WEBVTT 00:00.720 --> 00:01.630 All right, everyone. 00:01.750 --> 00:09.550 So next step is applying this random forest machine learning algorithm on the top of a message column 00:09.580 --> 00:14.150 or I will say our training data, say XStream and likely. 00:15.020 --> 00:15.360 All right. 00:15.760 --> 00:23.410 So here we are dealing with just tax data only, and we just cannot prove all those actual data directly 00:23.410 --> 00:25.450 on a machine learning algorithm. 00:25.450 --> 00:31.630 V, how to convert some sort of encoding which will convert all those textual data into some sort of 00:32.020 --> 00:32.650 numbers. 00:33.010 --> 00:39.310 Now, to do that, there are a number of Mattos are available like a bag of four Smardon what I would 00:39.310 --> 00:46.830 say DFI RDF model and some of the deep learning based advance models like a vote to work for a global 00:46.830 --> 00:47.320 actors. 00:47.740 --> 00:55.630 So all those techniques eventually try to convert all those textual data into kinds of number which 00:55.630 --> 01:00.400 will try to preserve all those semantic relationship exist between the data. 01:01.230 --> 01:06.400 The deep learning mis model of Bisley works much better with accuracy. 01:06.790 --> 01:14.480 But we'll go with the one hand coding technique which will convert all your text data into kind of D.F. 01:14.530 --> 01:15.370 idea factor. 01:15.410 --> 01:17.830 That is nothing but a down frequency. 01:18.190 --> 01:20.370 Hannah, inverse documents frequency. 01:20.410 --> 01:28.150 So multiplication of most of this will give you some number and gives you some score the presence of 01:28.150 --> 01:30.430 some particular word in a document. 01:30.820 --> 01:35.410 So you can consider every single record here as a kind of document. 01:36.070 --> 01:37.060 So what do you do? 01:37.390 --> 01:41.000 Let me first import those PEF ideas. 01:42.030 --> 01:42.970 So from. 01:46.180 --> 01:46.900 SBA loan. 01:49.500 --> 01:51.180 Not feature extraction. 01:52.110 --> 01:53.100 Let's import. 01:58.170 --> 02:00.260 So feature sticks and inside Taito. 02:00.360 --> 02:08.820 So we to go like a tax model and it will be at D.F. Ideas, rectories it. 02:09.710 --> 02:10.020 All right. 02:10.050 --> 02:11.020 So that is the one input. 02:11.610 --> 02:14.620 Another one is we are going to apply this addendum forest already. 02:14.850 --> 02:18.420 So that is a part of our Skillern and Simbel Martin. 02:21.710 --> 02:22.980 And let me import 02:28.170 --> 02:29.000 random forest. 02:29.100 --> 02:29.660 Classified. 02:30.210 --> 02:32.280 So first will apply this DFI leaf. 02:32.460 --> 02:35.180 And then random forest classified. 02:35.250 --> 02:37.830 So that is a kind of very much pipelined process. 02:38.100 --> 02:43.770 So what we'll do applying this, both things individually, we can create one pipeline object. 02:44.190 --> 02:47.970 So that is also part of this Aspelin pipeline. 02:49.560 --> 02:50.820 We help import this. 02:52.930 --> 02:53.410 Pipeline. 02:54.160 --> 03:01.220 All right, now, instead of creating this two object, we will create a pipeline object. 03:01.600 --> 03:07.480 And instead the pipeline object, we will just pass most of this object, PEF, IDF, recognize it and 03:07.510 --> 03:08.890 then the forest classify it. 03:09.430 --> 03:11.620 So let me make pipeline object. 03:15.290 --> 03:18.560 And here we will pass it as a list. 03:19.190 --> 03:27.020 This two objects, DFI, Dave, vectorized object and a random forest classified object so we can pass 03:27.020 --> 03:29.390 both of them as like couples. 03:31.670 --> 03:39.500 So it will be a let's give the name like the IDF and corresponding its object. 03:41.090 --> 03:41.810 Next is. 03:46.120 --> 03:52.630 Obviously, a classic fire that is nothing but a random forest classified object. 03:53.260 --> 03:55.130 Now this number is classified. 03:55.130 --> 03:57.660 It contains a lot of hyper parameter. 