WEBVTT 00:00.120 --> 00:02.360 Hello developers, and welcome back. 00:02.400 --> 00:08.000 In this section we're going to build together the house price prediction application. 00:08.000 --> 00:11.920 We're going to mix machine learning with Android. 00:11.920 --> 00:18.160 We have two parts to work on the Android and the machine learning part. 00:18.200 --> 00:25.600 We need to train the model using TensorFlow library, testing the model actual versus predicted, and 00:25.600 --> 00:29.320 exporting the model as Tflite extension. 00:29.560 --> 00:34.120 Then we need to move to the Android Studio build UI. 00:34.200 --> 00:35.320 With Jetpack Compose. 00:35.320 --> 00:42.640 We're going to build our application with Jetpack Compose, add the Tflite TensorFlow model to the assets 00:42.640 --> 00:46.560 folder and using the model to predict prices. 00:46.600 --> 00:52.800 Now let's see the steps involved in training the models with TensorFlow. 00:52.840 --> 00:59.320 It's similar to what we've done before, but with simple and minor changes. 00:59.320 --> 01:02.080 But at first, what is TensorFlow? 01:02.240 --> 01:10.630 TensorFlow is an open source library and end to end platform for machine learning deployed by Google. 01:10.790 --> 01:19.790 It's primarily used for numerical computation and building and deploying machine learning models, especially 01:19.790 --> 01:21.710 deep neural networks. 01:21.710 --> 01:26.630 We have that many features and characteristics for TensorFlow. 01:26.910 --> 01:33.670 Like open source, it's freely available and maintained by a large community, allowing for widespread 01:33.870 --> 01:35.510 use and development. 01:35.590 --> 01:43.670 Deep learning focus while versatile for various machine learning tasks, it's particularly well suited 01:43.670 --> 01:51.070 for deep learning applications such as image recognition, neutral natural language processing, and 01:51.070 --> 01:52.550 predictive analytics. 01:52.550 --> 01:58.510 And by the way, we're going to use TensorFlow later for image recognition and object detection. 01:58.550 --> 02:07.220 TensorFlow utilizes dataflow graphs where nodes represent mathematical operations and edges represent 02:07.220 --> 02:13.140 multidimensional data arrays called tensors and flow between them. 02:13.180 --> 02:21.420 It can run on various platforms, including CPUs, GPUs, mobile devices, embedded systems, and specialized 02:21.460 --> 02:27.900 tensor processing units TPU and also we can run it on microcontrollers. 02:28.100 --> 02:36.980 TensorFlow offers a rich set of tools, libraries like Keras for high level API and community resources 02:36.980 --> 02:44.580 to facilitate the entire machine learning workflow, from data acquisition and model building to training 02:44.580 --> 02:45.780 and deployment. 02:45.780 --> 02:48.300 We're gonna use it from scratch. 02:48.300 --> 02:56.140 We're gonna install this library, import it, and we're gonna start with the first step by loading 02:56.340 --> 02:57.860 the data set. 02:58.260 --> 03:01.460 The first step is the data set. 03:01.660 --> 03:04.420 Load and preprocess your data. 03:04.460 --> 03:11.620 This may include tasks such as Normalization, resizing, augmentation, and splitting into training, 03:11.620 --> 03:14.220 validation, and test sets. 03:14.380 --> 03:20.540 Create TensorFlow data objects for efficient data handling during training. 03:20.780 --> 03:26.140 Then we're going to move to the second step which is build and compile the model. 03:26.340 --> 03:34.980 Define your model architecture using TensorFlow, Keras dot sequential or the functional API. 03:35.300 --> 03:41.980 Choose the appropriate levels that you need and then compile the model by specifying an optimizer, 03:42.260 --> 03:45.260 a loss function, and the other metrics. 03:45.260 --> 03:52.460 As we learned before about the MSE and the the MSE error and the R squared. 03:52.500 --> 03:54.580 Did you remember in the previous videos? 03:54.580 --> 03:59.620 Then we're going to go to the third step which is training the model. 03:59.660 --> 04:06.620 Use model dot fit method to train your model on the prepared training data. 04:06.780 --> 04:13.290 Provide validation data to monitor performance during training and detect overfitting. 04:13.490 --> 04:21.330 Consider using callbacks like model Checkpoint to save the best model weights, or early stopping to 04:21.370 --> 04:22.730 prevent overfitting. 04:22.890 --> 04:29.570 Then we're gonna move to the fourth step, which is save and convert model. 04:29.570 --> 04:32.890 We need to evaluate the model after training. 04:32.930 --> 04:40.690 Evaluate the model's performance on set on test sets using Model.evaluate method to get metrics like 04:40.690 --> 04:48.930 loss and accuracy, and then save the trained model to the dot to the TensorFlow format, which is dot 04:48.970 --> 04:49.570 PB. 04:49.810 --> 04:55.450 So save your model in TensorFlow Savedmodel format using Model.save method. 04:55.450 --> 05:00.970 This saves the model's architecture, weights and training configuration. 05:01.090 --> 05:03.810 By the way, this is the TensorFlow format. 05:03.810 --> 05:05.570 It's like a folder. 05:05.890 --> 05:10.570 And inside the folder we have model architecture weights and training configuration. 05:10.570 --> 05:13.960 Don't worry we're gonna see them in deep, deep details. 05:13.960 --> 05:21.840 In the next videos for Android, we need to get a file with extension TF Lite. 05:21.920 --> 05:26.360 TF Lite, now known as our or Lite RT. 05:26.720 --> 05:34.480 This is a framework that enables on device artificial intelligence, AI, and machine learning by optimizing 05:34.480 --> 05:42.600 and deploying models onto resource constrained devices like mobile phones, embedded systems, and microcontrollers. 05:42.600 --> 05:52.600 And this is the target from this section to get the TF Lite model and include it in our application 05:52.600 --> 05:53.800 Android application. 05:53.840 --> 05:54.320 Okay. 05:54.560 --> 06:03.400 So this is the steps we're going to follow in the next videos in order to get a well-trained model and 06:03.400 --> 06:11.560 import it into our Android application and use it in order to get the predictions for the prices of 06:11.600 --> 06:12.680 the houses.