def create_env(env_id, output_path, seed=0):
rank = MPI.COMM_WORLD.Get_rank()
set_global_seeds(seed + 10000 * rank)
env = gym.make(env_id)
env = Monitor(env, os.path.join(output_path, str(rank)), allow_early_resets=True)
env.seed(seed)
return env
def _init():
env = gym.make(env_id,
cf_id=rank + 1,
n_obstacles=n_obstacles,
gazebo=gazebo,
gazebo_process=gazebo_process,
cf_process=cf_process)
env = Monitor(env, log_dir)
env.seed(seed + rank)
return env
def _init():
env = make_atari(env_id)
# if 'BoxWorld' in env_id:
# print('using wrap_boxworld!')
# env = wrap_boxworld(env, episode_life=False, clip_rewards=False, frame_stack=False, scale=False)
# else:
# env = wrap_deepmind(env, episode_life=False, clip_rewards=False, frame_stack=False, scale=False)
if useMonitor:
env = Monitor(env, log_dir + str(rank), allow_early_resets=True)
env.seed(seed + rank)
return env
def make_mujoco_env(env_id, seed, allow_early_resets=True):
"""
Create a wrapped, monitored gym.Env for MuJoCo.
:param env_id: (str) the environment ID
:param seed: (int) the inital seed for RNG
:param allow_early_resets: (bool) allows early reset of the environment
:return: (Gym Environment) The mujoco environment
"""
set_global_seeds(seed + 10000 * mpi_rank_or_zero())
env = gym.make(env_id)
env = Monitor(env, os.path.join(logger.get_dir(), '0'), allow_early_resets=allow_early_resets)
env.seed(seed)
return env
def _init():
# env = make_atari(env_id)
env = football_env.create_environment(
env_name=env_name,
stacked=True,
render=True if 'pixel' in rep else False,
# rewards=FLAGS.reward_experiment,
logdir=log_dir,
enable_goal_videos=FLAGS.dump_scores and (seed == 0),
enable_full_episode_videos=FLAGS.dump_full_episodes
and (seed == 0),
dump_frequency=50 if FLAGS.render and seed == 0 else 0)
if useMonitor:
env = Monitor(env, log_dir + str(rank), allow_early_resets=True)
env.seed(seed + rank)
return env
def _init():
if sb_version == "sb3":
set_random_seed(seed + rank)
else:
set_global_seeds(seed + rank)
# this seed will be overridden by the last but one statement of this function
env = make_custom_env(seed=0,
sb_version=sb_version,
env_kwargs=env_kwargs,
env_name=env_name,
algo_name=algo_name)
log_file = os.path.join(log_dir,
str(rank)) if log_dir is not None else None
info_keywords = ()
env = Monitor(env, log_file, info_keywords=info_keywords)
env.seed(seed + rank)
return env
from stable_baselines.bench import Monitor
from model import BnnPolicy
NUM_TIMESTEPS = 5_000_000
EVAL_FREQ = 100_000
EVAL_EPISODES = 100
NUM_TRIALS = 10
for n in range(4, NUM_TRIALS + 1):
LOGDIR = f"exp/expert/ppo-bnn-mujoco/{n}"
logger.configure(folder=LOGDIR)
env = gym.make("SlimeVolley-v0")
env = Monitor(env, LOGDIR, allow_early_resets=True)
env.seed(n)
model = PPO1(BnnPolicy,
env,
timesteps_per_actorbatch=4096,
clip_param=0.2,
entcoeff=0.0,
optim_epochs=10,
optim_stepsize=3e-4,
optim_batchsize=64,
gamma=0.99,
lam=0.95,
schedule='linear',
verbose=2)
eval_callback = EvalCallback(env,
def _init():
env = gym.make(env_id)
env = Monitor(env, filename=path)
env.seed(seed + rank)
return env
def _init():
env = make_atari(env_id)
env = Monitor(env, log_dir + str(rank), allow_early_resets=True)
env.seed(seed + rank)
return env
def _init():
env = gym.make(env_id, n_obstacles=1, avoidance_method='Heuristic')
env = Monitor(env, log_dir)
env.seed(seed + rank)
return env
def _init():
env = gym.make(env_id)
env = Monitor(env, log_dir)
env.seed(seed)
return env
