Mastering TensorFlow 2.x: Implement Powerful Neural Nets across Structured, Unstructured datasets and Time Series Data

Cover Page
Title Page
Copyright Page
Dedication Page
About the Author
About the Reviewer
Acknowledgement
Preface
Errata
Table of Contents
1. Getting Started with TensorFlow 2.x
    Introduction
    Structure
    Objective
    Installing TensorFlow 2.x
        Installation on Ubuntu
        Installing TensorFlow 2.x with GPU support
    Keras high-level APIs integration into TensorFlow
        Python binding
        Functional Spec
        Layer
        Keras graph
    Writing the first sample using TensorFlow 2.x
        Preprocess the data
            Build the model
            Compile the model
            Train the model
            Verifying predictions
    Low-level APIs
        Dataflow
        tf.Graph structure
        tf.Operation and tf.Tensor
    Image classification with CNN
    Conclusion
    Questions
    References
2. Machine Learning with TensorFlow 2.x
    Introduction
    Structure
    Objectives
    Basic classification with TensorFlow 2.x
        Clean the dataset
        Model creation and training
    Regression examples with TensorFlow 2.x
        The Boston Housing Price dataset
        Preparing the data
            Building the neural network
    Validating our approach using K-fold validation
    Overfitting and underfitting
        Sonar dataset
    Saving and restoring models
    Conclusion
    Question
3. Keras Based APIs
    Introduction
    Structure
    Objective
    Keras functional API
        Training, evaluation, and prediction using TensorFlow and Keras
            Using training and evaluation loops
    Loss, metrics, and an optimizer
        Regression problem
        Binary classification problem
        Implementation of loss functions in TensorFlow 2.x
        Mean Square Error (MSE)
        Sparse cross categorical entropy
        Cross categorical entropy
    Layers and models using Keras functional APIs
    Effective TensorFlow 2.x
        Reorganization of namespaces
            Deprecated APIs
            Eager execution
            Functions replace sessions
            Removing globals
            Control flow
            Combine tf.data.Dataset and tf.function
    Conclusion
    Questions
    Code listing
    References
4. Convolutional Neural Networks
    Introduction
    Structure
    Objective
    Introduction to convolutional networks
        Short comings of fully-connected networks
    Convolution in TensorFlow 2.7
    Simple convolutional network with TensorFlow
    Building ResNet with TensorFlow
        ResNet
    Advanced computer vision techniques
        VGG architecture
        Learning from the VGG network
        Increased depth of feature maps
        VGG in TensorFlow
        Area under the curve for VGG16
        VGG19
        Inception V3
        InceptionV3 architecture
    TensorFlow Hub
    Summary
    Questions
    References
    Code listing
5. Recurrent Neural Networks
    Introduction
    Structure
    Objectives
    Basic concepts
    Simple recurrent neural network
        Building your first SimpleRNN network
            Weight constraints
    Gated recurrent unit
        Building your first GRU-based model
    Long-Term Short-Term Memory (LSTM)
        LSTM cell operations
            Text processing with LSTM
    Bidirectional LSTM
        Text processing with bidirectional LSTM
    Conclusion
    Points to remember
    Multiple choice questions
        Answers
    Questions
    Key terms
6. Time Series Forecasting with TensorFlow
    Introduction
    Structure
    Objectives
    Background
        Machine learning and time series
    Common patterns in time series
    First time series notebook with synthetic data
    Synthetic dataset
        Single layer model to predict time series
        Multiple layer model for predicting the value
    RNN for predicting the time series data
        Applying Lambda function
        Adjusting the learning rate dynamically
        LSTM and time series
        Process synthetic dataset with LSTM
    Conclusion
    References
    Points to remember
    Multiple choice questions
        Answers
    Questions
    Key terms
7. Distributed Training
    Introduction
    Structure
    Objectives
    Introduction to distributed TensorFlow
    Types of strategies
        Mirror strategy
        Multi-worker mirrored strategy
            Collective communication in a multi-worker mirrored strategy
            Code walk-through of MirroredStrategy
    Running MirroredStrategy on multi CPU AWS virtual machine
    Multi-worker training with Keras with two processes
    MNIST dataset and model definition
    Multi-worker configuration with two workers
    Environment variables and subprocesses in notebooks
        Strategy to be chosen
        Train the model
    Conclusion
    Questions
    References
8. Reinforcement Learning
    Introduction
    Structure
    Objective
    Sequential decision process: agent and the world
    Markov decision process
    Model free policy
    DQN networks
    Atari game with deep reinforcement learning - DQN
    Actor critic network
    Introducing TF-agents
    TF-agent
    DQN based agent for CartPole game
    SAC agent’s support in TF-agents
    Conclusion
    Questions
    Code listing
    References
    Appendix
        Policy evaluation
        Policy iteration
9. Techniques to do Model Optimization
    Introduction
    Structure
    Objective
    Background
    Extension to neural networks
    Tools available
    Advantages of model optimization
    Quantization
        Post-training quantization
        Overview
        Build a Fashion MNIST model
        Convert to a TensorFlow Lite model
        Run the TFLite models
        Evaluate the models
    Weight pruning
        Pruning some of the layers
        Pruning using sequential APIs
    Weight clustering
        Enabling Cluster weights
    Serializing the clustered model
    Weight clustering MNIST classification model
    Conclusion
    Questions
        Answers
    References
10. Generative Adversarial Networks
    Introduction
    Structure
    Objective
    Introducing GANs
    Discriminator
    Generator
    BCE cost function
        GAN for FASHION MNIST images
    DCGAN
        Generate and save images
    Problem with the loss function
    Earth mover’s distance
    WGAN
        Implementing WGAN
    Pix2Pix with Maps dataset
    Conclusion
    Multiple choice questions
        Answers
    Code listing
    References
Index