WEBVTT 00:01.160 --> 00:01.740 Hey, everyone. 00:01.800 --> 00:03.630 So let's continue our discussion on. 00:04.580 --> 00:11.600 This classification, which spacy project so we help me clean, was one of our polygonal reviews. 00:11.960 --> 00:16.510 Now we have to apply all those cleaning process on every single day. 00:16.670 --> 00:19.520 Like I said, but we are not going to do that. 00:20.150 --> 00:26.560 Instead of that, this feature engineering stuff in this, as you learn or I would say is cyclical and 00:26.560 --> 00:28.880 lively, we'll pick for us. 00:29.270 --> 00:32.290 And for that, we are going to use this D.F. idea, vectorized it. 00:32.450 --> 00:33.710 And that is one pipeline. 00:34.220 --> 00:39.770 And then we are going to apply this support vector machine, machine learning algorithm. 00:39.800 --> 00:46.650 So let me execute this all necessary inputs and I will tell you what is this B.S. idea rectories. 00:47.390 --> 00:54.860 So this is like another feature encoding technique where we are going to convert all our domes. 00:54.980 --> 00:58.800 Or I would say the original documents are poisonously reviews. 00:58.870 --> 01:03.230 What is it that goes into kind of some sort of numbers? 01:03.650 --> 01:07.760 So if you'll just search for the ideas. 01:09.260 --> 01:11.690 Yes, that's a good explanation, I guess. 01:11.930 --> 01:12.410 Let me know. 01:15.260 --> 01:22.550 So based on this particular formula, they are calculating how much important individual vote in individual 01:23.120 --> 01:23.660 document. 01:24.260 --> 01:26.150 So for Tom De. 01:27.390 --> 01:35.410 A document how much its importance is and has quote of this D.F. idea will decide those things. 01:35.800 --> 01:40.960 So Tom Frequency's kind of how many times there will be cause in a document. 01:41.720 --> 01:44.400 And this was document frequency Liko. 01:45.140 --> 01:45.980 How many times? 01:45.980 --> 01:50.120 Some particular time, of course, across all those documents. 01:50.420 --> 01:57.570 And if you just multiplied, it will be kind of very much normalized stuff that many will some frequently 01:57.570 --> 02:00.860 are occurring were always occurs in an English dictionary. 02:01.100 --> 02:02.800 So it will do lower rates today. 02:03.260 --> 02:07.410 And those which are the very that word will view higher rates today. 02:07.760 --> 02:10.830 And that's why this idea component had it. 02:11.210 --> 02:13.070 So we delivered this particular formula. 02:13.140 --> 02:18.380 We're converting all those of reviews into kind of numbers. 02:18.680 --> 02:24.880 And for that, we are going to use this idea, recognize that a class from this has a loan library. 02:25.370 --> 02:32.490 Now, while converting those kneeing, we are going to pass this function, whatever the data are. 02:32.630 --> 02:39.780 Tax data, planning function we have created and we will tell this EAF ideas, vectorized class that 02:40.270 --> 02:46.370 like converting before that when you do all those kinds of tokenization tokenized. 02:47.630 --> 02:51.300 You pass through everything with this function. 02:51.870 --> 02:53.640 So let me execute it. 02:55.570 --> 02:58.220 And it will create B.F., IDF rectories it class. 02:58.580 --> 03:02.390 So this D.F. IDF vectorized class will do two things. 03:02.780 --> 03:05.540 First, it will be applied on this function. 03:06.170 --> 03:11.030 So tokenization will then according to whatever staff we help define here. 03:11.510 --> 03:18.470 And after that, those D.F. IDF formula will be applied to calculate actual DFAT of school for each 03:18.470 --> 03:21.320 individual done in different documents. 03:21.860 --> 03:23.180 And let me have any. 03:24.370 --> 03:28.330 All right, let's create a support vector machine, classify it. 03:29.910 --> 03:33.820 All right, so next thing is we have a data available with us. 03:34.240 --> 03:36.010 We have a machine learning algorithm. 03:36.010 --> 03:36.800 I will be test. 03:37.300 --> 03:39.760 Next thing is we need to train this model. 03:40.360 --> 03:42.520 And for that, we have to split our data. 03:43.060 --> 03:44.410 We are going to split the data. 03:44.440 --> 03:49.720 We Delp of this brain and this patient underscore split and 20 percentage of data. 03:49.750 --> 03:52.590 We are going to keep it inside that texting pockets. 03:53.020 --> 03:59.010 So if you just observe the shape of this green dataset and it basically does it find it in that causes 03:59.020 --> 04:03.940 a part of pasting buckets and two one nine eight is a part of draining buckets. 04:04.510 --> 04:06.700 So just observe the first few hours of training. 