[Feature Request] Make check_env assertions in regard to observation_space more actionable

[FieteO]

🚀 Feature

When checking the implementation of a custom gym with check_env there are a number of ways to fail the validation.
For the observation_space, the dtype or the shape could be wrong (as indicated in #746, #921 and #932) that will all lead to the same AssertionError:

AssertionError: The observation returned by the `reset()` method does not match the given observation space

This is telling the user that something is wrong, but it's not clear what exactly is causing the validation failure.

Out of my head, I could think of two more actionable error messages that would help a user understand what is wrong:
For the wrong shape:

AssertionError: The observation returned by the `reset()` method does not match the given observation space. Expected shape: (x,y), actual: (x,z)

For the wrong dtype:

AssertionError: The observation returned by the `reset()` method does not match the given observation space. Array was of type: float64, expected float32

Arguably there are more failure modes associated with this error.
(Since I ran into this and have assured that the above reasons do not apply)

P.S. If you have suggestions what else can let this assertion fail, I would appreciate that to help me find my issue...

Motivation

No response

Pitch

No response

Alternatives

No response

Additional context

No response

Checklist

• I have checked that there is no similar issue in the repo

[araffin]

Hello,
sounds like a good improvement, feel free to submit a PR ;)

[FieteO]

I justed started taking a look in the code :D
Can you tell me, what observation_space.contains(obs) is checking exactly?

    assert observation_space.contains(
        obs
    ), f"The observation returned by the `{method_name}()` method does not match the given observation space"

I suppose contains is this method, which is defined as

    def contains(self, x) -> bool:
        """Return boolean specifying if x is a valid member of this space."""
        if not isinstance(x, np.ndarray):
            logger.warn("Casting input x to numpy array.")
            try:
                x = np.asarray(x, dtype=self.dtype)
            except (ValueError, TypeError):
                return False

        return bool(
            np.can_cast(x.dtype, self.dtype)
            and x.shape == self.shape
            and np.all(x >= self.low)
            and np.all(x <= self.high)
        )

[FieteO]

@araffin I have now submitted a PR. Since I have irreparably broken my env (something with setuptools or wheels I suppose), would you be so kind as to run the other required checks on your end?

[araffin]

Closed by #1400