record, I just keep it inside the. 03:23.870 --> 03:24.570 Cleaning bucket. 03:24.890 --> 03:31.970 So I'm going to take the data from reviewing the text and put this error so it will take first 700 recalled 03:31.970 --> 03:39.280 because of this her come on and let's say it will put it inside the reviews, not drain file. 03:39.770 --> 03:40.880 Don't worry about the expense. 03:41.000 --> 03:41.870 It just the name. 03:41.870 --> 03:50.540 There is nothing like extensive and from the last or Tail-End will take another to ninety nine because 03:50.540 --> 03:53.870 I think we have a nine hundred and ninety nine records are available. 03:54.320 --> 03:56.360 So what will do reviews. 03:56.360 --> 04:00.550 Not the last two ninety nine data point. 04:01.130 --> 04:02.960 We will put it into texting. 04:03.590 --> 04:08.270 So that will be inside the reviews, not what it will be a test. 04:09.100 --> 04:09.770 All right. 04:10.070 --> 04:15.580 Now if you go to text classification will have now three files. 04:15.890 --> 04:21.540 So we are going to work with these reviews that trained first to build a model and then we'll work with 04:21.540 --> 04:23.000 the reviews that test. 04:24.950 --> 04:32.960 So our task is to create a classification model with which reviews doctrine before that, I want to 04:33.260 --> 04:41.420 show you that if we want to run this fast library directly from this spot, we can do home UNGEI and 04:41.420 --> 04:45.560 we have four stacks inside the first text via one more text. 04:45.560 --> 04:49.790 And you can see every time when we want to fight this campaign, we'll fight it like this. 04:50.010 --> 04:57.110 Otherwise you can include this fast text inside your party variable also so supervised, supervised 04:57.350 --> 04:57.920 learning. 04:58.190 --> 05:00.260 We are going to train our classifier. 05:01.450 --> 05:06.770 So come on, which we are going to use for training, our classifier will be a supervisor. 05:06.970 --> 05:10.510 So it will be supervised. 05:12.760 --> 05:14.860 Now, this requires two things. 05:15.340 --> 05:16.800 One will be your input file. 05:17.080 --> 05:20.380 So obviously our input file will be reviewed doctrine. 05:20.860 --> 05:26.510 And the next one will be what is your output, where you want to put your model. 05:27.130 --> 05:29.710 So let me provide the option. 05:29.710 --> 05:33.100 Or if you don't know what are the options, you can give it with the supervisor. 05:33.100 --> 05:38.800 Come on, you can just simply press with supervisor enter and you will have a whole lot of options are 05:38.800 --> 05:39.280 available. 05:39.280 --> 05:39.970 Which supervisor? 05:40.210 --> 05:46.870 So these are the mandatory options and remaining all might be optional options out there, because as 05:46.870 --> 05:50.200 you are going to train on which data set, you must provide it. 05:50.200 --> 05:55.300 So that's why have an input and an output if the MASKEW argument while. 05:56.230 --> 06:04.750 Training is classified, so one argument will be hyphen input that will be nothing but play reviews, 06:04.750 --> 06:12.700 not train, and the next one will be output to hyphen output and. 06:13.640 --> 06:18.680 What are the output finally, we can give any filename, so I'm going to give you like a model, let's 06:18.680 --> 06:22.340 say model and the second one under review. 06:22.670 --> 06:27.760 Let me give just the model and the score one that's perfectly fine, legendary. 06:27.800 --> 06:34.520 And it will start building a model for us, a supervised learning model based on this class text library 06:34.760 --> 06:40.110 and whatever data we provided of a training model of RVO 700 datapoint. 06:41.540 --> 06:42.080 All right. 06:42.090 --> 06:43.670 So we'll just see it again. 06:43.700 --> 06:48.650 So it just immediately did a very fast all those training. 06:48.650 --> 06:55.520 But if you just do that, you will have model one dot mean and model and it's got one dot. 06:56.030 --> 07:02.360 So as a model, it has created a two model and it's got one dot bin and model and it's called one dot 07:02.840 --> 07:04.040 file it has created. 