03:58.420 --> 04:01.300 So you can see you can see here. 04:02.890 --> 04:08.760 There are a lot of Hyperdynamics to exist, so if you want to improve this model, you can do this hyper 04:08.760 --> 04:09.230 parameter. 04:09.700 --> 04:14.830 So let's just make it some hyper parameters, like let's say and in this estimate, stimulus. 04:17.330 --> 04:18.760 So lean forward to Ray Lewis and Ray. 04:19.230 --> 04:21.050 Let just make it even handed on. 04:22.010 --> 04:23.690 So that we be an estimate. 04:24.260 --> 04:26.670 Let's just make it four timing, let's say 10. 04:28.520 --> 04:33.710 And let me assign it to classify it as an object. 04:35.680 --> 04:42.790 And let just treat our input training data, because one object got created, you need to train. 04:42.990 --> 04:46.570 And for training, there is a uniform sets of EPA that are available. 04:46.700 --> 04:49.130 Now, this has scale and library. 04:49.450 --> 04:51.370 So that will be a fit function. 04:51.940 --> 04:53.220 So here we are going to pass. 04:53.380 --> 04:56.110 Explain Tanah Lightering. 04:58.270 --> 05:01.070 All right, let me execute it. 05:01.520 --> 05:05.330 And that training process will start and you can see for creating. 05:06.360 --> 05:10.350 Ten different estimates are I would say that is a 10 different. 05:10.780 --> 05:12.120 This is in 3D will create. 05:12.480 --> 05:18.060 It has immediately given us the results so we can even go with a let's say and date also. 05:18.560 --> 05:23.910 And let me defined as classified once again, let just fit it once more. 05:25.460 --> 05:27.520 All right, so immediately we've got the result. 05:27.950 --> 05:30.650 That means our training is finished. 05:31.130 --> 05:35.120 Let's try to evaluate what model that how good our model is. 05:35.540 --> 05:36.680 So model, it's created. 05:37.100 --> 05:41.390 Our training process is what we need to use this classifier object. 05:42.710 --> 05:44.960 And we are going to use this Braddick matter. 05:46.200 --> 05:50.370 So prediction, we are going to apply on packs and escort testing. 05:50.820 --> 05:57.540 That means the input they stay w will pass it and it will give us some kind of prediction that our model 05:57.540 --> 05:59.510 has predicted this values. 06:00.260 --> 06:01.680 Let me executive. 06:03.070 --> 06:09.740 And if you're just this by and this court test and why in the school, right. 06:10.660 --> 06:13.060 Side by side comparison, you would be able to see. 06:13.840 --> 06:14.260 All right. 06:14.290 --> 06:20.070 So first, it has displayed this vibrate and then it has displayed this light prediction. 06:20.110 --> 06:23.390 So first Hispan, then we have a spam, then spam. 06:24.070 --> 06:24.390 All right. 06:24.430 --> 06:27.610 So analytically, how we can compare both of this. 06:27.870 --> 06:28.170 Is a. 06:29.650 --> 06:37.000 And to do those thing, there is a one definite matricide available to get the accuracy behind your 06:37.390 --> 06:39.040 classification problem. 06:39.700 --> 06:42.420 So one is like a full detailed classification record. 06:42.440 --> 06:47.610 So it will be hard if you just want to know about the accuracy score or if you want to know about the 06:48.010 --> 06:50.100 one specific measurement criteria later. 06:50.890 --> 06:51.970 Confusing matrix. 06:53.920 --> 06:55.930 So let's try to find all of them. 06:55.960 --> 06:59.260 So from Haski line. 07:03.270 --> 07:04.420 Not metrics. 07:05.130 --> 07:06.240 We are going to import. 07:09.520 --> 07:14.080 Classification Report 101 is accuracy scored. 07:16.920 --> 07:21.910 And one more is confusing matrix right now. 07:22.590 --> 07:30.040 Whatever it predicted data in our test data, we will be supplying and we'll try to find all those main 07:30.070 --> 07:31.890 criteria, how good our model is. 07:32.310 --> 07:35.390 So one is we can use like a accuracy score. 07:36.460 --> 07:39.350 So first, what argument you need to pass like a virus. 07:39.400 --> 07:43.270 That means your ground level and next is that you are protected. 