datawhalechina/easy-rl

Easy-RL, nicknamed the "Mushroom Book," is a Chinese-language textbook and tutorial series on reinforcement learning, a branch of machine learning where a program learns to make decisions by trying actions and receiving rewards or penalties. The mushroom name is a nod to Super Mario: the idea is that reading this book gives you a power-up, letting you explore reinforcement learning with growing confidence rather than being overwhelmed by its mathematical complexity. The content is drawn from several well-known Chinese university lecture series, primarily a deep reinforcement learning course by Professor Hung-yi Lee of National Taiwan University, known for making technical subjects accessible through game-based examples such as teaching an AI to play Atari games. Additional chapters pull from an introductory reinforcement learning course by Professor Bolei Zhou and a hands-on practical series by a world-champion reinforcement learning practitioner. The textbook covers 13 chapters progressing from fundamental concepts, through Markov decision processes (a mathematical framework for sequential decision-making), into specific algorithms including Q-learning, DQN (and its variants like Double DQN and Dueling DQN), policy gradient methods, the PPO algorithm, actor-critic methods, imitation learning, and handling sparse rewards. Each chapter comes with exercises and most include companion Jupyter Notebook code files so readers can run the algorithms and see them working directly. The book has been published in print by a Chinese publisher (People's Posts and Telecommunications Press) and is available from major Chinese book retailers. A free PDF version is available from the GitHub releases page. The online readable version is continuously updated at the project's website. All content is released under a Creative Commons BY-NC-SA 4.0 license, meaning it can be freely shared and adapted for non-commercial purposes with attribution.