TensorFlow in Action

Name: TensorFlow in Action
ISBN: 9787121309120

黄文坚 / Tang YuanPublishing House of Electronics Industry2017

3.6

Word count: 361K words
Chapters: 12
ISBN: 9787121309120

Report an error Find this book

Summary

TensorFlow in Action is co-authored by two TensorFlow developers, Huang Wenjian and Tang Yuan, and written against the TensorFlow 1.0 API for Chinese readers who want to master the framework end to end, from principles to code. The book is clearly structured: the first two chapters introduce TensorFlow's basic concepts, programming model (computational graphs, Session, Variable, Placeholder, and more), and design philosophy, and compare it side by side with mainstream deep learning frameworks such as Caffe, Theano, MXNet, and Torch. Chapters 3 and 4 start from installation and compilation and use handwritten digit recognition, Softmax regression, autoencoders, and multilayer perceptrons (MLPs) for a first hands-on practice. Chapters 5 and 6 explain the principles and implementation of convolutional neural networks and reproduce classic models one by one in code—AlexNet, VGGNet, Google Inception Net, and ResNet—while tracing the development of CNNs. Chapter 7 covers Word2Vec, LSTM-based language models, and bidirectional LSTM classifiers; Chapter 8 moves into deep reinforcement learning, implementing a policy network and a value network; Chapter 9 introduces TensorBoard visualization, single-machine multi-GPU parallelism, and distributed parallelism; and the final two chapters cover TF.Learn (Estimator, DataFrame, Monitors) and the statistical distributions, Layer modules, and performance profiler tfprof in TF.Contrib. Complete downloadable source code accompanies the book, with an emphasis on "code that runs and can be modified," making it one of the earlier technical books in China to systematically explain TensorFlow engineering practice.

Editor's review

这是一本“作者身份”相当硬的技术书：两位作者本身就是 TensorFlow 生态的参与者，唐源是 TensorFlow 的 Committer、TF.Learn 的作者之一，因此书中对框架设计意图的解释常常一句顶别处三段。它最大的价值在于“代码能跑、模型能复现”——把 AlexNet、VGGNet、Inception Net、ResNet、Word2Vec、LSTM、策略网络与估值网络这一整条深度学习模型谱系，用同一套 TensorFlow 代码风格串了起来，读者顺着抄一遍、改一遍，对深度学习的理解会从概念层面下沉到工程层面。后半部分讲 TensorBoard、多 GPU 与分布式并行、TF.Learn 与 TF.Contrib，是很多入门书略过、而工业界天天要用的内容，这部分内容当年填补了中文资料的空缺。当然也要如实提醒：全书基于 TensorFlow 1.0 的 API，如今 tf.Session、tf.placeholder 等写法在 TF2 中已被大幅重构，因此它的定位更适合“学模型结构与工程思路”，而不是“照着书搭今天的新项目”。对想理解深度学习框架如何运作、想补扎实基本功的读者，它仍然是一本值得翻开的中文经典。

Who it's for

A good fit for

已掌握 Python，想系统入门深度学习框架的开发者
需要理解 CNN、RNN、强化学习模型代码实现的学生与研究者
从事算法工程，想了解多 GPU 与分布式训练、TensorBoard 的工程师
读过理论书，希望动手跑通经典模型的转行者

Not a good fit for

完全零编程基础、需要从 Python 语法学起的读者
只想学 TensorFlow 2.x / Keras 最新写法、不愿做版本迁移的人
寻找前沿大模型、Transformer 内容的读者

Reading tips

先读第 1、2 章建立计算图与编程模型的概念，不要跳过框架对比部分。
第 3 章起务必动手跑代码，建议按书末提示到博文视点官网下载配套源码。
遇到 API 报错时优先查 TensorFlow 2.x 官方迁移文档，把书中写法对照翻译一遍。
读完每章模型后尝试只改数据集或超参数复现一次，检验是否真正掌握。

