def to_base_env(env,
make_env=None,
num_envs=1,
remote_envs=False,
async_remote_envs=False):
"""Wraps any env type as needed to expose the async interface."""
from ray.rllib.env.remote_vector_env import RemoteVectorEnv
if (remote_envs or async_remote_envs) and num_envs == 1:
raise ValueError(
"Remote envs only make sense to use if num_envs > 1 "
"(i.e. vectorization is enabled).")
if remote_envs and async_remote_envs:
raise ValueError("You can only specify one of remote_envs or "
"async_remote_envs.")
if not isinstance(env, BaseEnv):
if isinstance(env, MultiAgentEnv):
if remote_envs:
env = RemoteVectorEnv(make_env,
num_envs,
multiagent=True,
sync=True)
elif async_remote_envs:
env = RemoteVectorEnv(make_env,
num_envs,
multiagent=True,
sync=False)
else:
env = _MultiAgentEnvToBaseEnv(make_env=make_env,
existing_envs=[env],
num_envs=num_envs)
elif isinstance(env, ExternalEnv):
if num_envs != 1:
raise ValueError(
"ExternalEnv does not currently support num_envs > 1.")
env = _ExternalEnvToBaseEnv(env)
elif isinstance(env, VectorEnv):
env = _VectorEnvToBaseEnv(env)
else:
if remote_envs:
env = RemoteVectorEnv(make_env,
num_envs,
multiagent=False,
sync=True)
elif async_remote_envs:
env = RemoteVectorEnv(make_env,
num_envs,
multiagent=False,
sync=False)
else:
env = VectorEnv.wrap(
make_env=make_env,
existing_envs=[env],
num_envs=num_envs,
action_space=env.action_space,
observation_space=env.observation_space)
env = _VectorEnvToBaseEnv(env)
assert isinstance(env, BaseEnv), env
return env
def to_base_env(env, make_env=None, num_envs=1, remote_envs=False):
"""Wraps any env type as needed to expose the async interface."""
if remote_envs and num_envs == 1:
raise ValueError(
"Remote envs only make sense to use if num_envs > 1 "
"(i.e. vectorization is enabled).")
if not isinstance(env, BaseEnv):
if isinstance(env, MultiAgentEnv):
if remote_envs:
raise NotImplementedError(
"Remote multiagent environments are not implemented")
env = _MultiAgentEnvToBaseEnv(make_env=make_env,
existing_envs=[env],
num_envs=num_envs)
elif isinstance(env, ExternalEnv):
if num_envs != 1:
raise ValueError(
"ExternalEnv does not currently support num_envs > 1.")
env = _ExternalEnvToBaseEnv(env)
elif isinstance(env, VectorEnv):
env = _VectorEnvToBaseEnv(env)
else:
env = VectorEnv.wrap(make_env=make_env,
existing_envs=[env],
num_envs=num_envs,
remote_envs=remote_envs,
action_space=env.action_space,
observation_space=env.observation_space)
env = _VectorEnvToBaseEnv(env)
assert isinstance(env, BaseEnv)
return env
def to_base_env(
env: EnvType,
make_env: Callable[[int], EnvType] = None,
num_envs: int = 1,
remote_envs: bool = False,
remote_env_batch_wait_ms: int = 0,
policy_config: PartialTrainerConfigDict = None,
) -> "BaseEnv":
"""Wraps any env type as needed to expose the async interface."""
from ray.rllib.env.remote_vector_env import RemoteVectorEnv
if remote_envs and num_envs == 1:
raise ValueError(
"Remote envs only make sense to use if num_envs > 1 "
"(i.e. vectorization is enabled).")
if not isinstance(env, BaseEnv):
if isinstance(env, MultiAgentEnv):
if remote_envs:
env = RemoteVectorEnv(
make_env,
num_envs,
multiagent=True,
remote_env_batch_wait_ms=remote_env_batch_wait_ms)
else:
env = _MultiAgentEnvToBaseEnv(make_env=make_env,
existing_envs=[env],
num_envs=num_envs)
elif isinstance(env, ExternalEnv):
if num_envs != 1:
raise ValueError(
"External(MultiAgent)Env does not currently support "
"num_envs > 1. One way of solving this would be to "
"treat your Env as a MultiAgentEnv hosting only one "
"type of agent but with several copies.")
