Train ML Regression Models & Build Predictive Android Apps

Explore how machine learning uses data and algorithms to imitate human learning and make predictions. See dog breed recognition and salary prediction to illustrate data collection and model training.

Understand the three machine learning types—supervised, unsupervised, and reinforcement—and terms like data set, training data, testing data, features, and labels. See how supervised learning uses labeled data to predict outcomes.

Identify the two main types of supervised learning—regression and classification—and see how continuous labels like salary and categorical ones like gender are predicted from features such as height and weight.

Learn unsupervised learning with unlabeled data and clustering to discover patterns, and explore reinforcement learning as a trial-and-error method where AI agents learn from rewards and penalties.

Explore deep learning and artificial neural networks, including neurons and layers, to extract patterns, train models, and understand classification and regression with gender and house price examples.

Explore how a neural network predicts house price from area and rooms, using feed forwarding, backpropagation, and epoch-based training to adjust weights.

Explore basic deep learning concepts, including supervised and unsupervised learning, a neural network, feed forwarding and back propagation, learning rate, loss function, overfitting, and dropout.

Explore Google Colab as a browser-based, zero-configuration Python environment to train machine learning models for mobile apps, using code cells and notebooks with free GPU access and easy sharing.

Learn the basic syntax of Python, understand variables and data types, and practice using Google Colab notebooks to train machine learning models for mobile apps.

Learn how to create and delete Python numbers, print values, and inspect types; explore integer, long, float, and complex numbers, including declaring complex numbers with real and imaginary parts.

Discover how to declare Python strings with single quotes, print values, access characters by index, slice substrings, repeat strings, and concatenate using plus operators.

Explore Python lists, the versatile data type that can hold mixed values and nested lists. Learn indexing, slicing, and common methods like append, insert, remove, count, sort, and reverse.

Learn Python tuples and dictionaries, including tuple immutability, indexing for element access, and dicts with unique keys and key-based retrieval for building Android price prediction apps.

Discover Python basics for Android apps, covering if/elif/else conditionals, while and for loops, list and string iteration, and user input.

Learn how to read and write files in Python using the open function, with modes r, w, wb, b, and a, including creating, closing, and reading content.

Explore NumPy, the Python library for creating and manipulating large multi-dimensional arrays and matrices, and learn to import as np, create arrays, and inspect shape, size, and dtype.

Learn to create numpy arrays with np.zeros, np.ones, and np.random, fill with np.full, and reshape or flatten them for model training and dummy data.

Explore numpy operators, starting with unary operations to find min, max, and sum of an array, then apply binary operators for element-wise math and dot products.

Explore matrix multiplication in numpy using dot and matmul on two-dimensional arrays, then learn to sort matrices row-wise or column-wise with numpy’s sort function.

Discover pandas for python data manipulation and analysis by converting lists and dictionaries into data frames, a two-dimensional tabular structure with named columns like name and age.

Load csv data into a pandas data frame using read_csv, upload nba.csv in google colab, and inspect the 458-row, nine-column dataset of nba players for use in machine learning models.

Learn to handle missing values represented as NaN in a pandas data frame, using isnull, fill, and dropna, and check the data shape.

Explore matplotlib, a Python library for 2D plots of arrays, and learn to visualize data with plots, bars, and scatters using plt, in Google Colab.

Learn how to load and display images with matplotlib, load images with imread, and view image data as numpy arrays to visualize image shapes and values.

Discover how to import TensorFlow, declare and manipulate variables and arrays with tf.variable and the assign function, and create constants with tf.constant for mobile and cross-platform models.

Explore shaping and reshaping TensorFlow Lite tensors, inspect shapes like 2x3, rank and index elements, convert to numpy, and perform elementwise operations such as division and multiplication.

Explore tensor operations in TensorFlow, including transposing matrices, performing matmul with proper shapes, and typecasting, then learn ragged tensors with varying row lengths.

Explore tf.reduce_mean on TensorFlow Lite variables to compute means for scalars and 2D tensors, and specify axis to obtain column-wise or row-wise averages.

Generate and control random tensors in TensorFlow, with normal and uniform distributions, seeds for reproducibility, and basic tensor operations like add, concatenate, and finding min/max indices.

Save and restore thousands of tensor values using TensorFlow checkpoints, tracking training progress and restoring model state when needed.

Discover how TensorFlow Lite enables on-device machine learning for Android and iOS apps, converting trained models to TF Lite format and running them with the interpreter at the edge.

Train a simple regression model in Colab using TensorFlow, Keras, and numpy, then convert it to TensorFlow Lite for mobile apps.

Test a trained linear regression model by validating predictions against expected outputs, then convert the model to TensorFlow Lite using the Keras converter and save as linear.tf lite.

