Error when running example PR #84 (custom PettingZoo environment)

[tommasomarzi]

Hi!

I'd like to add a custom environment. To do this, I first tried to follow the instructions in the PR "[NOMRG] Example: adding custom PettingZoo env #84" to get familiar with the integration of a new environment. However, after cloning the repo and running run.py I get the error:

File "/Users/tommaso/Phd/current_projects/benchmarl_tests/BenchMARL/venv/lib/python3.10/site-packages/torchrl/envs/transforms/transforms.py", line 7484, in _generate_episode_reward_spec raise ValueError( ValueError: The in_key: reward is not present in the reward spec Composite( player: Composite( reward: UnboundedContinuous( shape=torch.Size([1, 3, 1]), space=ContinuousBox( low=Tensor(shape=torch.Size([1, 3, 1]), device=cpu, dtype=torch.float32, contiguous=True), high=Tensor(shape=torch.Size([1, 3, 1]), device=cpu, dtype=torch.float32, contiguous=True)), device=cpu, dtype=torch.float32, domain=continuous), device=cpu, shape=torch.Size([1, 3]), data_cls=None), device=cpu, shape=torch.Size([1]), data_cls=None).

Could you help me with this? I'm using an environment with python 3.10.10, torch 2.9.0, and torchRL 0.10.0.

Thanks in advance!

[Xmaster6y]

Hey,

I do not get this error, so I suppose you used your custom environment there, right?

There seems to be a problem with your reward specs, which, in the example, is handled by PettingZooWrapper with _make_specs and _make_group_specs. Maybe take it from there; otherwise, if you can provide a minimal reproduction code, that would be great.

[tommasomarzi]

Hi, thanks for the quick reply!

I do not get this error, so I suppose you used your custom environment there, right?

I was actually using the custom environment from the example. However, I rebuilt the virtual environment from scratch, and now it seems to work.

Do you have any recommended practices for installing the required packages (e.g., specific versions or tools to use)?

Also, I have a more general question: when building a custom environment, would you recommend always using the PettingZooWrapper, or does that depend on the specific environment?

[Xmaster6y]

uv is really the way to go for project management. And regarding your other question I don't have experienced much with custom envs so idk, but PettingZooWrapper seems easy.

[tommasomarzi]

Ok, thank you for your help!