About the publisher

Publishing House of Electronics Industry China · Founded in 1982

The Publishing House of Electronics Industry is a comprehensive publisher distinguished by its focus on electronic information and related fields. Judging from the more than 3,000 titles in its catalogue, its publishing covers programming and design, artificial intelligence, computer science, finance and economics, photography, and social sciences, attending both to cutting-edge technology and to general readers. Representative titles include Factor Investing: Methods and Practice, x86 Assembly Language: From Real Mode to Protected Mode, The RISC-V Reader, Artificial Intelligence: Speech Recognition, Understanding and Practice, Phased Array Radar Technology, Radar Target Characteristics, and 100 Go Mistakes and How to Avoid Them, most of which centre on technical methods and engineering practice. Its authors include Shi Chuan, Li Zhong, Wang Xiaobo, David Patterson, Andrew Waterman, and Yu Zihao, among others from China and abroad. Owing to limited public information, no further publishing features are elaborated here.

Chapter 1 TensorFlow Basics
Chapter 2 TensorFlow Versus Other Deep Learning Frameworks
Chapter 3 First Steps with TensorFlow
Chapter 4 Implementing Autoencoders and Multilayer Perceptrons with TensorFlow
Chapter 5 Implementing Convolutional Neural Networks with TensorFlow
Chapter 6 Implementing Classic Convolutional Neural Networks with TensorFlow
Chapter 7 Implementing Recurrent Neural Networks and Word2Vec with TensorFlow
Chapter 8 Implementing Deep Reinforcement Learning with TensorFlow
Chapter 9 TensorBoard, Multi-GPU Parallelism, and Distributed Parallelism
Chapter 10 TF.Learn from Beginner to Expert
Chapter 11 Other Components of TF.Contrib
References

FAQ

这本书用的是 TensorFlow 1.x 还是 2.x？

全书基于 TensorFlow 1.0 版本的 API 编写，出版于 2017 年。如今 TF2 已默认启用 Eager Execution 并整合 Keras，书中 tf.Session、tf.placeholder 等写法需要改造或使用兼容模式，学习时建议同时参考官方迁移指南。

零基础能直接读这本书吗？

不太建议。书中假定读者具备 Python 编程能力，并了解机器学习的基本概念（如损失函数、梯度下降、过拟合），直接上手容易卡在环境和数学部分。

书里的源代码怎么获取？

出版社在书内说明可通过博文视点官网（broadview.com.cn）下载书中配套的大段源代码，按书名检索对应图书页面即可找到资源下载入口。

两大作者是谁？

黄文坚为 TensorFlow Contributor，曾任明略数据技术合伙人，从事金融风控与数据挖掘；唐源是 TensorFlow、XGBoost、MXNet 的 committer，也是 TF.Learn、ggfortify 的作者之一。

这本书现在还值得读吗？

如果目标是快速上手公司的新项目，建议直接学 TF2 / PyTorch 的现行文档；但若想弄懂深度学习框架的底层机制、经典网络结构与分布式训练的组织方式，这本书的讲解依然清晰而有参考价值。

Comments

Find this book

This site is an index: you start “Search all sources” in your browser, and the entries it lists all point to third-party sites; please judge their availability yourself.

About the author

黄文坚 Author

Huang Wenjian is an author about whom limited public information is available; what can be confirmed is mainly his connection to computer technology writing. Only one work by him is held in the library, TensorFlow in Action (TensorFlow Shizhan), which is commonly classified under programming and design. Judging from the title and category, the book addresses the practical use of the TensorFlow deep learning framework and belongs to the field of technical introductions and engineering practice rather than theoretical monographs or academic research. It can therefore be inferred that the author focuses on explaining framework usage, model construction, and code implementation clearly, in a direct and pragmatic style aimed at readers who wish to learn by doing. For beginners and engineers, such a book is valuable for lowering the barrier to entry and offering workable, referenceable procedures. Because details such as dates of birth and death, nationality, education, and professional experience remain unconfirmed, this profile does not cover his personal life and only describes objectively the subject matter and orientation of the available work.