1 00:00:00,420 --> 00:00:05,250 So now you have created a where we base our model. 2 00:00:05,640 --> 00:00:11,220 We have predicted why best and wide-screen values using that model. 3 00:00:11,970 --> 00:00:18,150 Now it's time to evaluate the performance of our model using these projected values. 4 00:00:19,510 --> 00:00:25,080 There are two important metrics which we use to evaluate our model performance. 5 00:00:25,710 --> 00:00:28,230 The first one is mean squared error. 6 00:00:29,100 --> 00:00:36,060 This is the mean of is square off or a deviation of predicted values from the actual values. 7 00:00:37,700 --> 00:00:41,180 The second parameter is R-squared value. 8 00:00:41,760 --> 00:00:46,740 This is the goodness of fit for the modern artist square. 9 00:00:46,940 --> 00:00:53,030 Lies between zero and one one sense for perfect fit. 10 00:00:53,420 --> 00:01:00,560 That means we are able to justify all the variations in our Y variable using our model. 11 00:01:01,790 --> 00:01:03,220 And zero means no fake. 12 00:01:03,320 --> 00:01:12,050 We are not able to identify any deviations and by using our model generally are the square lies between 13 00:01:12,050 --> 00:01:14,600 zero point four two zero point eight. 14 00:01:14,850 --> 00:01:20,180 For good models and zero point eight annable for excellent models. 15 00:01:21,470 --> 00:01:24,380 There is no such range for mean squared error. 16 00:01:25,190 --> 00:01:27,380 This is an absolute number. 17 00:01:27,740 --> 00:01:31,790 You cannot compare it across different projects. 18 00:01:32,240 --> 00:01:40,940 This depends on your current dataset and you can use MASC to evaluate performance of different models 19 00:01:41,000 --> 00:01:42,070 on the same dataset. 20 00:01:42,710 --> 00:01:45,200 Now it is very easy to calculate. 21 00:01:45,230 --> 00:01:47,690 This is escort's using a skillern. 22 00:01:48,490 --> 00:01:54,290 First you have to import mean a squared error and are to underscore the score from Escalon. 23 00:01:54,320 --> 00:01:57,710 Metrics will just run this. 24 00:01:57,710 --> 00:01:58,110 Come on. 25 00:02:00,320 --> 00:02:05,990 Now to calculate the MASC, you just separate mean under score squared under score error. 26 00:02:06,770 --> 00:02:13,240 Then you have to give your actual Y values and you are credited y values. 27 00:02:13,790 --> 00:02:22,040 We have already predicted our Y values in our last lecture, so we will use just y under score tests 28 00:02:22,070 --> 00:02:23,330 and by underscore tests. 29 00:02:23,330 --> 00:02:25,520 Underscore pride as our arguments. 30 00:02:26,710 --> 00:02:33,830 Zukin can see the mean squared error on our test dataset is one hundred and eleven million. 31 00:02:35,350 --> 00:02:42,190 We can only use mini skirt edit to calculate performance of different models on our similar task. 32 00:02:42,340 --> 00:02:45,520 So right now this has no meaning for us. 33 00:02:46,060 --> 00:02:50,530 Let's move on to artist square value for our treene dataset. 34 00:02:51,010 --> 00:02:52,730 We have the predicted values. 35 00:02:52,740 --> 00:02:58,560 Are Wyandotte score green on that score, pride and the actual user in the score screen. 36 00:02:59,770 --> 00:03:03,850 So our artist square value is point A three. 37 00:03:04,300 --> 00:03:08,020 This means our model is performing great. 38 00:03:09,430 --> 00:03:10,960 Now let's calculate. 39 00:03:11,110 --> 00:03:13,720 Artists could well lose on our test data. 40 00:03:14,260 --> 00:03:21,550 So remember, we have created our model on this train data and we cannot use this train data to evaluate 41 00:03:21,550 --> 00:03:22,300 the performance. 42 00:03:22,330 --> 00:03:28,420 We have to give the test data that we have kept aside while creating our model. 43 00:03:29,620 --> 00:03:35,220 The artist square value for our test dataset is zero point six five. 44 00:03:36,690 --> 00:03:42,360 You should always look at the best artists good values who really would do more than performance, since 45 00:03:42,540 --> 00:03:45,610 you have not used this as their tough during model training. 46 00:03:46,680 --> 00:03:52,600 And we are getting artists to value more for our train, that asset as cumber bellard as there does 47 00:03:52,600 --> 00:03:58,290 it, because it is obvious we have trained our model on our train does it and it will always perform 48 00:03:58,290 --> 00:04:01,380 better on the dataset on which it has trained. 49 00:04:01,980 --> 00:04:09,780 So it is obvious to get artists to value great for our crane dataset then our estimates say.