def _init() -> gym.Env:
env = gym.make(env_id)
# Create folder if needed
if log_dir is not None:
os.makedirs(log_dir, exist_ok=True)
env = Monitor(env, log_dir)
env.seed(seed + rank)
return env
def _init():
set_random_seed(seed + rank)
env = gym.make(env_id, **env_kwargs)
# Wrap first with a monitor (e.g. for Atari env where reward clipping is used)
log_file = os.path.join(log_dir,
str(rank)) if log_dir is not None else None
# Monitor success rate too for the real robot
info_keywords = ('is_success', ) if 'NeckEnv' in env_id else ()
env = Monitor(env, log_file, info_keywords=info_keywords)
# Dict observation space is currently not supported.
# https://github.com/hill-a/stable-baselines/issues/321
# We allow a Gym env wrapper (a subclass of gym.Wrapper)
if wrapper_class:
env = wrapper_class(env)
env.seed(seed + rank)
return env
def _init() -> gym.Env:
env = gym.make(env_id)
env = Monitor(env, log_dir)
env.seed(seed + rank)
return env
def _init():
env = gym.make(env_id, reward_type="dense")
env = gym.wrappers.FlattenObservation(env)
env = Monitor(env)
env.seed(seed + rank)
return env
