def run(env_id, seed, noise_type, layer_norm, evaluation, **kwargs):
# Configure things.
rank = MPI.COMM_WORLD.Get_rank()
if rank != 0:
logger.set_level(logger.DISABLED)
# Create envs.
env = gym.make(env_id)
logger.debug("Env info")
logger.debug(env.__doc__)
logger.debug("-" * 20)
gym.logger.setLevel(logging.WARN)
if evaluation and rank == 0:
if kwargs['eval_env_id']:
eval_env_id = kwargs['eval_env_id']
else:
eval_env_id = env_id
eval_env = gym.make(eval_env_id)
# del eval_env_id from kwargs
del kwargs['eval_env_id']
else:
eval_env = None
# Parse noise_type
action_noise = None
param_noise = None
nb_actions = env.action_space.shape[-1]
for current_noise_type in noise_type.split(','):
current_noise_type = current_noise_type.strip()
if current_noise_type == 'none':
pass
elif 'adaptive-param' in current_noise_type:
_, stddev = current_noise_type.split('_')
param_noise = AdaptiveParamNoiseSpec(
initial_stddev=float(stddev),
desired_action_stddev=float(stddev))
elif 'normal' in current_noise_type:
_, stddev = current_noise_type.split('_')
action_noise = NormalActionNoise(mu=np.zeros(nb_actions),
sigma=float(stddev) *
np.ones(nb_actions))
elif 'ou' in current_noise_type:
_, stddev = current_noise_type.split('_')
action_noise = OrnsteinUhlenbeckActionNoise(
mu=np.zeros(nb_actions),
sigma=float(stddev) * np.ones(nb_actions))
elif 'epsnorm' in current_noise_type:
_, stddev, epsilon = current_noise_type.split('_')
action_noise = EpsilonNormalActionNoise(mu=np.zeros(nb_actions),
sigma=float(stddev) *
np.ones(nb_actions),
epsilon=float(epsilon))
else:
raise RuntimeError(
'unknown noise type "{}"'.format(current_noise_type))
# Configure components.
memory = Memory(limit=int(1e6),
action_shape=env.action_space.shape,
observation_shape=env.observation_space.shape)
critic = Critic(layer_norm=layer_norm)
actor = Actor(nb_actions, layer_norm=layer_norm)
# Seed everything to make things reproducible.
seed = seed + 1000000 * rank
tf.reset_default_graph()
# importing the current skill configs
if kwargs['look_ahead'] and kwargs['skillset']:
skillset_file = __import__("HER.skills.%s" % kwargs['skillset'],
fromlist=[''])
my_skill_set = SkillSet(skillset_file.skillset)
else:
my_skill_set = None
set_global_seeds(seed)
env.seed(seed)
if eval_env is not None:
eval_env.seed(seed)
# Disable logging for rank != 0 to avoid noise.
if rank == 0:
logger.info('rank {}: seed={}, logdir={}'.format(
rank, seed, logger.get_dir()))
start_time = time.time()
training.train(env=env,
eval_env=eval_env,
param_noise=param_noise,
action_noise=action_noise,
actor=actor,
critic=critic,
memory=memory,
my_skill_set=my_skill_set,
**kwargs)
env.close()
if eval_env is not None:
eval_env.close()
if rank == 0:
logger.info('total runtime: {}s'.format(time.time() - start_time))
def run(env_id, seed, noise_type, layer_norm, evaluation, **kwargs):
# Configure things.
rank = MPI.COMM_WORLD.Get_rank()
if rank != 0:
logger.set_level(logger.DISABLED)
