WEBVTT 00:00.080 --> 00:01.080 Welcome back. 00:01.120 --> 00:04.000 We get out of range. 00:04.320 --> 00:05.400 Impossible. 00:05.400 --> 00:06.680 Predicted mileage. 00:06.840 --> 00:08.040 And this is wrong. 00:08.320 --> 00:15.280 This is due to wrong feature order, wrong scale constants or data entry bug. 00:15.520 --> 00:19.440 Let me check the feature order and constants. 00:19.600 --> 00:22.720 So let me go back to the car features. 00:22.840 --> 00:27.160 Split the screen and here we have the order. 00:27.160 --> 00:30.440 Let me check the order with scaling features. 00:30.640 --> 00:32.920 Cylinder displacement. 00:33.080 --> 00:35.320 Horsepower weight. 00:35.360 --> 00:38.280 They are the same acceleration. 00:38.280 --> 00:41.000 Then model year, then origin one, two and three. 00:41.280 --> 00:44.080 Acceleration model year, origin one, two and three. 00:44.440 --> 00:48.960 Those for the standard deviation and the mean. 00:48.960 --> 00:50.560 So here cylinders. 00:50.560 --> 00:51.120 Displacement. 00:51.160 --> 00:51.760 Horsepower. 00:51.760 --> 00:52.080 Weight. 00:52.080 --> 00:55.400 Acceleration model year one two and three. 00:55.440 --> 00:57.560 And those are the same. 00:57.600 --> 00:58.080 Okay. 00:58.480 --> 01:02.560 So the order is not the problem. 01:02.560 --> 01:05.040 Let's see the input data. 01:05.240 --> 01:10.340 So let's scroll down to here for the statistical summary. 01:10.340 --> 01:12.900 And let me put it like this. 01:13.140 --> 01:19.140 The statistical mean for cylinders 5.47. 01:19.260 --> 01:23.260 Displacement one nine for horsepower one zero. 01:23.300 --> 01:27.860 For weight 2977. 01:28.060 --> 01:29.100 And acceleration. 01:29.700 --> 01:32.980 The acceleration is the error. 01:33.180 --> 01:36.940 So here we get the error of 15.5. 01:36.980 --> 01:40.300 For the model year is 75. 01:40.500 --> 01:43.380 The origin is 1.57. 01:43.380 --> 01:45.820 Let me check the feature standard. 01:45.820 --> 01:54.820 So you see guys that any error your data will be missed and the model will be generating random and 01:54.820 --> 01:56.460 wrong predicted data. 01:56.500 --> 01:57.900 The standard for the first 01:57.900 --> 02:08.580 1.710438849, 2.7, 3.6 and 0.8. 02:08.860 --> 02:12.420 So this is good for our application. 02:12.460 --> 02:15.180 Now let's run our application again. 02:15.180 --> 02:18.730 And by the way you can debug the data. 02:18.730 --> 02:22.490 So scroll down to here in the Predict mileage. 02:22.490 --> 02:26.610 Let me debug the data the scaled data. 02:26.610 --> 02:33.650 So debugging the scaled features let me log scaling cylinders features. 02:33.650 --> 02:36.890 With scaling you can put like this. 02:37.050 --> 02:40.330 Or you can print all the features. 02:40.570 --> 02:51.970 So print LN scaled features features with scaling dot for each indexed go and set the index and the 02:51.970 --> 02:52.730 value. 02:52.770 --> 02:55.010 Print the feature and the index. 02:55.330 --> 02:57.170 Let's run our application. 02:57.170 --> 02:57.930 Here we go. 02:57.970 --> 02:59.930 This is our application open. 02:59.930 --> 03:01.330 Look at clear. 03:01.330 --> 03:02.210 Look at. 03:02.370 --> 03:08.490 And here we have the data entered predict MPG. 03:08.690 --> 03:13.730 And here we get a new value and it's acceptable. 03:13.730 --> 03:16.810 So now it's 34 mpg. 03:16.970 --> 03:20.330 Let's select European predict mileage. 03:20.330 --> 03:21.970 It's 25. 03:22.010 --> 03:29.190 Let's select Japanese and predict the MPEG and it's working fine. 03:29.190 --> 03:30.950 Let's go back to American. 03:31.230 --> 03:33.190 Let's predict the features. 03:33.190 --> 03:39.110 And here we get the scaled features zero -0.83. 03:39.310 --> 03:44.030 Let's compare them with our Google Colab. 03:44.230 --> 03:55.110 So here if we take the same sample new car four cylinders 150 100 horsepower, 3015 acceleration and 03:55.110 --> 03:58.430 model year 76 origin American run. 03:58.430 --> 04:02.230 It will give me 20.54. 04:02.470 --> 04:05.790 And this is a big difference between our application. 04:05.830 --> 04:09.030 Those are similar to what we've done. 04:09.230 --> 04:12.630 Select American predict and it's 34. 04:12.630 --> 04:14.270 And this is a big problem. 04:14.310 --> 04:23.110 Maybe there is here in the standard deviations and the mean we need to modify those origins because 04:23.270 --> 04:26.070 the features are the same guys. 04:26.230 --> 04:29.390 So here we have a lot of features. 04:29.390 --> 04:33.070 The first six features are same as before. 04:33.070 --> 04:35.900 We need to configure the origin. 04:36.140 --> 04:40.220 So let me test again with this code. 04:40.260 --> 04:48.620 So here under this go and start anew and paste X train origin one, origin two and origin three. 04:48.660 --> 04:51.540 Get the mean and the standard deviation of them. 04:51.700 --> 04:53.180 And here we go. 04:53.220 --> 04:57.740 We get the standard deviation and the mean of them. 04:57.780 --> 05:00.780 So those are for the standard deviation. 05:00.780 --> 05:02.580 Let me correct them. 05:02.580 --> 05:05.500 So here the standard deviation. 05:05.500 --> 05:07.020 Replace it by this. 05:07.220 --> 05:10.780 Then this and this. 05:10.940 --> 05:14.060 Also we need to correct the mean. 05:14.100 --> 05:27.940 Now our mission to get the feature means here if we see guys that the problem is origin are of integer. 05:28.060 --> 05:33.060 And we splitted the origin to three columns. 05:33.060 --> 05:39.060 So we need to get the data from the three columns, not from the one column.