WEBVTT 00:00.080 --> 00:08.120 After the successful loading of the model from the assets folder, let's create a new function that 00:08.360 --> 00:10.080 makes the predictions. 00:10.240 --> 00:20.120 So function predict mileage takes car features as a parameter and returns a float. 00:20.360 --> 00:23.480 The float means the mpg try. 00:23.520 --> 00:27.520 Let's start with a try and catch block. 00:27.640 --> 00:31.160 Prepare input tensor with exact scaling. 00:31.400 --> 00:34.360 So here prepare input tensor. 00:34.400 --> 00:41.040 Val input buffer equals to tensor flow buffer dot. 00:41.080 --> 00:44.040 Create fixed size. 00:44.200 --> 00:46.280 Set the parameter. 00:46.320 --> 00:48.720 The first parameter as the input shape. 00:48.760 --> 00:53.320 The second parameter is data type of type float. 00:53.440 --> 00:57.320 So data type dot Float32. 00:57.360 --> 01:00.400 This line of code is very important. 01:00.760 --> 01:08.780 Creating a properly shaped and typed container, which is the tensor buffer that holds the data. 01:09.100 --> 01:12.180 The input data your model expects. 01:12.380 --> 01:22.740 Think of it as creating the exact mold or form that you need to pour your data into before giving it 01:22.740 --> 01:24.020 to the model. 01:24.020 --> 01:25.380 So this is the shape. 01:25.380 --> 01:31.500 This is the mold, the form that you need, uh, for pouring the data into it. 01:31.700 --> 01:32.260 Okay. 01:32.300 --> 01:35.580 So here we are getting the input buffer. 01:35.780 --> 01:38.020 This is a very important step. 01:38.380 --> 01:47.900 The tensor buffer is a helper class from the TensorFlow Lite support library that wraps tensor data, 01:47.940 --> 01:56.100 that its purpose is to provide an easy way to create and manipulate multi-dimensional arrays tensors 01:56.260 --> 01:58.660 that Tflite models understand. 01:58.780 --> 02:07.740 Okay, so this is the tensor buffer, like a specialized data box designed specifically for ML model 02:07.740 --> 02:10.860 input output create fixed sized. 02:10.900 --> 02:17.640 This is a function creating a TensorFlow buffer with a fixed predefined shape. 02:17.920 --> 02:26.960 The buffer size can't change after creation, and this is common for models that expect inputs of consistent 02:26.960 --> 02:27.480 size. 02:27.480 --> 02:35.560 For example fixed image dimensions, fixed number of features as in our example, and the input shape 02:35.800 --> 02:38.360 where an integer uh array. 02:38.400 --> 02:44.200 Defining dimensions and shapes that we defined before and this is of type. 02:44.400 --> 02:45.520 Let me show you. 02:45.560 --> 02:48.400 It's an int array of 1 to 9. 02:48.400 --> 02:51.440 As I said to you one sample nine features. 02:51.440 --> 02:53.160 We need to pass nine features. 02:53.440 --> 03:01.640 And this input buffer and Float32 specifies the data types of the elements in the TensorFlow 32 bit 03:01.640 --> 03:03.160 floating point numbers. 03:03.320 --> 03:07.840 Most common for ML models I want from you to focus with me. 03:07.840 --> 03:11.560 Now here is a very important step. 03:11.560 --> 03:20.360 I'll show you the features without scaling at first, and then we're going to move to make the scaled 03:20.360 --> 03:23.910 features according to our Google Colab. 03:23.910 --> 03:26.910 So features without scaling. 03:27.110 --> 03:30.630 And this is the main purpose behind scaling. 03:30.630 --> 03:36.550 So I'll create the features, I'll create the app, run the app and everything will work fine. 03:36.550 --> 03:43.950 But I'll show you the importance of scaling features if you didn't understand it in the Google Colab. 03:43.990 --> 03:52.470 Don't worry, I'm going to show you here and I am very confident that you should understand it well. 03:52.510 --> 03:53.190 Val. 03:53.310 --> 03:56.550 Features without scaling. 03:56.710 --> 03:58.510 This is the name of the variable. 03:58.710 --> 04:00.910 It's a float array of. 04:01.430 --> 04:11.950 And here let me start with the car features in order the cylinders the displacement the car features 04:11.950 --> 04:19.430 horsepower, weight, acceleration, model year and the origin. 04:19.430 --> 04:25.870 So car features dot origin one, origin two and origin three. 04:26.110 --> 04:31.970 Okay, we need to pass this this array into our input buffer. 04:31.970 --> 04:36.370 So input buffer dot load data. 04:36.410 --> 04:40.650 Load array data features without scaling. 04:40.770 --> 04:43.770 Then we prepare the output tensor. 04:43.890 --> 04:46.450 Val output buffer. 04:46.450 --> 04:52.290 As we did in the input buffer we will create for the output buffer tensor buffer dot. 04:52.290 --> 04:53.690 Create fixed size. 04:53.810 --> 04:58.050 Set the output shape and float32 this. 04:58.210 --> 05:00.970 Then we need to run the inference. 05:01.210 --> 05:03.010 Start with the interpreter. 05:03.050 --> 05:08.570 Run and set the input buffer buffer and the output buffer to output buffer. 05:08.890 --> 05:10.170 This is the interpreter. 05:10.170 --> 05:17.890 This is the run function that we have to pass two parameters the input buffer and the output buffer. 05:17.890 --> 05:20.090 Let's get the prediction results. 05:20.090 --> 05:32.650 So here val results equals to output buffer dot float array val prediction equals to results at index 05:32.650 --> 05:33.250 zero. 05:33.290 --> 05:36.050 Let me explain the output buffer. 05:36.110 --> 05:42.550 This is a tensor buffer that contains the model's prediction output and the float array. 05:42.790 --> 05:47.110 This property extracts the underlying data as a float array. 05:47.270 --> 05:54.870 What it does converts the tensors internal representation into a simple Kotlin array that you can easily 05:54.870 --> 05:55.750 work with. 05:55.790 --> 06:00.270 And here the prediction equals to results at index zero. 06:00.470 --> 06:03.910 Accesses the first element of the result array. 06:04.070 --> 06:12.150 For regression models like your car mileage prediction app, the output is typically a single value. 06:12.150 --> 06:21.070 So for this we use the result at index zero and the first element of the results array return the prediction 06:21.110 --> 06:23.150 result the prediction. 06:23.150 --> 06:33.110 And here we need to use the catch close e exception, throw runtime exception and print the message 06:33.150 --> 06:33.670 okay. 06:33.830 --> 06:38.870 This is how we use the function predict mileage. 06:38.870 --> 06:42.550 And this is the job of our application.