search

Stanford-ILIAD / PantheonRL

PantheonRL is a package for training and testing multi-agent reinforcement learning environments. PantheonRL supports cross-play, fine-tuning, ad-hoc coordination, and more.
MIT License
121 stars 18 forks source link

About the conflict between Stable Baselines and PantheonRL #9

Closed skydvn closed 1 year ago

skydvn commented 1 year ago

Dear authors, I run the examples of pettingzoo on colab. However I still face these problems.

2 Using cpu device Wrapping the env with a Monitor wrapper Wrapping the env in a DummyVecEnv.

AssertionError Traceback (most recent call last) in 10 print(env.n_players) 11 for i in range(env.n_players - 1): ---> 12 partner = OnPolicyAgent(PPO('MlpPolicy', env.getDummyEnv(i), verbose=1)) 13 14 # The second parameter ensures that the partner is assigned to a certain

5 frames /usr/local/lib/python3.8/dist-packages/stable_baselines3/common/vec_env/util.py in obs_space_info(obs_space) 65 subspaces = {i: space for i, space in enumerate(obs_space.spaces)} 66 else: ---> 67 assert not hasattr(obs_space, "spaces"), f"Unsupported structured space '{type(obs_space)}'" 68 subspaces = {None: obs_space} 69 keys = []

AssertionError: Unsupported structured space '<class 'gymnasium.spaces.dict.Dict'>'

bsarkar321 commented 1 year ago

Hi! Thanks for letting us know about this issue. The reason for this behavior is a change in PettingZoo's requirements since it now uses OpenAI's gym version 0.27.0 whereas SB3 (and PantheonRL) use 0.21.0. I'm in the process of writing a patch for this, which will be out in a few days.

bsarkar321 commented 1 year ago

Ok, I think the latest commits should resolve this issue!