def __init__(self,
name: str,
n_repeat_action: int = 1,
n_workers: int = 8,
blocking: bool = True,
**kwargs):
"""
:param name: Name of the Environment
:param env_class: Class of the environment to be wrapped.
:param n_workers: number of workers that will be used.
:param blocking: step the environments asynchronously.
:param args: args of the environment that will be parallelized.
:param kwargs: kwargs of the environment that will be parallelized.
"""
super(ParallelDMControl, self).__init__(name=name)
envs = [
ExternalDMControl(name=name,
n_repeat_action=n_repeat_action,
**kwargs) for _ in range(n_workers)
]
self._batch_env = BatchEnv(envs, blocking)
self._env = DMControlEnv(name,
n_repeat_action=n_repeat_action,
**kwargs)
self.observation_space = self._env.observation_space
self.action_space = self._env.action_space
class ParallelDMControl(Environment):
"""Wrap a dm_control environment to be stepped in parallel."""
def __init__(self, name: str, n_repeat_action: int=1, n_workers: int=8,
blocking: bool=True, **kwargs):
"""
:param name: Name of the Environment
:param env_class: Class of the environment to be wrapped.
:param n_workers: number of workers that will be used.
:param blocking: step the environments asynchronously.
:param args: args of the environment that will be parallelized.
:param kwargs: kwargs of the environment that will be parallelized.
"""
super(ParallelDMControl, self).__init__(name=name)
envs = [ExternalDMControl(name=name, n_repeat_action=n_repeat_action, **kwargs)
for _ in range(n_workers)]
self._batch_env = BatchEnv(envs, blocking)
self._env = DMControlEnv(name, n_repeat_action=n_repeat_action, **kwargs)
self.observation_space = self._env.observation_space
self.action_space = self._env.action_space
def __getattr__(self, item):
return getattr(self._env, item)
def step_batch(self, actions: np.ndarray, states: np.ndarray=None,
n_repeat_action: [np.ndarray, int]=None):
return self._batch_env.step_batch(actions=actions, states=states,
n_repeat_action=n_repeat_action)
def step(self, action: np.ndarray, state: np.ndarray=None, n_repeat_action: int=None):
return self._env.step(action=action, state=state, n_repeat_action=n_repeat_action)
def reset(self, return_state: bool = True, blocking: bool=True):
state, obs = self._env.reset(return_state=True)
self.sync_states()
return state, obs if return_state else obs
def get_state(self):
return self._env.get_state()
def set_state(self, state):
self._env.set_state(state)
self.sync_states()
def sync_states(self):
self._batch_env.sync_states(self.get_state())
def __init__(self,
name: str,
n_repeat_action: int = 1,
height: float = 100,
width: float = 100,
wrappers=None,
n_workers: int = 8,
blocking: bool = True,
**kwargs):
"""
:param name: Name of the Environment
:param env_class: Class of the environment to be wrapped.
:param n_workers: number of workers that will be used.
:param blocking: step the environments asynchronously.
:param args: args of the environment that will be parallelized.
:param kwargs: kwargs of the environment that will be parallelized.
"""
super(ParallelRetro, self).__init__(name=name)
envs = [
ExternalRetro(name=name,
n_repeat_action=n_repeat_action,
height=height,
width=width,
wrappers=wrappers,
**kwargs) for _ in range(n_workers)
]
self._batch_env = BatchEnv(envs, blocking)
self._env = RetroEnvironment(name,
n_repeat_action=n_repeat_action,
height=height,
width=width,
wrappers=wrappers,
**kwargs)
self._env.init_env()
self.observation_space = self._env.observation_space
self.action_space = self._env.action_space