Train a Tensorflow Lite model from a simple single-input, single-output dataset by building a one-neuron network, selecting an optimizer and loss function, training over epochs, and converting to tflite.

Analyze a trained tflite model to inspect its input and output shapes using Natron. Confirm a single input of type float 32 and a single output value.

Create Android Studio project with Java, design gui with an edit text, a button, and text view in constraint layout, and initialize them to test a regression model y=2x-1.

Integrate a tflite regression model in android studio by adding an assets folder with linear.tflite, configuring a TensorFlow dependency, and loading the model with a TensorFlow Lite interpreter using ByteBuffer.

Pass user input from a text field to a TensorFlow Lite model in Android, convert to float, run inference, and display the output in a TextView.

Integrate a tflite regression model in an Android app by loading the model as a byte buffer from assets, initializing TensorFlow Lite interpreter, then performing inference and displaying the result.

Train a regression model to predict automobile fuel efficiency, covering data collection, preprocessing, normalization, one-hot encoding, neural network training, evaluation, and tflite conversion for Android apps.

Train a real-world regression model to predict fuel efficiency from car features, using the auto mpg dataset from Kaggle with mpg as the label and nine attributes, downloadable as csv.

Upload the dataset to Google Colab, import pandas and TensorFlow, load the csv with read_csv, select needed columns, handle missing values, and preview with tail to prepare for training.

Load the csv, check missing values with isna and sum, identify six missing horsepower entries, and drop or replace them to create a clean dataset for model training.

Apply one hot encoding to the origin column, creating USA, Europe, and Japan features and removing the origin column to improve model accuracy.

Scale numeric columns to a common range with normalization before training a model, preventing area from biasing house price predictions from area and rooms, and noting normalization types.

Normalize dataset features by subtracting the mean and dividing by the standard deviation to bring training and test data to a common scale, improving model training.

Train a fuel efficiency prediction model in TensorFlow using Keras sequential with two 64-neuron hidden layers and a single-neuron output, using MSE and MAE metrics.

Evaluate the trained model on the 20% test data, compare predictions to actual values, and note small MAE and MSE; then convert and save as automobile.tflite for mobile use.

Train a real-world regression model on a Kaggle dataset, apply one-hot encoding and normalization, use an 80/20 split to predict miles per gallon, and convert to tflite for Android apps.

Clone the starter android project from GitHub or course resources, import it in Android Studio, and run the emulator app to integrate the trained fuel efficiency model.

Explore the android starter code, including main activity and user interface layout, cylinder input, origin spinner with USA, Europe, and Japan, and a tf lite model setup for price prediction.

Load a TensorFlow Lite model in Android by creating an interpreter, loading automobile.tflite into a mapped ByteBuffer, inputting text fields, one-hot encoding regions, and normalizing with mean and standard deviation.

Normalize input data in the android app using training statistics. Subtract each feature’s mean and divide by its standard deviation for all nine inputs before predicting the price.

Pass nine inputs to a TensorFlow Lite model in Android with a 2d float array, preserve the defined order, and perform inference to obtain a single output, the mpg.

Test the fuel efficiency prediction android app on the simulator, inputting features from the Colab testing dataset to generate a prediction that closely matches the actual value.

Review how the fuel efficiency app is built: import starter code, load the model, and initialize the interpreter. Normalize inputs, select region, run inference, and display results.

Explore data collection for a house price prediction model, preprocess with normalization and one-hot encoding, train an artificial neural network, evaluate and convert to tflite for an Android app.

Explore obtaining a house price dataset from Kaggle, review features like longitude, latitude, house age, rooms, and ocean proximity, and prepare housing.csv in Google Colab for training.

Train a tflite house price model by loading a 20,000-row housing csv in Google Colab, handling missing values, one-hot encoding ocean proximity with pandas, and performing 80/20 train-test split.

Normalize the data, build a neural network with three hidden layers (64, 64, 1), train with an optimizer and loss function, and evaluate mean squared error and mean absolute error.

Add more layers and increase neurons in the house price prediction model, retrain with more parameters and convert to tflite, then evaluate against testing data to assess overfitting.

Clone the starter code from GitHub into Android Studio, install on emulator, and review the gui with inputs like longitude, latitude, house age, total rooms, population, income, and ocean proximity.

Examine the strata app interface and code, load a house price prediction tflite model in Android, and initialize inputs including longitude, latitude, and ocean proximity.

Normalize Android input data for house price prediction by converting inputs to floats, encoding ocean proximity categories, and applying per column mean and standard deviation to 13 features.

Pass normalized 13 input features to the house price prediction model via a 2d input array and a 1x1 output array, run inference, and display the predicted price.

Test the application with the test dataset, predict house prices, and compare predictions to actual values, then retrain the model to improve accuracy.