Few days ago I found very good book for rllib framework, I know that it can look like advertisement but I spent a lot of time find book about rllib practical issues. This book shows step by step how to use rllib.
I hope it will be useful:
Mastering Reinforcement Learning with Python
Here is Github repo for the book , rllib examples start from Chapter 6: example
I recommend the book to better understand the code from github.
Keep in mind that Ray 2.0 is on the horizon and some API’s may not be stable over that change. The book is a little “older”.
For example the
config['learning_starts'] = 5000 is moving and being renamed.
If you want to learn about RLlib right now this is not too bad, since you can transition to Ray 2.0 later and take most of your knowledge with you.
Max Pumperla is also writing a book on Ray that will feature RLlib and that we can look forward to.
@arturn , Thank You fot the info about the book. Of course You are right that API will be changed, but for the beginner the main problem is to understand main ideas and main schemas how to use different algorithms with different envs, how to customize envs, how to use masking, how to configure cluster. There are so many configuration dictionaries that in some point beginner doesn’t know what configuration for what functionality is.
For example, is your config dictionary for cluster, experiment, algorithm, model, hyperparemeters tuning scheduler or something else.