class PolicyEvaluator(EvaluatorInterface):
"""Common ``PolicyEvaluator`` implementation that wraps a ``PolicyGraph``.
This class wraps a policy graph instance and an environment class to
collect experiences from the environment. You can create many replicas of
this class as Ray actors to scale RL training.
This class supports vectorized and multi-agent policy evaluation (e.g.,
VectorEnv, MultiAgentEnv, etc.)
Examples:
>>> # Create a policy evaluator and using it to collect experiences.
>>> evaluator = PolicyEvaluator(
... env_creator=lambda _: gym.make("CartPole-v0"),
... policy_graph=PGPolicyGraph)
>>> print(evaluator.sample())
SampleBatch({
"obs": [[...]], "actions": [[...]], "rewards": [[...]],
"dones": [[...]], "new_obs": [[...]]})
>>> # Creating policy evaluators using optimizer_cls.make().
>>> optimizer = SyncSamplesOptimizer.make(
... evaluator_cls=PolicyEvaluator,
... evaluator_args={
... "env_creator": lambda _: gym.make("CartPole-v0"),
... "policy_graph": PGPolicyGraph,
... },
... num_workers=10)
>>> for _ in range(10): optimizer.step()
>>> # Creating a multi-agent policy evaluator
>>> evaluator = PolicyEvaluator(
... env_creator=lambda _: MultiAgentTrafficGrid(num_cars=25),
... policy_graphs={
... # Use an ensemble of two policies for car agents
... "car_policy1":
... (PGPolicyGraph, Box(...), Discrete(...), {"gamma": 0.99}),
... "car_policy2":
... (PGPolicyGraph, Box(...), Discrete(...), {"gamma": 0.95}),
... # Use a single shared policy for all traffic lights
... "traffic_light_policy":
... (PGPolicyGraph, Box(...), Discrete(...), {}),
... },
... policy_mapping_fn=lambda agent_id:
... random.choice(["car_policy1", "car_policy2"])
... if agent_id.startswith("car_") else "traffic_light_policy")
>>> print(evaluator.sample())
MultiAgentBatch({
"car_policy1": SampleBatch(...),
"car_policy2": SampleBatch(...),
"traffic_light_policy": SampleBatch(...)})
"""
@DeveloperAPI
@classmethod
def as_remote(cls, num_cpus=None, num_gpus=None, resources=None):
return ray.remote(num_cpus=num_cpus,
num_gpus=num_gpus,
resources=resources)(cls)
@DeveloperAPI
def __init__(self,
env_creator,
policy_graph,
policy_mapping_fn=None,
policies_to_train=None,
tf_session_creator=None,
batch_steps=100,
batch_mode="truncate_episodes",
episode_horizon=None,
preprocessor_pref="deepmind",
sample_async=False,
compress_observations=False,
num_envs=1,
observation_filter="NoFilter",
clip_rewards=None,
clip_actions=True,
env_config=None,
model_config=None,
policy_config=None,
worker_index=0,
monitor_path=None,
log_dir=None,
log_level=None,
callbacks=None,
input_creator=lambda ioctx: ioctx.default_sampler_input(),
input_evaluation=frozenset([]),
output_creator=lambda ioctx: NoopOutput(),
remote_worker_envs=False,
async_remote_worker_envs=False):
"""Initialize a policy evaluator.
Arguments:
env_creator (func): Function that returns a gym.Env given an
EnvContext wrapped configuration.
policy_graph (class|dict): Either a class implementing
PolicyGraph, or a dictionary of policy id strings to
(PolicyGraph, obs_space, action_space, config) tuples. If a
dict is specified, then we are in multi-agent mode and a
policy_mapping_fn should also be set.
policy_mapping_fn (func): A function that maps agent ids to
policy ids in multi-agent mode. This function will be called
each time a new agent appears in an episode, to bind that agent
to a policy for the duration of the episode.
policies_to_train (list): Optional whitelist of policies to train,
or None for all policies.
tf_session_creator (func): A function that returns a TF session.
