Trouble implementing custom environment

[allenjeffreywu]

Hey team, I am having a tough time implementing CityLearn with MARLlib. I believe my first attempt was flawed. This is attempt number 2. I cannot get mappo to work with it, and it looks like I will have to edit the internals of MARLlib to make it work (which I want to avoid as MUCH as possible). My current issue is with action spaces in cc_mlp.py . My action space is a Gym Dict and the shape of it just never seems to work. I am curious about the opponent_actions input and what creates it, as well as ways to debug this without print statements. I am also curious as to the expected shape of opponent_actions_ls in the end before it is torch.cat()

Here is my repo:
https://github.com/allenjeffreywu/MARLlib_CityLearn2

Please take a look if possible or at least answer my questions to help with development. I hoped that MARLlib would be simple enough to implement but it looks like it isn't.

[Morphlng]

From what I understand, the observation space and action space in MARLlib is for a single agent, instead a dict for each agent. In other words, it does not support heterogeneous obs/action space.

[zxnga]

hey @allenjeffreywu I'm trying to work with the same environment as you, did you make it work ? i have troubles (see #239)

[allenjeffreywu]

@TheoZan No, I did not get it to work.