import numpy as np
import multiprocessing as mp
from xuance.environment.utils import space2shape
from xuance.environment.vector_envs.vector_env import VecEnv
from xuance.environment.vector_envs import clear_mpi_env_vars, flatten_list, CloudpickleWrapper
[docs]
def worker(remote, parent_remote, env_fn_wrappers, env_seed: int = None):
def step_env(env, action):
obs, reward_n, terminated, truncated, info = env.step(action)
if all(terminated.values()) or truncated:
obs_reset, info_reset = env.reset()
info["reset_obs"] = obs_reset
info["reset_avail_actions"] = info_reset['avail_actions']
info["reset_state"] = info_reset['state']
return obs, reward_n, terminated, truncated, info
parent_remote.close()
if env_seed is None:
envs = [env_fn_wrapper() for env_fn_wrapper in env_fn_wrappers.x]
else:
envs = [env_fn_wrapper(env_seed=env_seed + i_env) for i_env, env_fn_wrapper in enumerate(env_fn_wrappers.x)]
try:
while True:
cmd, data = remote.recv()
if cmd == 'step':
remote.send([step_env(env, action) for env, action in zip(envs, data)])
elif cmd == 'reset':
remote.send([env.reset() for env in envs])
elif cmd == 'render':
remote.send([env.render(data) for env in envs])
elif cmd == 'close':
remote.send([env.close() for env in envs])
remote.close()
break
elif cmd == 'get_env_info':
env_info = envs[0].env_info
remote.send(CloudpickleWrapper(env_info))
elif cmd == 'get_groups_info':
env_info = envs[0].groups_info
remote.send(CloudpickleWrapper(env_info))
else:
raise NotImplementedError
except KeyboardInterrupt:
print('SubprocVecEnv worker: got KeyboardInterrupt')
finally:
for env in envs:
env.close()
[docs]
class SubprocVecMultiAgentEnv(VecEnv):
"""
VecEnv that runs multiple environments in parallel in subproceses and communicates with them via pipes.
Recommended to use when num_envs > 1 and step() can be a bottleneck.
"""
def __init__(self, env_fns, env_seed, context='spawn', in_series=1):
"""
Arguments:
env_fns: iterable of callables - functions that create environments to run in subprocesses. Need to be cloud-pickleable
in_series: number of environments to run in series in a single process
(e.g. when len(env_fns) == 12 and in_series == 3, it will run 4 processes, each running 3 envs in series)
"""
self.waiting = False
self.closed = False
self.in_series = in_series
num_envs = len(env_fns)
assert num_envs % in_series == 0, "Number of envs must be divisible by number of envs to run in series"
self.n_remotes = num_envs // in_series
env_fns = np.array_split(env_fns, self.n_remotes)
ctx = mp.get_context(context)
self.remotes, self.work_remotes = zip(*[ctx.Pipe() for _ in range(self.n_remotes)])
self.ps = [ctx.Process(target=worker, args=(work_remote, remote, CloudpickleWrapper(env_fn)))
for (work_remote, remote, env_fn) in zip(self.work_remotes, self.remotes, env_fns)]
if env_seed is None:
self.ps = [ctx.Process(target=worker, args=(work_remote, remote, CloudpickleWrapper(env_fn)))
for (work_remote, remote, env_fn) in zip(self.work_remotes, self.remotes, env_fns)]
else:
self.ps = [ctx.Process(target=worker, args=(work_remote, remote, CloudpickleWrapper(env_fn),
env_seed + ith_remote * in_series))
for (ith_remote, work_remote, remote, env_fn) in zip(
range(self.n_remotes), self.work_remotes, self.remotes, env_fns)]
for p in self.ps:
p.daemon = True # if the main process crashes, we should not cause things to hang
with clear_mpi_env_vars():
p.start()
for remote in self.work_remotes:
remote.close()
self.remotes[0].send(('get_env_info', None))