07:43.870 --> 07:49.390 So that will be a white test that is able to truly win and wipe predicted. 07:49.480 --> 07:50.970 Will be it predictive, is it? 07:51.850 --> 07:53.050 And if you execute it. 07:54.200 --> 07:58.420 You'll be able to see almost ninety four point sixty five percent of your case. 07:58.780 --> 07:59.830 We got it right. 07:59.860 --> 08:02.460 That means our model is predicted. 08:02.950 --> 08:04.030 Very good reason. 08:04.120 --> 08:10.150 If you just multiply it with our total number of records, that will be around. 08:11.110 --> 08:12.410 How many testing that code? 08:12.460 --> 08:14.640 That is it, 449 testing records. 08:15.190 --> 08:17.500 So if you multiply this number by 08:20.170 --> 08:25.900 forty nine, we will get 425, sort of 449. 08:26.170 --> 08:27.940 We know for 425 samples. 08:27.970 --> 08:28.870 We got it right. 08:29.230 --> 08:31.240 That means our model is quite accurate. 08:31.270 --> 08:33.400 We got enough good accuracy. 08:34.090 --> 08:38.710 We can even calculate this confusing metrics also so, so confusing metrics. 08:38.740 --> 08:46.600 Give us a little bit more detail stuff related to our classification problem and how accurate and a 08:46.600 --> 08:47.800 good our model is. 08:49.420 --> 08:51.010 And let just pass. 08:51.310 --> 08:52.240 Same like earlier. 08:52.480 --> 08:55.480 Why test and why. 08:56.940 --> 08:57.090 Right. 08:58.210 --> 08:58.630 All right. 08:59.170 --> 09:06.080 So now as we are dealing here with a binary classification problem, it will return us to cross two 09:06.250 --> 09:06.820 metrics. 09:07.030 --> 09:12.130 So in know diagonal elements, you can see it has radically correctly. 09:12.520 --> 09:16.000 So in a first case, let's consider it like a. 09:17.850 --> 09:26.340 Ham, then we always pan ham and then we spend so in off diagonal elements that that is a 24 in the 09:26.340 --> 09:26.790 numbers. 09:27.000 --> 09:35.880 That's where we got it wrong, that our data was ham and our system as predicted, like a spam, whereas 09:36.170 --> 09:39.540 sometimes we eat has predicted like a spam. 09:39.780 --> 09:41.280 But our data was a ham. 09:41.710 --> 09:42.730 So otherwise I was. 09:43.410 --> 09:47.650 But you know, on diagonal elements 226 and one ninety nine. 09:48.210 --> 09:51.950 This is the case where we completely got it right. 09:52.920 --> 09:57.510 Now, if you're dealing with, let's say, multiclass classification problem, let's say fateless classification 09:57.510 --> 09:57.900 problem. 09:58.170 --> 10:04.800 So in that case, you would have got this confusing matrix as a fake CROSSFIRE matrix because on a horizontally 10:04.920 --> 10:09.950 Aldo's through label will be given a vertical, it will be given a radically limit. 10:10.790 --> 10:15.090 Now, if you want to get more information about this confusing matrix, you can just Google around it 10:15.440 --> 10:17.040 and you will get much more idea. 10:17.890 --> 10:18.220 All right. 10:18.270 --> 10:19.980 So one more is classification. 10:22.170 --> 10:27.880 So it seemed like earlier we can pass here whiteish in a way, Fred. 10:29.440 --> 10:30.630 And let me executed. 10:31.840 --> 10:38.460 All right, so it doesn't display, you know, beautifully, Meinhardt, so we can just wrap it around 10:38.460 --> 10:39.400 a print function. 10:40.800 --> 10:41.290 All right. 10:41.380 --> 10:45.440 So somewhat more detailed explanation about our model. 10:45.850 --> 10:48.310 They are given like a precision recall. 10:48.610 --> 10:49.260 Have fun, Scott. 10:49.450 --> 10:51.340 And a support for what category? 10:51.340 --> 10:52.620 Ham and spam. 10:53.920 --> 10:54.470 All right. 10:54.520 --> 10:57.280 So that is how accurate and good our model is. 10:57.400 --> 10:58.330 A random forest. 10:58.960 --> 11:01.630 Now, can you improve it or can you apply some other model? 11:01.660 --> 11:06.640 So in the next video, we will see applying another model support regular machine. 11:06.910 --> 11:08.190 So see you next video.