This is optional and only useful with TFPolicyGraph.
batch_steps (int): The target number of env transitions to include
in each sample batch returned from this evaluator.
batch_mode (str): One of the following batch modes:
"truncate_episodes": Each call to sample() will return a batch
of at most `batch_steps * num_envs` in size. The batch will
be exactly `batch_steps * num_envs` in size if
postprocessing does not change batch sizes. Episodes may be
truncated in order to meet this size requirement.
"complete_episodes": Each call to sample() will return a batch
of at least `batch_steps * num_envs` in size. Episodes will
not be truncated, but multiple episodes may be packed
within one batch to meet the batch size. Note that when
`num_envs > 1`, episode steps will be buffered until the
episode completes, and hence batches may contain
significant amounts of off-policy data.
episode_horizon (int): Whether to stop episodes at this horizon.
preprocessor_pref (str): Whether to prefer RLlib preprocessors
("rllib") or deepmind ("deepmind") when applicable.
sample_async (bool): Whether to compute samples asynchronously in
the background, which improves throughput but can cause samples
to be slightly off-policy.
compress_observations (bool): If true, compress the observations.
They can be decompressed with rllib/utils/compression.
num_envs (int): If more than one, will create multiple envs
and vectorize the computation of actions. This has no effect if
if the env already implements VectorEnv.
observation_filter (str): Name of observation filter to use.
clip_rewards (bool): Whether to clip rewards to [-1, 1] prior to
experience postprocessing. Setting to None means clip for Atari
only.
clip_actions (bool): Whether to clip action values to the range
specified by the policy action space.
env_config (dict): Config to pass to the env creator.
model_config (dict): Config to use when creating the policy model.
policy_config (dict): Config to pass to the policy. In the
multi-agent case, this config will be merged with the
per-policy configs specified by `policy_graph`.
worker_index (int): For remote evaluators, this should be set to a
non-zero and unique value. This index is passed to created envs
through EnvContext so that envs can be configured per worker.
monitor_path (str): Write out episode stats and videos to this
directory if specified.
log_dir (str): Directory where logs can be placed.
log_level (str): Set the root log level on creation.
callbacks (dict): Dict of custom debug callbacks.
input_creator (func): Function that returns an InputReader object
for loading previous generated experiences.
input_evaluation (list): How to evaluate the policy performance.
This only makes sense to set when the input is reading offline
data. The possible values include:
- "is": the step-wise importance sampling estimator.
- "wis": the weighted step-wise is estimator.
- "simulation": run the environment in the background, but
use this data for evaluation only and never for learning.
output_creator (func): Function that returns an OutputWriter object
for saving generated experiences.
remote_worker_envs (bool): If using num_envs > 1, whether to create
those new envs in remote processes instead of in the current
process. This adds overheads, but can make sense if your envs
are very CPU intensive (e.g., for StarCraft).
async_remote_worker_envs (bool): Similar to remote_worker_envs,
but runs the envs asynchronously in the background.
"""
if log_level:
logging.getLogger("ray.rllib").setLevel(log_level)
if worker_index > 1:
disable_log_once_globally() # only need 1 evaluator to log
elif log_level == "DEBUG":
enable_periodic_logging()
env_context = EnvContext(env_config or {}, worker_index)
policy_config = policy_config or {}
self.policy_config = policy_config
self.callbacks = callbacks or {}
self.worker_index = worker_index
model_config = model_config or {}
policy_mapping_fn = (policy_mapping_fn
or (lambda agent_id: DEFAULT_POLICY_ID))
if not callable(policy_mapping_fn):
raise ValueError(
"Policy mapping function not callable. If you're using Tune, "
"make sure to escape the function with tune.function() "
"to prevent it from being evaluated as an expression.")