self.env_info = self.remotes[0].recv().x
self.viewer = None
VecEnv.__init__(self, num_envs, self.env_info['observation_space'], self.env_info['action_space'])
self.agents = self.env_info['agents']
self.num_agents = self.env_info['num_agents']
self.state_space = self.env_info['state_space'] # Type: Box
self.buf_state = [np.zeros(space2shape(self.state_space)) for _ in range(self.num_envs)]
self.buf_obs = [{} for _ in range(self.num_envs)]
self.buf_avail_actions = [{} for _ in range(self.num_envs)]
self.buf_info = [{} for _ in range(self.num_envs)]
self.actions = None
self.max_episode_steps = self.env_info['max_episode_steps']
self.remotes[0].send(('get_groups_info', None))
self.groups_info = self.remotes[0].recv().x
[docs]
def reset(self):
self._assert_not_closed()
for remote in self.remotes:
remote.send(('reset', None))
result = [remote.recv() for remote in self.remotes]
result = flatten_list(result)
obs, info = zip(*result)
self.buf_obs = list(obs)
self.buf_info = list(info)
self.buf_state = [info[e]['state'] for e in range(self.num_envs)]
self.buf_avail_actions = [info[e]['avail_actions'] for e in range(self.num_envs)]
return list(obs), list(info)
[docs]
def step_async(self, actions):
self._assert_not_closed()
actions = np.array_split(actions, self.n_remotes)
for remote, action in zip(self.remotes, actions):
remote.send(('step', action))
self.waiting = True
[docs]
def step_wait(self):
self._assert_not_closed()
results = [remote.recv() for remote in self.remotes]
results = flatten_list(results)
self.waiting = False
obs, rewards, terminated, truncated, info = zip(*results)
self.buf_obs = list(obs)
self.buf_info = list(info)
self.buf_state = [info[e]['state'] for e in range(self.num_envs)]
self.buf_avail_actions = [info[e]['avail_actions'] for e in range(self.num_envs)]
return list(obs), list(rewards), list(terminated), list(truncated), list(info)
[docs]
def render(self, mode):
self._assert_not_closed()
for pipe in self.remotes:
pipe.send(('render', mode))
imgs = [pipe.recv() for pipe in self.remotes]
imgs = flatten_list(imgs)
return imgs
def _assert_not_closed(self):
assert not self.closed, "Trying to operate on a SubprocVecEnv after calling close()"
def __del__(self):
if not self.closed:
self.close()
[docs]
class SubprocVecEnv_StarCraft2(SubprocVecMultiAgentEnv):
def __init__(self, env_fns, env_seed, context='spawn', in_series=1):
super(SubprocVecEnv_StarCraft2, self).__init__(env_fns, env_seed, context, in_series)
self.num_enemies = self.env_info['num_enemies']
self.battles_game = np.zeros(self.num_envs, np.int32)
self.battles_won = np.zeros(self.num_envs, np.int32)
self.dead_allies_count = np.zeros(self.num_envs, np.int32)
self.dead_enemies_count = np.zeros(self.num_envs, np.int32)
[docs]
def step_wait(self):
self._assert_not_closed()
results = [remote.recv() for remote in self.remotes]
results = flatten_list(results)
self.waiting = False
obs, rewards, terminated, truncated, info = zip(*results)
self.buf_obs = list(obs)
self.buf_info = list(info)
self.buf_state = [info[e]['state'] for e in range(self.num_envs)]
self.buf_avail_actions = [info[e]['avail_actions'] for e in range(self.num_envs)]
for i in range(self.num_envs):
if all(terminated[i].values()) or truncated[i]:
self.battles_game[i] += 1
if info[i]['battle_won']:
self.battles_won[i] += 1
self.dead_allies_count[i] += info[i]['dead_allies']
self.dead_enemies_count[i] += info[i]['dead_enemies']
return list(obs), list(rewards), list(terminated), list(truncated), list(info)