Deep Learning and the Game of Go

It started out good, albeit lacking in code commentary. Starting at chapter 4, which has pretty critical components to understand, it becomes difficult to follow with no indication of where the code should go in terms of file structure. From what I can tell, the chapter has no mention of any file in which and of the code snippets should go. There's no perspective on how the code in the chapter fits at a higher level.

This book does a good walkthrough of the basics of machine learning for establishing the go game front end, plus standards used for exchanging Go information, both overall game and move by move basis. Then it adds how to set up a deep learning system.

I've read most ML books out there and none are as informative as this one.It'll teach you the basics of ML, how you can encode features and outputs using DNN is invaluable and an important topic which is often missed in most books. Go is used as an excellent case study and the historical design decisions are well documented and give you an amazing insider look of how to think about solving new ML problems.

Although the title is "Deep Learning and the Game of Go", the focus (and strength) of this book is Deep Reinforcement learning (DRL), one of the hottest Machine Learning topics right now. The authors present the major new DRL algorithms by applying them to the ancient game of Go, which - and this is what makes this an enjoyable read - neither so easy that you constantly ask yourself "what do I need these fancy techniques for?" nor so hard as to require Google-scale hardware to get to end up with a bit that you can perceive to play intelligently (though not at the AlphaGo level). One of the strengths of the book compared to others on DRL is that it does not shy away from the question of how to obtain training data, preprocess it and encode the game states, which makes it feel quite self-contained and useful on its own.To get the most out of this book you should be a proficient software developer (not necessarily in Python), both to make sense of the numerous code snippets and to get the examples to run (great care has been taken to present self-contained and runnable code in the first few chapters, but expect to run into some minor hiccups with the code in later sections), but also to have a go at adapting the ideas to another game (or games?) of your choice.You should also have at least some interest in Go, but it's not necessary to be able to actually play Go (I myself was familiar with the rules, but apart from that am a total Go beginner).

Overall, I thought this book was pretty good. I've read a bunch of the O'Reilly books, but I thought this one approached the topic in a better way, for the most part.On the one hand, it makes a lot of assumptions, primarily about your coding ability, familiarity with machine learning and mathematics generally.On the other hand, it doesn't really get bogged down with tangents. The authors did a good job focusing on the task at hand, and methodically go through building up a deep learning system that comes together to play Go, while still touching on the things you need to know.Ultimately, I thought this book was pretty good, and I recommend it.

Good hands-on tutorial. However, the reinforcement learning part (last two chapters) hides the forest of the whole RL field by only showing one tree.