Foundations of Deep Reinforcement Learning: Theory and Practice in Python

Epilogue

This book began by formulating an RL problem as an MDP. Parts I and II introduced the main families of deep RL algorithms that can be used to solve MDPs—policy-based, value-based, and combined methods. Part III focused on the practicalities of training agents, covering topics such as debugging, neural network architecture, and hardware. We also included a deep RL almanac containing information about hyperparameters and algorithm performance for some classic control and Atari environments from OpenAI Gym.

It was fitting to end the book by taking a look at environment design since this is an important part of using deep RL in practice. Without environments, there is nothing for an agent to solve. Environment design is a large and interesting ...