self.env_creator = env_creator
self.sample_batch_size = batch_steps * num_envs
self.batch_mode = batch_mode
self.compress_observations = compress_observations
self.preprocessing_enabled = True
self.env = _validate_env(env_creator(env_context))
if isinstance(self.env, MultiAgentEnv) or \
isinstance(self.env, BaseEnv):
def wrap(env):
return env # we can't auto-wrap these env types
elif is_atari(self.env) and \
not model_config.get("custom_preprocessor") and \
preprocessor_pref == "deepmind":
# Deepmind wrappers already handle all preprocessing
self.preprocessing_enabled = False
if clip_rewards is None:
clip_rewards = True
def wrap(env):
env = wrap_deepmind(env,
dim=model_config.get("dim"),
framestack=model_config.get("framestack"))
if monitor_path:
env = _monitor(env, monitor_path)
return env
else:
def wrap(env):
if monitor_path:
env = _monitor(env, monitor_path)
return env
self.env = wrap(self.env)
def make_env(vector_index):
return wrap(
env_creator(
env_context.copy_with_overrides(
vector_index=vector_index, remote=remote_worker_envs)))
self.tf_sess = None
policy_dict = _validate_and_canonicalize(policy_graph, self.env)
self.policies_to_train = policies_to_train or list(policy_dict.keys())
if _has_tensorflow_graph(policy_dict):
if (ray.is_initialized()
and ray.worker._mode() != ray.worker.LOCAL_MODE
and not ray.get_gpu_ids()):
logger.info("Creating policy evaluation worker {}".format(
worker_index) +
" on CPU (please ignore any CUDA init errors)")
with tf.Graph().as_default():
if tf_session_creator:
self.tf_sess = tf_session_creator()
else:
self.tf_sess = tf.Session(config=tf.ConfigProto(
gpu_options=tf.GPUOptions(allow_growth=True)))
with self.tf_sess.as_default():
self.policy_map, self.preprocessors = \
self._build_policy_map(policy_dict, policy_config)
else:
self.policy_map, self.preprocessors = self._build_policy_map(
policy_dict, policy_config)
self.multiagent = set(self.policy_map.keys()) != {DEFAULT_POLICY_ID}
if self.multiagent:
if not (isinstance(self.env, MultiAgentEnv)
or isinstance(self.env, BaseEnv)):
raise ValueError(
"Have multiple policy graphs {}, but the env ".format(
self.policy_map) +
"{} is not a subclass of MultiAgentEnv?".format(self.env))
self.filters = {
policy_id: get_filter(observation_filter,
policy.observation_space.shape)
for (policy_id, policy) in self.policy_map.items()
}
if self.worker_index == 0:
logger.info("Built filter map: {}".format(self.filters))
# Always use vector env for consistency even if num_envs = 1
self.async_env = BaseEnv.to_base_env(
self.env,
make_env=make_env,
num_envs=num_envs,
remote_envs=remote_worker_envs,
async_remote_envs=async_remote_worker_envs)
self.num_envs = num_envs
if self.batch_mode == "truncate_episodes":
unroll_length = batch_steps
pack_episodes = True
elif self.batch_mode == "complete_episodes":
unroll_length = float("inf") # never cut episodes
pack_episodes = False # sampler will return 1 episode per poll
else:
raise ValueError("Unsupported batch mode: {}".format(
self.batch_mode))
self.io_context = IOContext(log_dir, policy_config, worker_index, self)
self.reward_estimators = []
for method in input_evaluation:
if method == "simulation":
logger.warning(
"Requested 'simulation' input evaluation method: "
"will discard all sampler outputs and keep only metrics.")
