Source code for stable_baselines3.common.vec_env.subproc_vec_env

import multiprocessing as mp
import warnings
from collections import OrderedDict
from typing import Any, Callable, Dict, List, Optional, Sequence, Tuple, Type, Union

import gymnasium as gym
import numpy as np
from gymnasium import spaces

from stable_baselines3.common.vec_env.base_vec_env import (
from stable_baselines3.common.vec_env.patch_gym import _patch_env

def _worker(
    remote: mp.connection.Connection,
    parent_remote: mp.connection.Connection,
    env_fn_wrapper: CloudpickleWrapper,
) -> None:
    # Import here to avoid a circular import
    from stable_baselines3.common.env_util import is_wrapped

    env = _patch_env(env_fn_wrapper.var())
    reset_info: Optional[Dict[str, Any]] = {}
    while True:
            cmd, data = remote.recv()
            if cmd == "step":
                observation, reward, terminated, truncated, info = env.step(data)
                # convert to SB3 VecEnv api
                done = terminated or truncated
                info["TimeLimit.truncated"] = truncated and not terminated
                if done:
                    # save final observation where user can get it, then reset
                    info["terminal_observation"] = observation
                    observation, reset_info = env.reset()
                remote.send((observation, reward, done, info, reset_info))
            elif cmd == "reset":
                observation, reset_info = env.reset(seed=data)
                remote.send((observation, reset_info))
            elif cmd == "render":
            elif cmd == "close":
            elif cmd == "get_spaces":
                remote.send((env.observation_space, env.action_space))
            elif cmd == "env_method":
                method = getattr(env, data[0])
                remote.send(method(*data[1], **data[2]))
            elif cmd == "get_attr":
                remote.send(getattr(env, data))
            elif cmd == "set_attr":
                remote.send(setattr(env, data[0], data[1]))  # type: ignore[func-returns-value]
            elif cmd == "is_wrapped":
                remote.send(is_wrapped(env, data))
                raise NotImplementedError(f"`{cmd}` is not implemented in the worker")
        except EOFError:

[docs]class SubprocVecEnv(VecEnv): """ Creates a multiprocess vectorized wrapper for multiple environments, distributing each environment to its own process, allowing significant speed up when the environment is computationally complex. For performance reasons, if your environment is not IO bound, the number of environments should not exceed the number of logical cores on your CPU. .. warning:: Only 'forkserver' and 'spawn' start methods are thread-safe, which is important when TensorFlow sessions or other non thread-safe libraries are used in the parent (see issue #217). However, compared to 'fork' they incur a small start-up cost and have restrictions on global variables. With those methods, users must wrap the code in an ``if __name__ == "__main__":`` block. For more information, see the multiprocessing documentation. :param env_fns: Environments to run in subprocesses :param start_method: method used to start the subprocesses. Must be one of the methods returned by multiprocessing.get_all_start_methods(). Defaults to 'forkserver' on available platforms, and 'spawn' otherwise. """ def __init__(self, env_fns: List[Callable[[], gym.Env]], start_method: Optional[str] = None): self.waiting = False self.closed = False n_envs = len(env_fns) if start_method is None: # Fork is not a thread safe method (see issue #217) # but is more user friendly (does not require to wrap the code in # a `if __name__ == "__main__":`) forkserver_available = "forkserver" in mp.get_all_start_methods() start_method = "forkserver" if forkserver_available else "spawn" ctx = mp.get_context(start_method) self.remotes, self.work_remotes = zip(*[ctx.Pipe() for _ in range(n_envs)]) self.processes = [] for work_remote, remote, env_fn in zip(self.work_remotes, self.remotes, env_fns): args = (work_remote, remote, CloudpickleWrapper(env_fn)) # daemon=True: if the main process crashes, we should not cause things to hang # pytype: disable=attribute-error process = ctx.Process(target=_worker, args=args, daemon=True) # type: ignore[attr-defined] # pytype: enable=attribute-error process.start() self.processes.append(process) work_remote.close() self.remotes[0].send(("get_spaces", None)) observation_space, action_space = self.remotes[0].recv() super().__init__(len(env_fns), observation_space, action_space)
[docs] def step_async(self, actions: np.ndarray) -> None: for remote, action in zip(self.remotes, actions): remote.send(("step", action)) self.waiting = True
[docs] def step_wait(self) -> VecEnvStepReturn: results = [remote.recv() for remote in self.remotes] self.waiting = False obs, rews, dones, infos, self.reset_infos = zip(*results) return _flatten_obs(obs, self.observation_space), np.stack(rews), np.stack(dones), infos
[docs] def reset(self) -> VecEnvObs: for env_idx, remote in enumerate(self.remotes): remote.send(("reset", self._seeds[env_idx])) results = [remote.recv() for remote in self.remotes] obs, self.reset_infos = zip(*results) # Seeds are only used once self._reset_seeds() return _flatten_obs(obs, self.observation_space)
[docs] def close(self) -> None: if self.closed: return if self.waiting: for remote in self.remotes: remote.recv() for remote in self.remotes: remote.send(("close", None)) for process in self.processes: process.join() self.closed = True
[docs] def get_images(self) -> Sequence[Optional[np.ndarray]]: if self.render_mode != "rgb_array": warnings.warn( f"The render mode is {self.render_mode}, but this method assumes it is `rgb_array` to obtain images." ) return [None for _ in self.remotes] for pipe in self.remotes: # gather render return from subprocesses pipe.send(("render", None)) outputs = [pipe.recv() for pipe in self.remotes] return outputs
[docs] def get_attr(self, attr_name: str, indices: VecEnvIndices = None) -> List[Any]: """Return attribute from vectorized environment (see base class).""" target_remotes = self._get_target_remotes(indices) for remote in target_remotes: remote.send(("get_attr", attr_name)) return [remote.recv() for remote in target_remotes]
[docs] def set_attr(self, attr_name: str, value: Any, indices: VecEnvIndices = None) -> None: """Set attribute inside vectorized environments (see base class).""" target_remotes = self._get_target_remotes(indices) for remote in target_remotes: remote.send(("set_attr", (attr_name, value))) for remote in target_remotes: remote.recv()
[docs] def env_method(self, method_name: str, *method_args, indices: VecEnvIndices = None, **method_kwargs) -> List[Any]: """Call instance methods of vectorized environments.""" target_remotes = self._get_target_remotes(indices) for remote in target_remotes: remote.send(("env_method", (method_name, method_args, method_kwargs))) return [remote.recv() for remote in target_remotes]
[docs] def env_is_wrapped(self, wrapper_class: Type[gym.Wrapper], indices: VecEnvIndices = None) -> List[bool]: """Check if worker environments are wrapped with a given wrapper""" target_remotes = self._get_target_remotes(indices) for remote in target_remotes: remote.send(("is_wrapped", wrapper_class)) return [remote.recv() for remote in target_remotes]
def _get_target_remotes(self, indices: VecEnvIndices) -> List[Any]: """ Get the connection object needed to communicate with the wanted envs that are in subprocesses. :param indices: refers to indices of envs. :return: Connection object to communicate between processes. """ indices = self._get_indices(indices) return [self.remotes[i] for i in indices]
def _flatten_obs(obs: Union[List[VecEnvObs], Tuple[VecEnvObs]], space: spaces.Space) -> VecEnvObs: """ Flatten observations, depending on the observation space. :param obs: observations. A list or tuple of observations, one per environment. Each environment observation may be a NumPy array, or a dict or tuple of NumPy arrays. :return: flattened observations. A flattened NumPy array or an OrderedDict or tuple of flattened numpy arrays. Each NumPy array has the environment index as its first axis. """ assert isinstance(obs, (list, tuple)), "expected list or tuple of observations per environment" assert len(obs) > 0, "need observations from at least one environment" if isinstance(space, spaces.Dict): assert isinstance(space.spaces, OrderedDict), "Dict space must have ordered subspaces" assert isinstance(obs[0], dict), "non-dict observation for environment with Dict observation space" return OrderedDict([(k, np.stack([o[k] for o in obs])) for k in space.spaces.keys()]) elif isinstance(space, spaces.Tuple): assert isinstance(obs[0], tuple), "non-tuple observation for environment with Tuple observation space" obs_len = len(space.spaces) return tuple(np.stack([o[i] for o in obs]) for i in range(obs_len)) # type: ignore[index] else: return np.stack(obs) # type: ignore[arg-type]