TensorFlow in Action

This book is written based on the TensorFlow 1.0 API, aimed at developers and researchers who want to quickly turn deep learning ideas into runnable models. It begins with the basic principles of TensorFlow such as computation graphs, sessions and variables, and compares TensorFlow with other mainstream deep learning frameworks to help readers understand the design philosophy and applicable boundaries of this framework. In the practical part, the author uses complete code to implement various types of deep neural networks, including AutoEncoder, Multilayer Perceptron (MLP), convolutional neural networks (AlexNet, VGGNet, Inception Net, ResNet), Word2Vec word vectors, as well as recurrent neural networks (LSTM, bidirectional RNN) and the policy network and value network in deep reinforcement learning. It also specifically explains TensorBoard visualization, single-machine multi-GPU parallelism, distributed parallel training, and the use of other TensorFlow components such as TF.Learn. The book emphasizes runnable code and cases close to industrial scenarios; after reading, readers can independently build and train common deep learning models and understand the engineering practice points in multi-GPU and distributed environments. Note that the examples are based on the TensorFlow 1.x API and differ considerably from later 2.x versions in sessions, graph mode and other aspects, so appropriate adaptation is needed when working through them.