04:07.670 --> 04:08.110 All right. 04:08.590 --> 04:10.990 Next is we are going to create one pipeline object. 04:11.530 --> 04:14.170 And first, it will be possible. 04:14.980 --> 04:19.600 The first object of this D.F. idea and then the classify. 04:20.110 --> 04:21.980 So that will create one pipeline object. 04:22.690 --> 04:26.550 And let just fit our data in the sense that training is going to start. 04:26.710 --> 04:26.950 Now. 04:31.230 --> 04:37.710 So it's taking a good amount of time for the training purpose, so I'm just fast forwarding my video 04:37.910 --> 04:39.290 Pilates training. 04:39.780 --> 04:40.460 We call it. 04:43.620 --> 04:49.950 All right, so now you can see that training is fitness and it has written this pipeline object where 04:50.340 --> 04:56.940 the steps are like a two step one is terrified of step and and that one is a support vector machine 04:56.940 --> 04:57.200 step. 04:57.690 --> 04:58.960 Now, once the training is over. 04:59.040 --> 05:01.380 Next thing is to predict our pace. 05:01.680 --> 05:02.030 Listen. 05:02.640 --> 05:10.980 So for prediction of testing result, we are going to import this classification criteria like Accuracy 05:10.980 --> 05:14.220 Sport Classification Report and a confusing matrix. 05:15.060 --> 05:18.810 And let's apply prediction on our testing dataset. 05:18.870 --> 05:21.690 So it will return us all those prediction. 05:22.140 --> 05:25.110 Let's not confuse and metrics. 05:25.390 --> 05:29.090 So now here we are dealing with a binary classification problem. 05:29.100 --> 05:31.920 So it has come up with a two across two metrics. 05:32.340 --> 05:36.650 So in 201 cases had 221 cases. 05:36.690 --> 05:45.630 That is a national element where everything predicted was the right prediction that as v Mr. Prediction 05:45.630 --> 05:48.140 in case of people plus 70 recalls. 05:48.510 --> 05:53.670 If you need detailed explanation of this classification report, you can use this classification report 05:53.670 --> 05:57.690 and you can pass on all your grant route lewyn and predictions. 05:58.770 --> 06:03.450 So in this case, we are almost getting seventy seven percentage of accuracy. 06:04.560 --> 06:11.580 If you want to get this accuracy score, something like this total will be your total number of testing 06:11.700 --> 06:12.090 dataset. 06:12.180 --> 06:13.830 So we have a testing that does it. 06:13.860 --> 06:15.570 How much how many testing datasets? 06:16.200 --> 06:17.250 Find it interesting. 06:17.670 --> 06:24.690 So out of finding 50, there will be a 50 plus 78 already gone wrong. 06:25.200 --> 06:30.050 And our accuracy's code is seventy six point seventy two percentage. 06:30.480 --> 06:33.780 So almost down twenty three point five. 06:33.920 --> 06:36.110 What I would say twenty four percentage of his. 06:36.120 --> 06:37.230 We got it wrong. 06:37.950 --> 06:40.920 Now, let's break with some sample reviews. 06:41.450 --> 06:41.950 Let's see. 06:42.210 --> 06:43.350 Reviews will be Voll. 06:43.440 --> 06:43.700 Hi. 06:43.700 --> 06:46.710 I'm learning natural language processing a fun fashion. 06:47.160 --> 06:54.540 So if you just applied to all the classifier algorithm or pipeline algorithm, what it will do, it 06:54.540 --> 06:58.120 will try to convert with Delp of this D.F. idea. 06:58.320 --> 07:07.530 Vectorized it would all text processing function we created applied on this linear SBC or I would say 07:07.860 --> 07:12.450 support vector machine class and create a prediction for us. 07:12.900 --> 07:16.820 So the prediction is one that means it's the positive review. 07:17.670 --> 07:18.800 Let's predicted four. 07:19.020 --> 07:20.870 It's hard to learn new things. 07:21.020 --> 07:24.630 That means it's a kind of negative feelings we're giving. 07:24.690 --> 07:25.850 Let's predicate. 07:26.640 --> 07:27.780 So output is zero. 07:27.900 --> 07:29.880 That means it's a negative review. 07:30.530 --> 07:30.870 All right. 07:30.900 --> 07:33.240 So that's all about this combined. 07:33.450 --> 07:34.550 We use this. 07:34.560 --> 07:42.240 I am Libby, Yale and Amazon Review and applied a little differently, this tax classification or I 07:42.240 --> 07:46.340 would say sentiment analysis problem, positive or negative. 07:46.830 --> 07:54.560 And mainly we use a different thing, like a different spacy function and how we can apply those all 07:54.790 --> 08:01.300 specification as a pipeline process while using the different functions of this pastilla label. 08:02.070 --> 08:02.460 All right. 08:02.580 --> 08:03.930 That's all about this project. 08:03.960 --> 08:05.060 See you in the next video.