07:04.250 --> 07:06.470 That means a training process is over. 07:06.680 --> 07:08.480 Now we can use this model. 07:09.930 --> 07:16.980 To build a prediction, so let me just clear it and how we are going to predict it so far that same 07:17.760 --> 07:26.350 home and we have stacks and stacks so we can even create a place for it. 07:26.370 --> 07:28.120 So in Linux, it's impossible. 07:28.400 --> 07:31.410 So let me give you some Malia's name. 07:31.440 --> 07:34.400 Let's see some fast track. 07:34.420 --> 07:35.610 So it will be F.P.. 07:35.640 --> 07:40.620 I I'm just providing here and put this whole carmin inside the. 07:43.410 --> 07:50.420 So Saulius, God created, we can just search for aliens, so we have aliens, did it God created, 07:51.010 --> 07:54.180 yes, God created a physical look how long it takes. 07:54.840 --> 08:00.030 Now I can just simply write like F.T. But obviously I am not sure whether it will work in another time 08:00.030 --> 08:00.540 or not. 08:00.990 --> 08:01.590 So after. 08:02.280 --> 08:07.530 Yeah, it won't work here because it is specific to this one on only the cell. 08:07.830 --> 08:09.370 So let me just clear it. 08:09.720 --> 08:17.400 And now from onwards I'm going to write this feat so every word options are available for prediction. 08:17.400 --> 08:25.350 So we have one command is available like a real one, more commands like a prop and we have one more 08:25.350 --> 08:29.480 like a test so we can have our evolution on our testing dataset. 08:29.490 --> 08:31.950 But before that, let us test on our. 08:33.410 --> 08:44.100 Random data like let's give any data by hardcoded input so happy and I'm going to provide here a pretty 08:45.510 --> 08:51.880 hefty prediction, next one is on using my using which model they are going to do it. 08:51.900 --> 08:56.960 So that will be nothing but model one or model underscore one not been filed. 08:57.510 --> 09:05.250 And as we are not going to provide any text data, so I'll just provide hyphen and it will ask me for 09:05.430 --> 09:08.180 which data point you want to do the classification. 09:08.460 --> 09:16.290 So I will, I will just say like I love food and let's see what we classify. 09:17.070 --> 09:18.090 It is a level one. 09:18.090 --> 09:19.080 That means hillocks. 09:19.590 --> 09:21.240 I do not. 09:22.760 --> 09:28.250 Let me whoops, I do not allow food in a restaurant. 09:29.030 --> 09:33.880 Still, it is classified as the one we can even try with the chili. 09:34.580 --> 09:35.850 We don't have enough data. 09:36.040 --> 09:36.440 Might be. 09:36.440 --> 09:38.870 So I don't like four. 09:40.890 --> 09:47.220 Yes, so it has not taken into consideration, like, I guess do not part, and that's very important. 09:47.410 --> 09:50.220 What I can say, like I d like it. 09:50.430 --> 09:50.970 Let's see. 09:52.080 --> 09:58.170 Still level one only, but I'm not sure about that, but maybe we don't have sufficient data set, but 09:58.170 --> 10:00.420 that is how you can train your classifier. 10:00.900 --> 10:03.080 Let me just come out from here. 10:03.660 --> 10:11.430 Now, apart from that, you can have a predict probability so it will predict the probability that particular 10:11.430 --> 10:15.290 data points belongs to level one or it will be level zero. 10:16.360 --> 10:24.030 Next thing is, we can have a complete based on testing data, so we have all the testing data set. 10:24.040 --> 10:26.920 So you can see this is reviews the test. 10:27.280 --> 10:32.040 Let me open it and I'll show you some data set reviews or test. 10:33.010 --> 10:33.600 All right. 10:33.610 --> 10:36.010 So there are some there are some one. 10:36.790 --> 10:39.570 And let's just try one of them. 10:40.540 --> 10:47.830 So all 299 a record, it will do the calculation or do the prediction and it will finally give us the 10:47.830 --> 10:48.400 accuracy. 