env = _ExternalEnvToBaseEnv(env)
elif isinstance(env, VectorEnv):
env = _VectorEnvToBaseEnv(env)
else:
if remote_envs:
env = RemoteVectorEnv(
make_env,
num_envs,
multiagent=False,
remote_env_batch_wait_ms=remote_env_batch_wait_ms,
existing_envs=[env],
)
else:
env = VectorEnv.wrap(
make_env=make_env,
existing_envs=[env],
num_envs=num_envs,
action_space=env.action_space,
observation_space=env.observation_space,
policy_config=policy_config,
)
env = _VectorEnvToBaseEnv(env)
assert isinstance(env, BaseEnv), env
return env
def to_base_env(env,
make_env=None,
num_envs=1,
remote_envs=False,
async_remote_envs=False):
"""Wraps any env type as needed to expose the async interface."""
from ray.rllib.env.remote_vector_env import RemoteVectorEnv
if (remote_envs or async_remote_envs) and num_envs == 1:
raise ValueError(
"Remote envs only make sense to use if num_envs > 1 "
"(i.e. vectorization is enabled).")
if remote_envs and async_remote_envs:
raise ValueError("You can only specify one of remote_envs or "
"async_remote_envs.")
if not isinstance(env, BaseEnv):
if isinstance(env, MultiAgentEnv):
if remote_envs:
env = RemoteVectorEnv(
make_env, num_envs, multiagent=True, sync=True)
elif async_remote_envs:
env = RemoteVectorEnv(
make_env, num_envs, multiagent=True, sync=False)
else:
env = _MultiAgentEnvToBaseEnv(
make_env=make_env,
existing_envs=[env],
num_envs=num_envs)
elif isinstance(env, ExternalEnv):
if num_envs != 1:
raise ValueError(
"ExternalEnv does not currently support num_envs > 1.")
env = _ExternalEnvToBaseEnv(env)
elif isinstance(env, VectorEnv):
env = _VectorEnvToBaseEnv(env)
else:
if remote_envs:
env = RemoteVectorEnv(
make_env, num_envs, multiagent=False, sync=True)
elif async_remote_envs:
env = RemoteVectorEnv(
make_env, num_envs, multiagent=False, sync=False)
else:
env = VectorEnv.wrap(
make_env=make_env,
existing_envs=[env],
num_envs=num_envs,
action_space=env.action_space,
observation_space=env.observation_space)
env = _VectorEnvToBaseEnv(env)
assert isinstance(env, BaseEnv), env
return env
def wrap_async(env, make_env=None, num_envs=1):
"""Wraps any env type as needed to expose the async interface."""
if not isinstance(env, AsyncVectorEnv):
if isinstance(env, MultiAgentEnv):
env = _MultiAgentEnvToAsync(
make_env=make_env, existing_envs=[env], num_envs=num_envs)
elif isinstance(env, ExternalEnv):
if num_envs != 1:
raise ValueError(
"ExternalEnv does not currently support num_envs > 1.")
env = _ExternalEnvToAsync(env)
elif isinstance(env, VectorEnv):
env = _VectorEnvToAsync(env)
else:
env = VectorEnv.wrap(
make_env=make_env, existing_envs=[env], num_envs=num_envs)
env = _VectorEnvToAsync(env)
assert isinstance(env, AsyncVectorEnv)
return env
def wrap_async(env, make_env=None, num_envs=1):
"""Wraps any env type as needed to expose the async interface."""
if not isinstance(env, AsyncVectorEnv):
if isinstance(env, MultiAgentEnv):
env = _MultiAgentEnvToAsync(
make_env=make_env, existing_envs=[env], num_envs=num_envs)
elif isinstance(env, ServingEnv):
if num_envs != 1:
raise ValueError(
"ServingEnv does not currently support num_envs > 1.")
env = _ServingEnvToAsync(env)
elif isinstance(env, VectorEnv):
env = _VectorEnvToAsync(env)
else:
env = VectorEnv.wrap(
make_env=make_env, existing_envs=[env], num_envs=num_envs)
env = _VectorEnvToAsync(env)
assert isinstance(env, AsyncVectorEnv)
return env
def testNestedTupleVector(self):
self.doTestNestedTuple(
lambda _: VectorEnv.wrap(lambda i: NestedTupleEnv()))
def to_base_env(
env: EnvType,
make_env: Callable[[int], EnvType] = None,
num_envs: int = 1,
remote_envs: bool = False,
remote_env_batch_wait_ms: int = 0,
policy_config: Optional[PartialTrainerConfigDict] = None,
) -> "BaseEnv":
"""Converts an RLlib-supported env into a BaseEnv object.