# Create envs.
env = gym.make(env_id)
env = bench.Monitor(
env,
logger.get_dir() and os.path.join(logger.get_dir(), str(rank)))
gym.logger.setLevel(logging.WARN)
if evaluation and rank == 0:
eval_env = gym.make(env_id)
eval_env = bench.Monitor(eval_env,
os.path.join(logger.get_dir(), 'gym_eval'))
#env = bench.Monitor(env, None)
else:
eval_env = None
# Parse noise_type
action_noise = None
param_noise = None
nb_actions = env.action_space.shape[-1]
for current_noise_type in noise_type.split(','):
current_noise_type = current_noise_type.strip()
if current_noise_type == 'none':
pass
elif 'adaptive-param' in current_noise_type:
_, stddev = current_noise_type.split('_')
param_noise = AdaptiveParamNoiseSpec(
initial_stddev=float(stddev),
desired_action_stddev=float(stddev))
elif 'normal' in current_noise_type:
_, stddev = current_noise_type.split('_')
action_noise = NormalActionNoise(mu=np.zeros(nb_actions),
sigma=float(stddev) *
np.ones(nb_actions))
elif 'ou' in current_noise_type:
_, stddev = current_noise_type.split('_')
action_noise = OrnsteinUhlenbeckActionNoise(
mu=np.zeros(nb_actions),
sigma=float(stddev) * np.ones(nb_actions))
else:
raise RuntimeError(
'unknown noise type "{}"'.format(current_noise_type))
# Configure components.
memory = Memory(limit=int(1e6),
action_shape=env.action_space.shape,
observation_shape=env.observation_space.shape)
critic = Critic(layer_norm=layer_norm)
actor = Actor(nb_actions, layer_norm=layer_norm)
# Seed everything to make things reproducible.
seed = seed + 1000000 * rank
logger.info('rank {}: seed={}, logdir={}'.format(rank, seed,
logger.get_dir()))
tf.reset_default_graph()
set_global_seeds(seed)
env.seed(seed)
if eval_env is not None:
eval_env.seed(seed)
# Disable logging for rank != 0 to avoid noise.
if rank == 0:
start_time = time.time()
training.train(env=env,
eval_env=eval_env,
param_noise=param_noise,
action_noise=action_noise,
actor=actor,
critic=critic,
memory=memory,
**kwargs)
env.close()
if eval_env is not None:
eval_env.close()
if rank == 0:
logger.info('total runtime: {}s'.format(time.time() - start_time))
def run(env_id, seed, noise_type, layer_norm, evaluation, **kwargs):
# Configure things.
rank = MPI.COMM_WORLD.Get_rank()
if rank != 0:
logger.set_level(logger.DISABLED)
# Create envs.
env = gym.make(env_id)
# print(env.action_space.shape)
logger.info("Env info")
logger.info(env.__doc__)
logger.info("-" * 20)
gym.logger.setLevel(logging.WARN)
if evaluation and rank == 0:
if kwargs['eval_env_id']:
eval_env_id = kwargs['eval_env_id']
else:
eval_env_id = env_id
eval_env = gym.make(eval_env_id)
# del eval_env_id from kwargs
del kwargs['eval_env_id']
else:
eval_env = None
# Parse noise_type
action_noise = None
param_noise = None
tf.reset_default_graph()
## this is a HACK
if kwargs['skillset']:
# import HER.skills.set2 as skillset_file
skillset_file = __import__("HER.skills.%s" % kwargs['skillset'],
fromlist=[''])
my_skill_set = SkillSet(skillset_file.skillset)
nb_actions = my_skill_set.params + my_skill_set.len
else:
nb_actions = env.action_space.shape[-1]
for current_noise_type in noise_type.split(','):
current_noise_type = current_noise_type.strip()
if current_noise_type == 'none':
pass
elif 'adaptive-param' in current_noise_type:
_, stddev = current_noise_type.split('_')
param_noise = AdaptiveParamNoiseSpec(
initial_stddev=float(stddev),
desired_action_stddev=float(stddev))
elif 'normal' in current_noise_type:
_, stddev = current_noise_type.split('_')
action_noise = NormalActionNoise(mu=np.zeros(nb_actions),
sigma=float(stddev) *
np.ones(nb_actions))
elif 'ou' in current_noise_type:
_, stddev = current_noise_type.split('_')
action_noise = OrnsteinUhlenbeckActionNoise(
mu=np.zeros(nb_actions),
sigma=float(stddev) * np.ones(nb_actions))
elif 'epsnorm' in current_noise_type:
_, stddev, epsilon = current_noise_type.split('_')
action_noise = EpsilonNormalActionNoise(mu=np.zeros(nb_actions),
sigma=float(stddev) *
np.ones(nb_actions),
epsilon=float(epsilon))
elif 'pepsnorm' in current_noise_type:
_, stddev, epsilon = current_noise_type.split('_')
action_noise = EpsilonNormalParameterizedActionNoise(
mu=np.zeros(my_skill_set.num_params),
sigma=float(stddev) * np.ones(my_skill_set.num_params),
epsilon=float(epsilon),
discrete_actions_dim=my_skill_set.len)
else:
raise RuntimeError(
'unknown noise type "{}"'.format(current_noise_type))