10:48.760 --> 10:50.070 So let me try it. 10:50.120 --> 10:58.840 Every test that will be overcome on if we just execute it, it will tell us what are the few more things 10:58.840 --> 10:59.700 you need to provide. 10:59.710 --> 11:02.080 So that will be a model model. 11:02.080 --> 11:02.950 Model filename. 11:05.430 --> 11:14.730 So F.T. test the inviable model filename, so that will be nothing but model one that bin and let me 11:14.730 --> 11:18.150 provide filename, so that will be the test. 11:18.930 --> 11:19.580 All right. 11:19.890 --> 11:26.220 So you can see they have given that are total two hundred and ninety nine data points and four places 11:26.250 --> 11:29.500 and they are given this much and that is nothing but a recall. 11:29.580 --> 11:34.740 So they will be also delivered to Dessaix and present also there would be zero point thirty six. 11:35.460 --> 11:36.060 All right. 11:36.270 --> 11:44.940 Now we are just trying this model without using any modification of default parameter, but there are 11:44.940 --> 11:47.280 a lot of things you can always do. 11:47.670 --> 11:52.170 So let's go to our original commands of supervised learning. 11:52.890 --> 11:53.390 Yes. 11:53.420 --> 11:55.660 So that's our advice. 11:55.950 --> 11:59.250 Now, along with that, there are a number of options you can provide. 11:59.460 --> 12:06.180 So let's say you want to run your model for let's say there will be some default number of people for 12:06.180 --> 12:07.500 this model, Gaudron. 12:07.680 --> 12:13.200 But you can run it for, let's say, a thousand époque so you can have a provision like Let Me provided 12:13.200 --> 12:21.630 another model like Model and Escoto and I will do hyphen people, let's say Chozen. 12:21.930 --> 12:22.590 All right. 12:23.850 --> 12:30.570 Apart from that, let's say while conversion of text into a number of feature and coding or I would 12:30.570 --> 12:38.280 say feature engineering does, I want to use this and grammar so I have one more options, like a hyphen 12:38.280 --> 12:38.700 W. 12:40.520 --> 12:41.000 Void 12:43.910 --> 12:49.590 Ingrams, let's say, bigram model, I want to use it or I want to use Trigram model, I can use it. 12:49.610 --> 12:50.880 Let's give it a two. 12:51.290 --> 12:53.060 So this way there are a number of options. 12:53.060 --> 12:56.450 You can make it and it will create a brand new model for you. 12:57.420 --> 13:04.360 All right, so now you can see this many numbers of people take a little amount of time and we have 13:04.360 --> 13:08.960 a Bigram model it has created, let's test the accuracy again. 13:08.970 --> 13:11.820 So we are going to use this prediction. 13:12.270 --> 13:13.080 Yes, predict. 13:13.680 --> 13:17.580 But instead of using model in the score, one will create it now, model in this two. 13:18.030 --> 13:22.980 And due to that, you can see more than there's got to be not to backfill got created. 13:24.330 --> 13:33.720 Let me give Nene some reviews like I like your food here. 13:33.840 --> 13:34.740 So it has this. 13:35.700 --> 13:41.230 Predicts there will be a label and a second one, I'm not sure about it, but that's OK, maybe some 13:41.230 --> 13:42.920 more trainings are required to address it. 13:43.260 --> 13:48.350 And you can see we just come online from by writing a few comments. 13:48.360 --> 13:55.380 If you have a dataset, you can train your model with this vast text library, and it is one of the 13:55.380 --> 13:56.790 very high performance library. 13:57.890 --> 14:03.560 So this is mainly used for the tax classification task here, we used to classify the review between 14:03.770 --> 14:04.920 positive and negative. 14:05.210 --> 14:10.550 But for any kind of tax classification task, whether it's a binary classification problem or multiclass 14:10.550 --> 14:16.490 classification problem, always you can use it whenever you have a huge amount of tax rate available 14:16.500 --> 14:16.850 with you. 14:17.420 --> 14:17.960 All right. 14:17.990 --> 14:19.310 See you in the next video.