tf_agents.environments.gym_wrapper.spec_from_gym_space

Converts gym spaces into array specs.

tf_agents.environments.gym_wrapper.spec_from_gym_space(
    space: gym.Space,
    dtype_map: Optional[Dict[gym.Space, np.dtype]] = None,
    simplify_box_bounds: bool = True,
    name: Optional[Text] = None
) -> tf_agents.specs.BoundedArraySpec

Gym does not properly define dtypes for spaces. By default all spaces set their type to float64 even though observations do not always return this type. See: https://github.com/openai/gym/issues/527

To handle this we allow a dtype_map for setting default types for mapping spaces to specs.

observations. Not sure that we have a need for this yet.

space gym.Space to turn into a spec.
dtype_map A dict from spaces to dtypes to use as the default dtype.
simplify_box_bounds Whether to replace bounds of Box space that are arrays with identical values with one number and rely on broadcasting.
name Name of the spec.

A BoundedArraySpec nest mirroring the given space structure.

ValueError If there is an unknown space type.