Supported types for the given `env` arg are gym.Env, BaseEnv,
VectorEnv, MultiAgentEnv, or ExternalEnv.
The resulting BaseEnv is always vectorized (contains n
sub-environments) for batched forward passes, where n may also be 1.
BaseEnv also supports async execution via the `poll` and `send_actions`
methods and thus supports external simulators.
TODO: Support gym3 environments, which are already vectorized.
Args:
env: An already existing environment of any supported env type
to convert/wrap into a BaseEnv. Supported types are gym.Env,
BaseEnv, VectorEnv, MultiAgentEnv, ExternalEnv, or
ExternalMultiAgentEnv.
make_env: A callable taking an int as input (which indicates the
number of individual sub-environments within the final
vectorized BaseEnv) and returning one individual
sub-environment.
num_envs: The number of sub-environments to create in the
resulting (vectorized) BaseEnv. The already existing `env`
will be one of the `num_envs`.
remote_envs: Whether each sub-env should be a @ray.remote actor.
You can set this behavior in your config via the
`remote_worker_envs=True` option.
remote_env_batch_wait_ms: The wait time (in ms) to poll remote
sub-environments for, if applicable. Only used if
`remote_envs` is True.
policy_config: Optional policy config dict.
Returns:
The resulting BaseEnv object.
"""
from ray.rllib.env.remote_vector_env import RemoteVectorEnv
if remote_envs and num_envs == 1:
raise ValueError(
"Remote envs only make sense to use if num_envs > 1 "
"(i.e. vectorization is enabled).")
# Given `env` is already a BaseEnv -> Return as is.
if isinstance(env, BaseEnv):
return env
# `env` is not a BaseEnv yet -> Need to convert/vectorize.
# MultiAgentEnv (which is a gym.Env).
if isinstance(env, MultiAgentEnv):
# Sub-environments are ray.remote actors:
if remote_envs:
env = RemoteVectorEnv(
make_env,
num_envs,
multiagent=True,
remote_env_batch_wait_ms=remote_env_batch_wait_ms)
# Sub-environments are not ray.remote actors.
else:
env = _MultiAgentEnvToBaseEnv(
make_env=make_env, existing_envs=[env], num_envs=num_envs)
# ExternalEnv.
elif isinstance(env, ExternalEnv):
if num_envs != 1:
raise ValueError(
"External(MultiAgent)Env does not currently support "
"num_envs > 1. One way of solving this would be to "
"treat your Env as a MultiAgentEnv hosting only one "
"type of agent but with several copies.")
env = _ExternalEnvToBaseEnv(env)
# VectorEnv.
# Note that all BaseEnvs are also vectorized, but the user may want to
# define custom vectorization logic and thus implement a custom
# VectorEnv class.
elif isinstance(env, VectorEnv):
env = _VectorEnvToBaseEnv(env)
# Anything else: This usually implies that env is a gym.Env object.
else:
# Sub-environments are ray.remote actors:
if remote_envs:
# Determine, whether the already existing sub-env (could
# be a ray.actor) is multi-agent or not.
multiagent = ray.get(env._is_multi_agent.remote()) if \
hasattr(env, "_is_multi_agent") else False
env = RemoteVectorEnv(
make_env,
num_envs,
multiagent=multiagent,
remote_env_batch_wait_ms=remote_env_batch_wait_ms,
existing_envs=[env],
)
# Sub-environments are not ray.remote actors.
else:
env = VectorEnv.vectorize_gym_envs(
make_env=make_env,
existing_envs=[env],
num_envs=num_envs,
action_space=env.action_space,
observation_space=env.observation_space,
)
env = _VectorEnvToBaseEnv(env)
# Make sure conversion went well.
assert isinstance(env, BaseEnv), env
return env
def test_nested_tuple_vector(self):
self.do_test_nested_tuple(
lambda _: VectorEnv.wrap(lambda i: NestedTupleEnv()))
def test_nested_dict_vector(self):
self.do_test_nested_dict(
lambda _: VectorEnv.wrap(lambda i: NestedDictEnv()))
def convert_to_base_env(
env: EnvType,
make_env: Callable[[int], EnvType] = None,
num_envs: int = 1,
remote_envs: bool = False,
remote_env_batch_wait_ms: int = 0,
) -> "BaseEnv":
"""Converts an RLlib-supported env into a BaseEnv object.