# Configure components.
memory = Memory(limit=int(1e6),
action_shape=(nb_actions, ),
observation_shape=env.observation_space.shape)
if kwargs['newarch']:
critic = Critic(layer_norm=layer_norm, hidden_unit_list=[400, 300])
elif kwargs['newcritic']:
critic = NewCritic(layer_norm=layer_norm)
else:
critic = Critic(layer_norm=layer_norm)
if kwargs['skillset'] is None:
if kwargs['newarch']:
actor = Actor(discrete_action_size=env.env.discrete_action_size,
cts_action_size=nb_actions -
env.env.discrete_action_size,
layer_norm=layer_norm,
hidden_unit_list=[400, 300])
else:
actor = Actor(discrete_action_size=env.env.discrete_action_size,
cts_action_size=nb_actions -
env.env.discrete_action_size,
layer_norm=layer_norm)
my_skill_set = None
else:
# pass
# get the skillset and make actor accordingly
if kwargs['newarch']:
actor = Actor(discrete_action_size=my_skill_set.len,
cts_action_size=nb_actions - my_skill_set.len,
layer_norm=layer_norm,
hidden_unit_list=[400, 300])
else:
actor = Actor(discrete_action_size=my_skill_set.len,
cts_action_size=nb_actions - my_skill_set.len,
layer_norm=layer_norm)
# Seed everything to make things reproducible.
seed = seed + 1000000 * rank
logger.info('rank {}: seed={}, logdir={}'.format(rank, seed,
logger.get_dir()))
set_global_seeds(seed)
env.seed(seed)
if eval_env is not None:
eval_env.seed(seed)
# Disable logging for rank != 0 to avoid noise.
if rank == 0:
start_time = time.time()
training.train(env=env,
eval_env=eval_env,
param_noise=param_noise,
action_noise=action_noise,
actor=actor,
critic=critic,
memory=memory,
my_skill_set=my_skill_set,
**kwargs)
env.close()
if eval_env is not None:
eval_env.close()
if rank == 0:
logger.info('total runtime: {}s'.format(time.time() - start_time))
boolean_flag(parser, 'look-ahead', default=True)
parser.add_argument('--exploration-final-eps', type=float, default=0.001)
parser.add_argument('--num-samples', type=int, default=5)
boolean_flag(parser, 'select-action', default=True)
args = parser.parse_args()
# we don't directly specify timesteps for this script, so make sure that if we do specify them
# they agree with the other parameters
if args.num_timesteps is not None:
assert (args.num_timesteps == args.nb_epochs * args.nb_epoch_cycles *
args.nb_rollout_steps)
dict_args = vars(args)
del dict_args['num_timesteps']
return dict_args
if __name__ == '__main__':
args = parse_args()
if MPI.COMM_WORLD.Get_rank() == 0:
logger.configure(dir=args["log_dir"])
logger.set_level(logger.DEBUG)
logger.info(str(args))
# Run actual script.
try:
run(**args)
except KeyboardInterrupt:
print("Exiting!")
def run(env_id, seed, noise_type, layer_norm, evaluation, memory_size, factor,
**kwargs):
# Configure things.
rank = 0
if rank != 0:
logger.set_level(logger.DISABLED)
dologging = kwargs["dologging"]
# Create envs.
env = gym.make(env_id)
gym.logger.setLevel(logging.WARN)
if evaluation and rank == 0:
eval_env = gym.make(env_id)
else:
eval_env = None
# Parse noise_type
action_noise = None
param_noise = None
nb_actions = env.action_space.shape[-1]
for current_noise_type in noise_type.split(','):
current_noise_type = current_noise_type.strip()
if current_noise_type == 'none':
pass
elif 'adaptive-param' in current_noise_type:
_, stddev = current_noise_type.split('_')
param_noise = AdaptiveParamNoiseSpec(
initial_stddev=float(stddev),
desired_action_stddev=float(stddev))
elif 'normal' in current_noise_type:
_, stddev = current_noise_type.split('_')
action_noise = NormalActionNoise(mu=np.zeros(nb_actions),
sigma=float(stddev) *
np.ones(nb_actions))
elif 'ou' in current_noise_type:
_, stddev = current_noise_type.split('_')
action_noise = OrnsteinUhlenbeckActionNoise(
mu=np.zeros(nb_actions),
sigma=float(stddev) * np.ones(nb_actions))
else:
raise RuntimeError(
'unknown noise type "{}"'.format(current_noise_type))
