WEBVTT 00:00.080 --> 00:00.840 Welcome back. 00:00.840 --> 00:07.160 If you remember from the previous videos, we have our linear relationship price equals to three times 00:07.200 --> 00:09.040 size plus two plus noise. 00:09.320 --> 00:14.840 We give the model the size in square feet which is the input layer. 00:14.840 --> 00:17.720 We have only one number, one input. 00:18.040 --> 00:26.160 That dense layer will start learning w and B, which are the slope and the intercept respectively. 00:26.600 --> 00:33.160 Then in the output layer we get the price and calculate the price according to the learned equation. 00:33.160 --> 00:35.520 We take size. 00:35.520 --> 00:38.360 We take one number as input. 00:38.400 --> 00:42.480 We get price which is another number. 00:42.640 --> 00:44.560 So the input is one number. 00:44.560 --> 00:47.120 The output is one number. 00:47.160 --> 00:48.840 This is very important. 00:48.840 --> 00:53.880 You should understand what are the model input and output. 00:53.920 --> 00:58.440 Back to Android Studio we created this load model file. 00:58.600 --> 01:04.890 Now let's create the function that will do the Their prediction. 01:05.130 --> 01:11.930 So here function predict how many variables we need to pass. 01:11.970 --> 01:15.770 We need to pass only one variable as an input. 01:15.770 --> 01:23.410 So here the size input which is a float and how many outputs we get. 01:23.450 --> 01:26.610 We need to return only one float. 01:26.650 --> 01:27.210 Okay. 01:27.370 --> 01:30.610 So here we need to return a float. 01:30.650 --> 01:32.210 Let's start with val. 01:32.450 --> 01:45.250 Input size equals to array of float array of and we pass the size input here the shape to the array 01:45.250 --> 01:47.370 or shape to the array. 01:47.610 --> 01:49.090 Let me move it to here. 01:49.130 --> 01:51.890 Set the shape one one also. 01:52.090 --> 01:54.730 Val output price. 01:54.770 --> 01:59.010 Array of float array of zero. 01:59.170 --> 02:02.170 Also the shape is one one. 02:02.210 --> 02:04.410 Please pay attention with me. 02:04.450 --> 02:10.740 TensorFlow Lite models expect multi-dimensional arrays or tensors. 02:10.860 --> 02:20.740 Even if your model takes a single number exactly like our situation, we only need to pass the size 02:20.740 --> 02:22.900 of the house as an input. 02:22.900 --> 02:25.060 So we have only one single number. 02:25.220 --> 02:28.060 It must be wrapped in arrays. 02:28.100 --> 02:33.580 Float array of this function gives us a 1D array. 02:33.780 --> 02:43.540 We wrap the size input into 1D array, and in order to transform this 1D array into 2D array, we use 02:43.580 --> 02:51.540 float array of array of float array of x into transformed into 2D array. 02:51.740 --> 02:54.420 So the size input is float. 02:54.660 --> 02:58.020 We when we put it with float array of. 02:58.060 --> 03:03.100 This will transform our data our float into 1D array. 03:03.620 --> 03:08.860 If we wrapped inside array of, we get a 2D array. 03:09.100 --> 03:12.070 So guys this is a very important step. 03:12.070 --> 03:14.310 You should pay attention to it. 03:14.510 --> 03:19.670 We need to transform our float into a tensor. 03:19.710 --> 03:23.550 A tensor is a 2D array or multi-dimensional arrays. 03:23.590 --> 03:30.830 Okay so again guys the size input is float float array of is 1D array. 03:31.070 --> 03:41.270 Transforming the float into 1D array and array of transforming it into a 2D array similar to the output 03:41.270 --> 03:41.990 price. 03:42.030 --> 03:48.470 Okay, so here we are preparing an empty output container for TensorFlow Lite to fill in. 03:48.670 --> 03:58.270 It's initialized with zero f, but will be overwritten and shape one one again one prediction result 03:58.270 --> 03:59.750 for one input. 03:59.750 --> 04:05.310 So the shape one prediction result for one input. 04:05.350 --> 04:08.550 Then we need to run the interpreter. 04:08.550 --> 04:11.150 So run inference with the input data. 04:11.190 --> 04:17.120 Start with a return Try interpreter dot run. 04:17.160 --> 04:22.800 Pass the input size and the output price. 04:22.800 --> 04:27.560 Set the output price zero zero and the Logcat. 04:27.600 --> 04:34.680 Here if we have an error log ve failed to predict the interpreter dot run. 04:34.720 --> 04:42.640 Here we are using the interpreter that we initialized before and run it with the model. 04:42.640 --> 04:49.880 So here we are running the model and passing the input and the output the input input size. 04:49.880 --> 04:52.080 The output is the output price. 04:52.240 --> 05:01.760 Since interpreter is nullable, the safe call exclamation mark dot avoids crashes if it is not initialized. 05:01.760 --> 05:11.800 And here after inference, the result is stored inside the output, returning the prediction result. 05:11.800 --> 05:21.410 First row first column of the 2D array stores result in the first and only element of the 2D array, 05:21.410 --> 05:23.810 and this returns a float. 05:23.810 --> 05:31.210 And here we have error handling in case if there is any error, if anything goes wrong, it logs the 05:31.210 --> 05:33.570 error and returns null. 05:33.610 --> 05:44.170 In short, this function takes a single number which is the size input feeds it into your TensorFlow 05:44.410 --> 05:50.290 Lite TensorFlow Lite model and returns the predicted number. 05:50.410 --> 05:54.650 It's a wrapper that shapes the input correctly. 05:54.650 --> 05:58.130 Like as we learned here, size is a float. 05:58.290 --> 06:01.330 We transformed it to a 1D array. 06:01.370 --> 06:11.010 Then we transformed it to 2D array, runs the inference and extract the and returns the result safety, 06:11.170 --> 06:15.010 storing it into the first element of the 2D array.