WEBVTT 00:01.560 --> 00:08.000 Hylan, I hope you have learned a lot from these supervised learning and unsupervised listening sessions 00:09.030 --> 00:13.380 the Supervised Learning Project continues to produce. 00:14.280 --> 00:20.150 The first project is a regression project and the second one is a classification project. 00:21.090 --> 00:31.530 This particular review will give you an outline of how you should work on the project in case of supervised 00:31.530 --> 00:32.010 learning. 00:32.190 --> 00:35.480 As we already know, we have two types of variables. 00:36.030 --> 00:43.020 The first one is the value which we want to predict, and the second one are the different values using 00:43.020 --> 00:49.140 which we will learn certain ideas about the values and we will predict those. 00:49.860 --> 00:51.810 So what is the data which we have here? 00:56.800 --> 01:01.660 Here we have this entire dataset, which is about the whole state. 01:03.380 --> 01:12.110 The target here is the price column, and here you can see we have around twenty one thousand rows of 01:12.110 --> 01:16.160 data in this twenty one thousand rows of data. 01:16.170 --> 01:21.520 We have various details about the prices, which is the value which you need to predict. 01:22.280 --> 01:24.810 As you can see, this is a continuous value. 01:24.830 --> 01:27.590 So this should be a regression from. 01:28.940 --> 01:33.320 Now, there are a lot of columns which are present in this particular dataset. 01:34.700 --> 01:43.620 The values being I.B., the bedrooms, number of bathrooms, the adult, the house, the waterfront 01:43.640 --> 01:48.060 condition, if it is a waterfront, what is the condition level of the house? 01:48.290 --> 01:50.300 What is the grade of the house? 01:50.570 --> 01:55.490 And when it was built, when it was renovated, the zip code and so on. 01:56.450 --> 02:00.020 Using all of these details, you need to create. 02:01.140 --> 02:09.030 Features and then use them, as you can see, that there are no categorical columns pressing, you need 02:09.030 --> 02:16.500 not convert anything into the tables, but you can consider these your renovation. 02:18.080 --> 02:25.100 Zip code, as I've got to go to columns, I'm convert to renovation and find some relationship between 02:25.730 --> 02:27.220 the renovation. 02:27.530 --> 02:32.990 Similarly, you can find out some details about this date, what this date signifies. 02:33.320 --> 02:35.510 So this is what you can do. 02:39.640 --> 02:46.810 And similarly, here are some ideas now based on your learning, you would better know what you need 02:46.810 --> 02:47.930 to do with this idea. 02:49.150 --> 02:56.410 So now once you have this entire dataset, you will follow the process, which has been clearly defined 02:56.410 --> 02:57.930 and described to you. 02:59.780 --> 03:06.650 So you will find out different outliers and missing data, and in case there are certain outliers or 03:06.650 --> 03:11.340 missing data presented, then you will try to impute the values to it. 03:11.540 --> 03:16.260 Now, what values you should be imputing, you would have learned during the preparation. 03:17.030 --> 03:23.060 Similarly, you can create new columns and using these techniques, which we have discussed. 03:23.300 --> 03:27.260 You can remove the columns and do feature selection. 03:28.800 --> 03:35.040 Then you will be applying different algorithms and comparing the performances. 03:36.150 --> 03:43.000 We have very rewarded performances, not that you don't need to improve the model a lot. 03:43.950 --> 03:49.080 The model improvement is a dome, which you need to know that even. 03:50.040 --> 03:58.450 If a model you see at one point of time, is it the person using a linear model? 03:58.620 --> 04:00.750 I'm using a random for this. 04:00.750 --> 04:09.360 Let us see in case the eye could see changes to eighty eight point one, then it is not recommended 04:09.360 --> 04:17.520 to use a random forest because we are not gaining a lot in terms of accuracy, but we are increasing 04:17.520 --> 04:18.840 the complexity of the. 04:20.110 --> 04:27.520 So always consider how much you're increasing the complexity of the model in terms of how much you are 04:27.520 --> 04:29.050 gaining in terms of accuracy. 04:29.800 --> 04:32.050 Same thing goes for stacking. 04:32.560 --> 04:40.450 If you are random forest or you most orny, this model is giving you an IQ of see eighty eight point 04:40.450 --> 04:45.630 six percent and a stacking model gives you be eight point seven percent. 04:45.970 --> 04:53.500 Then go ahead and use the old model, the range of what is a used or whatever model you have used instead 04:53.500 --> 04:59.350 of going for the stacking model, because again, the stacking model will be a highly complex model, 04:59.530 --> 05:01.270 which you don't want to use. 05:02.580 --> 05:10.410 OK, so these are a few things which you need to keep in mind and also remember, there is no great 05:10.410 --> 05:11.380 or good model. 05:11.820 --> 05:15.700 A model should be having a good accuracy. 05:15.700 --> 05:21.090 And your target here is to learn how to implement the different algorithms. 05:21.720 --> 05:25.870 So do not skip to any specifical automatically. 05:26.100 --> 05:33.840 What I would recommend to you is to figure out the dataset and implement all the algorithms which you 05:33.840 --> 05:37.470 have learned and improve each and every algorithm. 05:38.770 --> 05:45.790 By this, you will learn how to fine tune different models and you will also learn how you can compare 05:45.790 --> 05:46.870 different models. 05:50.440 --> 05:59.440 During this entire course, I have used a lot of data sets, and this gives you an additional opportunity 05:59.740 --> 06:03.300 to use those data sets for you for the learning. 06:03.880 --> 06:09.510 So as a project, you can implement several algorithms on this particular dataset. 06:09.700 --> 06:16.060 And after you are done with the project as a future learning, you can pick out any datasets which I 06:16.060 --> 06:24.040 have shared and then start exploring them, stop walking on them and implementing different algorithms 06:24.040 --> 06:25.660 on top of those datasets. 06:26.520 --> 06:33.610 That is the reason why I have so many data sets so that at the end of the course you will have enough 06:33.610 --> 06:35.020 data to practice on. 06:35.900 --> 06:45.860 So I hope you will work hard on this project and improve the accuracy of the models and create great 06:45.860 --> 06:49.970 models and have a bright future using this morning. 06:50.660 --> 06:51.230 Thank you.