WEBVTT 00:00.080 --> 00:00.960 Welcome back. 00:00.960 --> 00:04.920 Now let's run our application and see if there is any error. 00:04.960 --> 00:08.960 Here I select the Xiaomi my physical device. 00:08.960 --> 00:12.520 Also you can select the emulators. 00:12.520 --> 00:17.840 But in order to prevent any errors I need to test on my real device. 00:17.840 --> 00:22.600 So here this is our application car mileage predictor. 00:22.840 --> 00:33.640 Set the cylinders for displacement 150 horsepower, weight acceleration and the model year. 00:33.840 --> 00:46.200 Let me test if I entered invalid model year for example 1999 okay so 90 select American predict mileage. 00:46.320 --> 00:53.560 And here we have this error model year must be between 70 and 82. 00:53.720 --> 00:59.200 So here our validation of errors is working fine. 00:59.200 --> 01:01.720 Let me make it as 76. 01:01.960 --> 01:03.120 This is good. 01:03.120 --> 01:05.800 Select American predict mpg. 01:06.080 --> 01:07.760 And here we go. 01:07.880 --> 01:12.720 Predicted mileage is 8184. 01:12.760 --> 01:24.120 This is not acceptable because we need to get, uh, data between and range between, uh, like 12 and 01:24.160 --> 01:26.040 25 or 30. 01:26.080 --> 01:28.960 So this is not acceptable. 01:28.960 --> 01:29.440 Why? 01:29.880 --> 01:35.480 Because we used data that are not scaled. 01:35.480 --> 01:43.680 So here I want from you to understand the importance of the scaled features. 01:43.720 --> 01:47.160 Again, guys, this is the most important thing. 01:47.200 --> 01:52.120 You should pay attention when training machine learning and models. 01:52.120 --> 01:54.120 You should make scales. 01:54.280 --> 02:02.010 This is your pay attention, your application is working fine and the model is predicting data as you 02:02.250 --> 02:02.850 need. 02:02.890 --> 02:09.610 But there is one thing you missed that is the scaled features. 02:09.730 --> 02:13.330 Back to our Google Colab here. 02:13.570 --> 02:21.450 If you remember, let me scroll down before training the model, we talked about feature scaling. 02:21.770 --> 02:28.090 This is especially crucial for neural networks and many other machine learning algorithms. 02:28.370 --> 02:34.050 It standardizes the range of independent features in the data to a common scale. 02:34.290 --> 02:44.210 So here the formula is very simple X, which is the data or the feature value minus mean over standard 02:44.210 --> 02:45.090 deviation. 02:45.330 --> 02:52.410 In this case, we need to get the mean for every feature and the standard deviation of every feature 02:52.410 --> 02:56.050 in order to transform our data to this. 02:56.210 --> 03:00.410 So this step is not included in our model. 03:00.410 --> 03:03.050 We need to include it in Android Studio. 03:03.050 --> 03:12.010 So we're going to do this as similar to this in Android Studio in order to get the correct and the exact 03:12.010 --> 03:12.770 values. 03:12.770 --> 03:23.250 So again guys, since we get an unacceptable data and unacceptable, um, like predictions and despite 03:23.250 --> 03:33.650 everything is working fine because we are not using data scaling, because weight is dominating over 03:33.650 --> 03:45.530 all the features we need to set, every feature has its own effect, has its own scale, and all the 03:45.650 --> 03:49.250 features has a common Unit. 03:49.250 --> 03:58.490 If you remember, guys, we talked about cylinders like in CC, horsepower in um, in HP, weight in 03:58.490 --> 04:01.890 pounds, acceleration in second and so on. 04:01.890 --> 04:06.490 So every feature is, has its own unit. 04:06.530 --> 04:12.810 We need to unify all those features into one scale. 04:12.850 --> 04:13.330 Okay. 04:13.570 --> 04:23.810 So I need from you to, uh, to focus with me and pay attention how I need to use standard scaler, 04:23.810 --> 04:28.530 how to use feature scaling in our Android application. 04:28.530 --> 04:33.050 So we're going to use this step in Android Studio. 04:33.090 --> 04:41.650 At first we need to get the standard deviation and the mean of every feature before we move to Android 04:41.650 --> 04:42.250 Studio. 04:42.250 --> 04:46.940 So let me scroll up here in the statistical summary. 04:46.940 --> 04:54.700 And step number three understanding data, we get the mean and the standard deviation. 04:54.700 --> 04:57.460 So std standard deviation and the mean. 04:57.500 --> 05:03.620 We need to um to subtract x from mean over standard deviation. 05:03.620 --> 05:12.180 So please pay attention cup those values the standard deviation and the mean of cylinders displacement 05:12.180 --> 05:15.100 horsepower weight acceleration model year and so on. 05:15.100 --> 05:18.100 So let me start with cylinders. 05:18.260 --> 05:24.900 So let me take here Open Android Studio go to Car Mileage Predictor. 05:25.140 --> 05:34.180 And here inside this class let me create a new variable of type array. 05:34.580 --> 05:38.260 So here feature scaling. 05:38.540 --> 05:41.540 This is the most important step. 05:41.780 --> 05:48.580 So private val feature means equals to float array of. 05:49.020 --> 05:51.820 And let's start with a cylinder. 05:51.860 --> 05:54.700 The cylinders is 5.4 05:54.940 --> 05:59.380 71939F. 05:59.580 --> 06:01.180 This is for cylinders. 06:01.180 --> 06:03.220 So let me mark it like this. 06:03.220 --> 06:05.020 Let's continue with the others. 06:25.980 --> 06:32.540 We can take the origin one mean similar to origin two mean and origin three mean. 06:32.580 --> 06:33.140 Okay. 06:33.300 --> 06:39.220 Later if we get errors we can modify those values okay. 06:39.260 --> 06:47.950 Now let's create another variable similar to what we've done before, but this time for the feature 06:48.190 --> 06:50.270 standard deviations. 06:50.270 --> 06:52.430 So float array of. 06:52.630 --> 06:58.430 Here we need to use the standard deviation of the data. 06:58.430 --> 07:01.310 So copy this 1.7. 07:01.350 --> 07:06.510 This is for cylinders standard deviation and so on. 07:15.990 --> 07:18.790 Is the cylinders standard deviation. 07:18.790 --> 07:21.350 This is for displacement. 07:21.390 --> 07:31.710 This is for horsepower the weight acceleration model year origin one, origin two and origin three. 07:31.750 --> 07:35.830 Those are the standard deviation and the feature means.