sample_async = True
elif method == "is":
ise = ImportanceSamplingEstimator.create(self.io_context)
self.reward_estimators.append(ise)
elif method == "wis":
wise = WeightedImportanceSamplingEstimator.create(
self.io_context)
self.reward_estimators.append(wise)
else:
raise ValueError(
"Unknown evaluation method: {}".format(method))
if sample_async:
self.sampler = AsyncSampler(self.async_env,
self.policy_map,
policy_mapping_fn,
self.preprocessors,
self.filters,
clip_rewards,
unroll_length,
self.callbacks,
horizon=episode_horizon,
pack=pack_episodes,
tf_sess=self.tf_sess,
clip_actions=clip_actions,
blackhole_outputs="simulation"
in input_evaluation)
self.sampler.start()
else:
self.sampler = SyncSampler(self.async_env,
self.policy_map,
policy_mapping_fn,
self.preprocessors,
self.filters,
clip_rewards,
unroll_length,
self.callbacks,
horizon=episode_horizon,
pack=pack_episodes,
tf_sess=self.tf_sess,
clip_actions=clip_actions)
self.input_reader = input_creator(self.io_context)
assert isinstance(self.input_reader, InputReader), self.input_reader
self.output_writer = output_creator(self.io_context)
assert isinstance(self.output_writer, OutputWriter), self.output_writer
logger.debug("Created evaluator with env {} ({}), policies {}".format(
self.async_env, self.env, self.policy_map))
@override(EvaluatorInterface)
def sample(self):
"""Evaluate the current policies and return a batch of experiences.
Return:
SampleBatch|MultiAgentBatch from evaluating the current policies.
"""
if log_once("sample_start"):
logger.info("Generating sample batch of size {}".format(
self.sample_batch_size))
batches = [self.input_reader.next()]
steps_so_far = batches[0].count
# In truncate_episodes mode, never pull more than 1 batch per env.
# This avoids over-running the target batch size.
if self.batch_mode == "truncate_episodes":
max_batches = self.num_envs
else:
max_batches = float("inf")
while steps_so_far < self.sample_batch_size and len(
batches) < max_batches:
batch = self.input_reader.next()
steps_so_far += batch.count
batches.append(batch)
batch = batches[0].concat_samples(batches)
if self.callbacks.get("on_sample_end"):
self.callbacks["on_sample_end"]({
"evaluator": self,
"samples": batch
})
# Always do writes prior to compression for consistency and to allow
# for better compression inside the writer.
self.output_writer.write(batch)
# Do off-policy estimation if needed
if self.reward_estimators:
for sub_batch in batch.split_by_episode():
for estimator in self.reward_estimators:
estimator.process(sub_batch)
if log_once("sample_end"):
logger.info("Completed sample batch:\n\n{}\n".format(
summarize(batch)))
if self.compress_observations:
if isinstance(batch, MultiAgentBatch):
for data in batch.policy_batches.values():
data["obs"] = [pack(o) for o in data["obs"]]
data["new_obs"] = [pack(o) for o in data["new_obs"]]
else:
batch["obs"] = [pack(o) for o in batch["obs"]]
batch["new_obs"] = [pack(o) for o in batch["new_obs"]]
return batch
@DeveloperAPI
@ray.method(num_return_vals=2)
def sample_with_count(self):
"""Same as sample() but returns the count as a separate future."""
