search

flow-project / flow

Computational framework for reinforcement learning in traffic control
MIT License
1.06k stars 374 forks source link

[Documentation] The readthedocs documentation is very limited #772

Open HenryJia opened 4 years ago

HenryJia commented 4 years ago

Sections on the code documentation is limited very much lacking detail. For example, the flow.envs documentation is pretty much empty https://flow.readthedocs.io/en/latest/flow.envs.html

It would be exceptionally useful if documentation was here as well as in the docstrings. I'd prefer not to have to dig through the codebase to find out which environment or network does what etc.

eugenevinitsky commented 4 years ago

Hi @HenryJia, thanks for pointing this out. While we figure out how to best upgrade the docs, the tutorials might still be helpful for you?

HenryJia commented 4 years ago

The tutorials are great for understanding the rough structure and setup. However, when I want to dig a little deeper and mess a little more (e..g. with what specific states I want to feed in to my neural networks), it's quite helpful to have the docs there. Also, the tutorials are a bit focused around using ray/rllib which I feel is a bit too black-box like for my liking. I want to be able to build my own networks and training from scratch on top of it

eugenevinitsky commented 4 years ago

Yeah, I don't disagree. We'll prioritize upgrading the docs.

eugenevinitsky commented 4 years ago

Btw, to make communication easier we have started up a slack channel for users to communciate more directly with devs and other users: https://join.slack.com/t/flow-users/shared_invite/enQtNzk5NzE1OTkyNTE0LTU0YmE5ZTVkZmJmNDg1Nzg2YThjOThmY2ZlYWE2ZjQ0NTE1NjE3Y2YwMTljY2U5M2Y3MDhmMWUzZDM3YWViNmE