Is it just me or does anyone else feel like rays quality of documentation is lacking? I like the concept of providing examples to show how to do certain things. But ray and rllib in particular have a complete lack of good API level documentation. Docs on a level similar to stable-baselines3 would make using and extending ray modules so much easier. PPO — Stable Baselines3 1.2.0a0 documentation
It would be much better if we had an API level documentation showing us what classes/methods are available what data types they take as input/arguments/return. Most real world problems can’t be solved using a couple of basic examples. We need to understand how to extend these objects to fit real world needs.
Not sure if this is going to get the right set of eyes. But that is my rant and I was just wondering does anyone else feel the same?
Oh this is awesome, I thought I had clicked on every link and couldn’t find it. I think the naming of “package reference” threw me off, the other ray modules are called “API reference”. Thank you for pointing this out to me.
I do think it would be beneficial to have at the end of each class/method definition a small example of how to use it. Sort of the way Numpy documents their API.
Some other things my team thinks would be beneficial would be to see some examples of extracting the final trained policy models and running data through it. It looks like RLlib does some internal data manipulation to feed data to the agents. I was able to get that to work by scouring the github to see what was happening.