batch = self.sample()
return batch, batch.count
@override(EvaluatorInterface)
def get_weights(self, policies=None):
if policies is None:
policies = self.policy_map.keys()
return {
pid: policy.get_weights()
for pid, policy in self.policy_map.items() if pid in policies
}
@override(EvaluatorInterface)
def set_weights(self, weights):
for pid, w in weights.items():
self.policy_map[pid].set_weights(w)
@override(EvaluatorInterface)
def compute_gradients(self, samples):
if log_once("compute_gradients"):
logger.info("Compute gradients on:\n\n{}\n".format(
summarize(samples)))
if isinstance(samples, MultiAgentBatch):
grad_out, info_out = {}, {}
if self.tf_sess is not None:
builder = TFRunBuilder(self.tf_sess, "compute_gradients")
for pid, batch in samples.policy_batches.items():
if pid not in self.policies_to_train:
continue
grad_out[pid], info_out[pid] = (
self.policy_map[pid]._build_compute_gradients(
builder, batch))
grad_out = {k: builder.get(v) for k, v in grad_out.items()}
info_out = {k: builder.get(v) for k, v in info_out.items()}
else:
for pid, batch in samples.policy_batches.items():
if pid not in self.policies_to_train:
continue
grad_out[pid], info_out[pid] = (
self.policy_map[pid].compute_gradients(batch))
else:
grad_out, info_out = (
self.policy_map[DEFAULT_POLICY_ID].compute_gradients(samples))
info_out["batch_count"] = samples.count
if log_once("grad_out"):
logger.info("Compute grad info:\n\n{}\n".format(
summarize(info_out)))
return grad_out, info_out
@override(EvaluatorInterface)
def apply_gradients(self, grads):
if log_once("apply_gradients"):
logger.info("Apply gradients:\n\n{}\n".format(summarize(grads)))
if isinstance(grads, dict):
if self.tf_sess is not None:
builder = TFRunBuilder(self.tf_sess, "apply_gradients")
outputs = {
pid:
self.policy_map[pid]._build_apply_gradients(builder, grad)
for pid, grad in grads.items()
}
return {k: builder.get(v) for k, v in outputs.items()}
else:
return {
pid: self.policy_map[pid].apply_gradients(g)
for pid, g in grads.items()
}
else:
return self.policy_map[DEFAULT_POLICY_ID].apply_gradients(grads)
@override(EvaluatorInterface)
def learn_on_batch(self, samples):
if log_once("learn_on_batch"):
logger.info(
"Training on concatenated sample batches:\n\n{}\n".format(
summarize(samples)))
if isinstance(samples, MultiAgentBatch):
info_out = {}
if self.tf_sess is not None:
builder = TFRunBuilder(self.tf_sess, "learn_on_batch")
for pid, batch in samples.policy_batches.items():
if pid not in self.policies_to_train:
continue
info_out[pid], _ = (
self.policy_map[pid]._build_learn_on_batch(
builder, batch))
info_out = {k: builder.get(v) for k, v in info_out.items()}
else:
for pid, batch in samples.policy_batches.items():
if pid not in self.policies_to_train:
continue
info_out[pid], _ = (
self.policy_map[pid].learn_on_batch(batch))
else:
info_out, _ = (
self.policy_map[DEFAULT_POLICY_ID].learn_on_batch(samples))
if log_once("learn_out"):
logger.info("Training output:\n\n{}\n".format(summarize(info_out)))
return info_out
@DeveloperAPI
def get_metrics(self):
"""Returns a list of new RolloutMetric objects from evaluation."""
out = self.sampler.get_metrics()
for m in self.reward_estimators:
out.extend(m.get_metrics())
return out
@DeveloperAPI
def foreach_env(self, func):
"""Apply the given function to each underlying env instance."""
envs = self.async_env.get_unwrapped()
if not envs:
return [func(self.async_env)]
else:
return [func(e) for e in envs]
@DeveloperAPI
def get_policy(self, policy_id=DEFAULT_POLICY_ID):
"""Return policy graph for the specified id, or None.
Arguments:
policy_id (str): id of policy graph to return.
"""
return self.policy_map.get(policy_id)
@DeveloperAPI
def for_policy(self, func, policy_id=DEFAULT_POLICY_ID):
"""Apply the given function to the specified policy graph."""
return func(self.policy_map[policy_id])
@DeveloperAPI
def foreach_policy(self, func):
"""Apply the given function to each (policy, policy_id) tuple."""
return [func(policy, pid) for pid, policy in self.policy_map.items()]
@DeveloperAPI
def foreach_trainable_policy(self, func):
"""Apply the given function to each (policy, policy_id) tuple.
This only applies func to policies in `self.policies_to_train`."""
return [
func(policy, pid) for pid, policy in self.policy_map.items()
if pid in self.policies_to_train
]
@DeveloperAPI
def sync_filters(self, new_filters):
"""Changes self's filter to given and rebases any accumulated delta.
