check_env function of gym is giving some assertion errors after converting environment using GymEnv

[AvisP]

Environment

• Grid2op version: 1.7.2
• System: macos 12.6.1

Bug description

The check_env() function of gym is failing after converting the environment using the GymEnv class. The check_env function calls

• check_reset_return_type() gives the output

200 result = env.reset()
--> 201 assert isinstance(
    202     result, tuple
    203 ), f"The result returned by `env.reset()` was not a tuple of the form `(obs, info)`, where `obs` is a observation and `info` is a dictionary containing additional information. Actual type: `{type(result)}`"
    204 assert (
    205     len(result) == 2
    206 ), f"Calling the reset method did not return a 2-tuple, actual length: {len(result)}"
    208 obs, info = result

AssertionError: The result returned by `env.reset()` was not a tuple of the form `(obs, info)`, where `obs` is a observation and `info` is a dictionary containing additional information. Actual type: `<class 'collections.OrderedDict'>``
    299 check_reset_options(env) 

-check_reset_seed() gives

ValueError                                Traceback (most recent call last)
Cell In [5], line 1
----> 1 check_reset_seed(gym_env)

File ~/........../L2PRN/lib/python3.10/site-packages/gym/utils/env_checker.py:78, in check_reset_seed(env)
     73 if "seed" in signature.parameters or (
     74     "kwargs" in signature.parameters
     75     and signature.parameters["kwargs"].kind is inspect.Parameter.VAR_KEYWORD
     76 ):
     77     try:
---> 78         obs_1, info = env.reset(seed=123)
     79         assert (
     80             obs_1 in env.observation_space
     81         ), "The observation returned by `env.reset(seed=123)` is not within the observation space."
     82         assert (
     83             env.unwrapped._np_random  # pyright: ignore [reportPrivateUsage]
     84             is not None
     85         ), "Expects the random number generator to have been generated given a seed was passed to reset. Mostly likely the environment reset function does not call `super().reset(seed=seed)`."

ValueError: too many values to unpack (expected 2)

How to reproduce

from grid2op.gym_compat import GymEnv
import grid2op
from gym import Env
from gym.utils.env_checker import check_env, check_reset_return_type, check_reset_options, check_reset_seed

from grid2op.Backend import PandaPowerBackend
bk_cls = PandaPowerBackend
    
env_name = "l2rpn_case14_sandbox"
training_env = grid2op.make(env_name, test=True, backend=bk_cls())  
gym_env = GymEnv(training_env)

check_reset_return_type(gym_env)

check_reset_seed(gym_env)

Temporary Solution

By Navigating to /......../L2PRN/lib/python3.10/site-packages/grid2op/gym_compat/gymenv.pyI temporarily fixed it by setting return_info=True and adding super().reset(seed=seed) in the reset function https://github.com/rte-france/Grid2Op/blob/51854262d02d96c1ee1798a78a86661d1ff971a1/grid2op/gym_compat/gymenv.py#L59

[BDonnot]

Hi again

You are facing this issue because grid2op is not totally compatible with recent gym versions.

I'm currently (did that this afternoon) working toward it but it takes a bit of time as gym changed a lot recently.

For now I guess using gum 0.21 is probably the best thing to do. In gym 0.26 they enforce their new api for reset, seed etc which I had no time to work on (it's just simple glue code but it should be done regardless)

Thanks for the script provided, I will add it in the tests tomorrow or the day after when i will have time to finish the integration.

Thanks

[AvisP]

Hi again!

Yeah the gym environment has changed a lot and is causing issues specially with more values being returned.

This is a minor issue. I just wanted to mention it but may need fixing eventually.

Thanks