tinyspace

tinyspace

Dataclass

class tinyspace.TinySpace(shape, dtype, low, high, cls=None, desc=None)

A simple and lightweight spaces implementation for RL environments, in place of gym.spaces.

Any Space is a TinySpace with the following args or a dict where each value is a TinySpace.

You can either use this dataclass or just define a Space using dicts, since tinyspace prefers subscriptable access. However, because a TinySpace is a dataclass, dot access also works.

Parameters

• shape (Tuple) – Shape of space. Use -1 or None for variable-length dimensions.

• dtype (Union[dtype, dtype]) – Data type of space.

• low (Union[int, float, ndarray, Tensor]) – Lower bound of space.

• high (Union[int, float, ndarray, Tensor]) – Upper bound of space.

• desc (Optional[str]) – Description of space.

• cls (Optional[str]) – Type of space, such as [“discrete”, “box”].

Typing

tinyspace.Space

alias of Union[tinyspace.TinySpace, Mapping[str, tinyspace.TinySpace]]

Creating

tinyspace.space_from_dict(d)

Parameters

d (Union[Dict[str, TypeVar(T)], Dict[str, Dict]]) – a dict to convert to a Space

tinyspace.from_gym_space(gym_space, to_tinyspace=True)

Parameters

• gym_space – A gym.Space.

• to_tinyspace (bool) – If True, then converts to TinySpace dataclass. If False, then leaves it dict.

Return type

Union[TinySpace, Mapping[str, TinySpace]]

tinyspace.to_gym_space(space)

Convert from tinyspace to gym.spaces.

Parameters

space (Union[TinySpace, Mapping[str, TinySpace]]) – The space to convert.

tinyspace.convert_gymenv_spaces(gym_env, to_tinyspace=True)

Convert a gym.Env to use tinyspace instead of gym.spaces.

Parameters

gym_env – A gym.Env.

Using

tinyspace.is_ndshape(space)

Checks if space.shape is “nd”, meaning atleast one dimension has variable length.

Parameters

space (Union[TinySpace, Mapping[str, TinySpace]]) – A Space to check.

Return type

bool

tinyspace.sample_from_space(space, zeros=True, batch_size=1, ndim_pad=1, to_torch_tensor=False)

Return type

Union[Dict, ndarray, Tensor]

tinyspace.collate_obs(obses, space, to_torch_tensor=False)

Parameters

• obses (Union[List, Tuple]) – Batch of observations which need to be collated.

• space (Union[TinySpace, Mapping[str, TinySpace]]) – A TinySpace or dict where each value is a TinySpace.

• to_torch_tensor (bool) – If True, then convert to torch.Tensor.