Supported types for the `env` arg are gym.Env, BaseEnv,
VectorEnv, MultiAgentEnv, ExternalEnv, or ExternalMultiAgentEnv.
The resulting BaseEnv is always vectorized (contains n
sub-environments) to support batched forward passes, where n may also
be 1. BaseEnv also supports async execution via the `poll` and
`send_actions` methods and thus supports external simulators.
TODO: Support gym3 environments, which are already vectorized.
Args:
env: An already existing environment of any supported env type
to convert/wrap into a BaseEnv. Supported types are gym.Env,
BaseEnv, VectorEnv, MultiAgentEnv, ExternalEnv, and
ExternalMultiAgentEnv.
make_env: A callable taking an int as input (which indicates the
number of individual sub-environments within the final
vectorized BaseEnv) and returning one individual
sub-environment.
num_envs: The number of sub-environments to create in the
resulting (vectorized) BaseEnv. The already existing `env`
will be one of the `num_envs`.
remote_envs: Whether each sub-env should be a @ray.remote actor.
You can set this behavior in your config via the
`remote_worker_envs=True` option.
remote_env_batch_wait_ms: The wait time (in ms) to poll remote
sub-environments for, if applicable. Only used if
`remote_envs` is True.
Returns:
The resulting BaseEnv object.
"""
from ray.rllib.env.remote_base_env import RemoteBaseEnv
from ray.rllib.env.external_env import ExternalEnv
from ray.rllib.env.multi_agent_env import MultiAgentEnv
from ray.rllib.env.vector_env import VectorEnv, VectorEnvWrapper
if remote_envs and num_envs == 1:
raise ValueError("Remote envs only make sense to use if num_envs > 1 "
"(i.e. vectorization is enabled).")
# Given `env` is already a BaseEnv -> Return as is.
if isinstance(env, (BaseEnv, MultiAgentEnv, VectorEnv, ExternalEnv)):
return env.to_base_env(
make_env=make_env,
num_envs=num_envs,
remote_envs=remote_envs,
remote_env_batch_wait_ms=remote_env_batch_wait_ms,
)
# `env` is not a BaseEnv yet -> Need to convert/vectorize.
else:
# Sub-environments are ray.remote actors:
if remote_envs:
# Determine, whether the already existing sub-env (could
# be a ray.actor) is multi-agent or not.
multiagent = ray.get(env._is_multi_agent.remote()) if \
hasattr(env, "_is_multi_agent") else False
env = RemoteBaseEnv(
make_env,
num_envs,
multiagent=multiagent,
remote_env_batch_wait_ms=remote_env_batch_wait_ms,
existing_envs=[env],
)
# Sub-environments are not ray.remote actors.
else:
# Convert gym.Env to VectorEnv ...
env = VectorEnv.vectorize_gym_envs(
make_env=make_env,
existing_envs=[env],
num_envs=num_envs,
action_space=env.action_space,
observation_space=env.observation_space,
)
# ... then the resulting VectorEnv to a BaseEnv.
env = VectorEnvWrapper(env)
# Make sure conversion went well.
assert isinstance(env, BaseEnv), env
return env
def test_vector_step(self):
env = VectorEnv.wrap(lambda _: MockEnvDictSubclass(), num_envs=3)
env.vector_step([0] * 3)
def test_nested_tuple_vector(self):
self.do_test_nested_tuple(
lambda _: VectorEnv.vectorize_gym_envs(lambda i: NestedTupleEnv())
)
def test_nested_dict_vector(self):
self.do_test_nested_dict(
lambda _: VectorEnv.vectorize_gym_envs(lambda i: NestedDictEnv())
)
def testNestedDictVector(self):
self.doTestNestedDict(
lambda _: VectorEnv.wrap(lambda i: NestedDictEnv()))
def testNestedDictVector(self):
self.doTestNestedDict(
lambda _: VectorEnv.wrap(lambda i: NestedDictEnv()))
def testNestedTupleVector(self):
self.doTestNestedTuple(
lambda _: VectorEnv.wrap(lambda i: NestedTupleEnv()))
def test_vector_step(self):
env = VectorEnv.vectorize_gym_envs(
make_env=lambda _: MockEnvDictSubclass(), num_envs=3)
env.vector_step([0] * 3)
