Mastering Reinforcement Learning with Python: Build next-generation, self-learning models using reinforcement learning techniques and best practices Paperback – 18 Dec. 2020

A very clear grounding on the core RL algorithms. The narrative is a good mix of theoretical, code and insights across a wide range of algorithms. The early chapters on Value based methods are a great combination of the theory, code explanation and intuition. The explanation of the algorithm methods is just about right balance, between heavy academic and practical insights.I am using this narrative to help me get back up to speed on an understanding policy based networks and specifically PPO algorithm for background in Unity ML-Agents, after some previous experience with DQN networks. So I am slightly disappointed that there are no code development on the initial Policy based methods, to grasp an understanding on how to implement REINFORCE and PPO from the narrative. But I guess there are plenty of implementations available online. Instead the book defers to the use of the Ray[RRLib] implementations to demonstrate the various algorithms, performance and an intuition between each. The latter sections drift into narrative and great references to frameworks, rather than offering code. Which is probably realistic for all the subjects that Author wishes to cover.Note the author advises on a Linux based setup, and the use of Ray[RLlib]. There is a note towards Github Issue on how to checkpoint and save the RLlib network models. After all the training compute, it is rather frustrating to discover that RLlib does not save the models by default. Many of us will have to adjust the samples from 50 workers, down to to 10, as I have to live in the real world of a basic gaming PC to perform local machine learning.This book covers an awful lot of ground and insight into core Reinforcement learning, that I have struggled to find expressed so well elsewhere.

Reinforcement leaning is a relatively new and uncommonly used algorithm that’s gaining some traction.Most books on RL focus on the theory extensively but not much on the code.This book covers both the theory and the code in good detail.I like the flow of this book also because it puts RL in context (Multi-Armed Bandits, Contextual Bandits and Markov Decision Process) then followed by solving the Reinforcement Learning ProblemWith this background, the book then explores Deep reinforcement learning and then finally applications.Thus, the book is unique in that it balances the theory, code and applications in a pragmatic and a practical manner

Great intro but needs a basic ML and probability background.

Best RL book I've ever read, clear with great examples and good balance of theory and applications