OpenSpielEnv¶
- torchrl.envs.OpenSpielEnv(*args, **kwargs)[source]¶
Google DeepMind OpenSpiel environment wrapper built with the game string.
GitHub: https://github.com/google-deepmind/open_spiel
Documentation: https://openspiel.readthedocs.io/en/latest/index.html
- Parameters:
game_string (str) – the name of the game to wrap. Must be part of
available_envs
.- Keyword Arguments:
device (torch.device, optional) – if provided, the device on which the data is to be cast. Defaults to
None
.batch_size (torch.Size, optional) – the batch size of the environment. Defaults to
torch.Size([])
.allow_done_after_reset (bool, optional) – if
True
, it is tolerated for envs to bedone
just afterreset()
is called. Defaults toFalse
.group_map (MarlGroupMapType or Dict[str, List[str]]], optional) – how to group agents in tensordicts for input/output. See
MarlGroupMapType
for more info. Defaults toALL_IN_ONE_GROUP
.categorical_actions (bool, optional) – if
True
, categorical specs will be converted to the TorchRL equivalent (torchrl.data.Categorical
), otherwise a one-hot encoding will be used (torchrl.data.OneHot
). Defaults toFalse
.return_state (bool, optional) – if
True
, “state” is included in the output ofreset()
andstep()
. The state can be given toreset()
to reset to that state, rather than resetting to the initial state. Defaults toFalse
.
- Variables:
available_envs – environments available to build
Examples
>>> from torchrl.envs import OpenSpielEnv >>> from tensordict import TensorDict >>> env = OpenSpielEnv("chess", return_state=True) >>> td = env.reset() >>> td = env.step(env.full_action_spec.rand()) >>> print(td) TensorDict( fields={ agents: TensorDict( fields={ action: Tensor(shape=torch.Size([2, 4672]), device=cpu, dtype=torch.int64, is_shared=False)}, batch_size=torch.Size([]), device=None, is_shared=False), next: TensorDict( fields={ agents: TensorDict( fields={ observation: Tensor(shape=torch.Size([2, 20, 8, 8]), device=cpu, dtype=torch.float32, is_shared=False), reward: Tensor(shape=torch.Size([2, 1]), device=cpu, dtype=torch.float32, is_shared=False)}, batch_size=torch.Size([2]), device=None, is_shared=False), current_player: Tensor(shape=torch.Size([]), device=cpu, dtype=torch.int32, is_shared=False), done: Tensor(shape=torch.Size([1]), device=cpu, dtype=torch.bool, is_shared=False), state: NonTensorData(data=FEN: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1 674 , batch_size=torch.Size([]), device=None), terminated: Tensor(shape=torch.Size([1]), device=cpu, dtype=torch.bool, is_shared=False)}, batch_size=torch.Size([]), device=None, is_shared=False)}, batch_size=torch.Size([]), device=None, is_shared=False) >>> print(env.available_envs) ['2048', 'add_noise', 'amazons', 'backgammon', ...]
reset()
can restore a specific state, rather than the initial state, as long asreturn_state=True
.>>> from torchrl.envs import OpenSpielEnv >>> from tensordict import TensorDict >>> env = OpenSpielEnv("chess", return_state=True) >>> td = env.reset() >>> td = env.step(env.full_action_spec.rand()) >>> td_restore = td["next"] >>> td = env.step(env.full_action_spec.rand()) >>> # Current state is not equal `td_restore` >>> (td["next"] == td_restore).all() False >>> td = env.reset(td_restore) >>> # After resetting, now the current state is equal to `td_restore` >>> (td == td_restore).all() True