def __init__(self, process: Process, worker_id: int, conn: Connection):
self.process = process
self.worker_id = worker_id
self.conn = conn
self.previous_step: EnvironmentStep = EnvironmentStep(None, {}, None)
self.previous_all_action_info: Dict[str, ActionInfo] = {}
self.waiting = False
def __init__(self, env: BaseEnv,
float_prop_channel: FloatPropertiesChannel):
super().__init__()
self.shared_float_properties = float_prop_channel
self.env = env
self.previous_step: EnvironmentStep = EnvironmentStep(None, {}, None)
self.previous_all_action_info: Dict[str, ActionInfo] = {}
def _postprocess_steps(
self, env_steps: List[EnvironmentResponse]
) -> List[EnvironmentStep]:
step_infos = []
timer_nodes = []
for step in env_steps:
payload: StepResponse = step.payload
env_worker = self.env_workers[step.worker_id]
new_step = EnvironmentStep(
env_worker.previous_step.current_all_brain_info,
payload.all_brain_info,
env_worker.previous_all_action_info,
)
step_infos.append(new_step)
env_worker.previous_step = new_step
if payload.timer_root:
timer_nodes.append(payload.timer_root)
if timer_nodes:
with hierarchical_timer("workers") as main_timer_node:
for worker_timer_node in timer_nodes:
main_timer_node.merge(
worker_timer_node, root_name="worker_root", is_parallel=True
)
return step_infos
def reset(
self, config: Dict[str, float] = None
) -> List[EnvironmentStep]: # type: ignore
if config is not None:
for k, v in config.items():
self.shared_float_properties.set_property(k, v)
all_brain_info = self.env.reset()
self.previous_step = EnvironmentStep(None, all_brain_info, None)
return [self.previous_step]
def reset(self, config: Optional[Dict] = None) -> List[EnvironmentStep]:
while any(ew.waiting for ew in self.env_workers):
if not self.step_queue.empty():
step = self.step_queue.get_nowait()
self.env_workers[step.worker_id].waiting = False
# First enqueue reset commands for all workers so that they reset in parallel
for ew in self.env_workers:
ew.send("reset", config)
# Next (synchronously) collect the reset observations from each worker in sequence
for ew in self.env_workers:
ew.previous_step = EnvironmentStep(None, ew.recv().payload, None)
return list(map(lambda ew: ew.previous_step, self.env_workers))
def reset(
self,
config: Dict[str, float] = None,
train_mode: bool = True,
custom_reset_parameters: Any = None,
) -> List[EnvironmentStep]: # type: ignore
all_brain_info = self.env.reset(
config=config,
train_mode=train_mode,
custom_reset_parameters=custom_reset_parameters,
)
self.previous_step = EnvironmentStep(None, all_brain_info, None)
return [self.previous_step]
def step(self) -> List[EnvironmentStep]:
all_action_info = self._take_step(self.previous_step)
self.previous_all_action_info = all_action_info
for brain_name, action_info in all_action_info.items():
self.env.set_actions(brain_name, action_info.action)
self.env.step()
all_brain_info = self._generate_all_brain_info()
step_brain_info = all_brain_info
step_info = EnvironmentStep(
self.previous_step.current_all_brain_info,
step_brain_info,
self.previous_all_action_info,
)
self.previous_step = step_info
return [step_info]
def step(self) -> List[EnvironmentStep]:
all_action_info = self._take_step(self.previous_step)
self.previous_all_action_info = all_action_info
actions = {}
values = {}
for brain_name, action_info in all_action_info.items():
actions[brain_name] = action_info.action
values[brain_name] = action_info.value
all_brain_info = self.env.step(vector_action=actions, value=values)
step_brain_info = all_brain_info
step_info = EnvironmentStep(
self.previous_step.current_all_brain_info,
step_brain_info,
self.previous_all_action_info,
)
self.previous_step = step_info
return [step_info]
def step(self) -> List[EnvironmentStep]:
all_action_info = self._take_step(self.previous_step)
self.previous_all_action_info = all_action_info
actions = {}
memories = {}
texts = {}
values = {}
for brain_name, action_info in all_action_info.items():
actions[brain_name] = action_info.action
memories[brain_name] = action_info.memory
texts[brain_name] = action_info.text
values[brain_name] = action_info.value
all_brain_info = self.env.step(actions, memories, texts, values)
step_brain_info = all_brain_info
step_info = EnvironmentStep(
self.previous_step.current_all_brain_info,
step_brain_info,
self.previous_all_action_info,
)
self.previous_step = step_info
return [step_info]
def __init__(self, env: BaseUnityEnvironment):
super().__init__()
self.env = env
self.previous_step: EnvironmentStep = EnvironmentStep(None, {}, None)
self.previous_all_action_info: Dict[str, ActionInfo] = {}