# Configure components.
single_train = False
ospace = env.observation_space
has_image = (not hasattr(ospace, 'shape')) or (not ospace.shape)
if has_image:
assert isinstance(env.observation_space, gym.spaces.Tuple)
env.observation_space.shape = [
x.shape for x in env.observation_space.spaces
]
#eval_env.observation_space.shape = [x.shape for x in eval_env.observation_space.spaces]
if rank == 0 or not single_train:
memory = Memory(limit=memory_size,
action_shape=env.action_space.shape,
observation_shape=env.observation_space.shape)
else:
memory = None
if has_image:
ignore = False
if ignore:
critic = IgnoreDepthCritic(layer_norm=layer_norm)
actor = IgnoreDepthActor(nb_actions, layer_norm=layer_norm)
else:
critic = DepthCritic(layer_norm=layer_norm)
if factor:
actor = FactoredDepthActor(nb_actions, layer_norm=layer_norm)
else:
actor = DepthActor(nb_actions, layer_norm=layer_norm)
else:
critic = Critic(layer_norm=layer_norm)
actor = Actor(nb_actions, layer_norm=layer_norm)
# Seed everything to make things reproducible.
seed = seed + 1000000 * rank
logger.info('rank {}: seed={}, logdir={}'.format(rank, seed,
logger.get_dir()))
tf.reset_default_graph()
set_global_seeds(seed)
env.seed(seed)
if eval_env is not None:
eval_env.seed(6)
# Disable logging for rank != 0 to avoid noise.
if rank == 0:
start_time = time.time()
testing.test(env=env,
eval_env=eval_env,
param_noise=param_noise,
action_noise=action_noise,
actor=actor,
critic=critic,
memory=memory,
**kwargs)
env.close()
if eval_env is not None:
eval_env.close()
if rank == 0:
logger.info('total runtime: {}s'.format(time.time() - start_time))
def run(env_id, seed, noise_type, layer_norm, evaluation, **kwargs):
# Configure things.
rank = 0
if rank != 0:
logger.set_level(logger.DISABLED)
dologging = kwargs["dologging"]
# Create envs.
env = gym.make(env_id)
gym.logger.setLevel(logging.WARN)
if evaluation and rank == 0:
eval_env = gym.make(env_id)
else:
eval_env = None
tf.reset_default_graph()
if kwargs['skillset']:
skillset_file = __import__("HER.skills.%s" % kwargs['skillset'],
fromlist=[''])
my_skill_set = SkillSet(skillset_file.skillset)
nb_actions = my_skill_set.params + my_skill_set.len
else:
nb_actions = env.action_space.shape[-1]
# Configure components.
memory = Memory(limit=int(1e6),
action_shape=env.action_space.shape,
observation_shape=env.observation_space.shape)
critic = Critic(layer_norm=layer_norm)
if kwargs['skillset'] is None:
actor = Actor(discrete_action_size=env.env.discrete_action_size,
cts_action_size=nb_actions -
env.env.discrete_action_size,
layer_norm=layer_norm)
my_skill_set = None
else:
# pass
# get the skillset and make actor accordingly
actor = Actor(discrete_action_size=my_skill_set.len,
cts_action_size=nb_actions - my_skill_set.len,
layer_norm=layer_norm)
# Seed everything to make things reproducible.
seed = seed + 1000000 * rank
logger.info('rank {}: seed={}, logdir={}'.format(rank, seed,
logger.get_dir()))
set_global_seeds(seed)
env.seed(seed)
if eval_env is not None:
eval_env.seed(seed)
# Disable logging for rank != 0 to avoid noise.
if rank == 0:
start_time = time.time()
testing.test(env=env,
eval_env=eval_env,
param_noise=None,
action_noise=None,
actor=actor,
critic=critic,
memory=memory,
my_skill_set=my_skill_set,
**kwargs)
env.close()
if eval_env is not None:
eval_env.close()
if rank == 0:
logger.info('total runtime: {}s'.format(time.time() - start_time))