Args:
new_filters (dict): Filters with new state to update local copy.
"""
assert all(k in new_filters for k in self.filters)
for k in self.filters:
self.filters[k].sync(new_filters[k])
@DeveloperAPI
def get_filters(self, flush_after=False):
"""Returns a snapshot of filters.
Args:
flush_after (bool): Clears the filter buffer state.
Returns:
return_filters (dict): Dict for serializable filters
"""
return_filters = {}
for k, f in self.filters.items():
return_filters[k] = f.as_serializable()
if flush_after:
f.clear_buffer()
return return_filters
@DeveloperAPI
def save(self):
filters = self.get_filters(flush_after=True)
state = {
pid: self.policy_map[pid].get_state()
for pid in self.policy_map
}
return pickle.dumps({"filters": filters, "state": state})
@DeveloperAPI
def restore(self, objs):
objs = pickle.loads(objs)
self.sync_filters(objs["filters"])
for pid, state in objs["state"].items():
self.policy_map[pid].set_state(state)
@DeveloperAPI
def set_global_vars(self, global_vars):
self.foreach_policy(lambda p, _: p.on_global_var_update(global_vars))
@DeveloperAPI
def export_policy_model(self, export_dir, policy_id=DEFAULT_POLICY_ID):
self.policy_map[policy_id].export_model(export_dir)
@DeveloperAPI
def export_policy_checkpoint(self,
export_dir,
filename_prefix="model",
policy_id=DEFAULT_POLICY_ID):
self.policy_map[policy_id].export_checkpoint(export_dir,
filename_prefix)
def _build_policy_map(self, policy_dict, policy_config):
policy_map = {}
preprocessors = {}
for name, (cls, obs_space, act_space,
conf) in sorted(policy_dict.items()):
logger.debug("Creating policy graph for {}".format(name))
merged_conf = merge_dicts(policy_config, conf)
if self.preprocessing_enabled:
preprocessor = ModelCatalog.get_preprocessor_for_space(
obs_space, merged_conf.get("model"))
preprocessors[name] = preprocessor
obs_space = preprocessor.observation_space
else:
preprocessors[name] = NoPreprocessor(obs_space)
if isinstance(obs_space, gym.spaces.Dict) or \
isinstance(obs_space, gym.spaces.Tuple):
raise ValueError(
"Found raw Tuple|Dict space as input to policy graph. "
"Please preprocess these observations with a "
"Tuple|DictFlatteningPreprocessor.")
with tf.variable_scope(name):
policy_map[name] = cls(obs_space, act_space, merged_conf)
if self.worker_index == 0:
logger.info("Built policy map: {}".format(policy_map))
logger.info("Built preprocessor map: {}".format(preprocessors))
return policy_map, preprocessors
def __del__(self):
if hasattr(self, "sampler") and isinstance(self.sampler, AsyncSampler):
self.sampler.shutdown = True
class RolloutWorker(EvaluatorInterface, ParallelIteratorWorker):
"""Common experience collection class.
This class wraps a policy instance and an environment class to
collect experiences from the environment. You can create many replicas of
this class as Ray actors to scale RL training.
This class supports vectorized and multi-agent policy evaluation (e.g.,
VectorEnv, MultiAgentEnv, etc.)
Examples:
>>> # Create a rollout worker and using it to collect experiences.
>>> worker = RolloutWorker(
... env_creator=lambda _: gym.make("CartPole-v0"),
... policy=PGTFPolicy)
>>> print(worker.sample())
SampleBatch({
"obs": [[...]], "actions": [[...]], "rewards": [[...]],
"dones": [[...]], "new_obs": [[...]]})
>>> # Creating a multi-agent rollout worker
>>> worker = RolloutWorker(
... env_creator=lambda _: MultiAgentTrafficGrid(num_cars=25),
... policies={
... # Use an ensemble of two policies for car agents
... "car_policy1":
... (PGTFPolicy, Box(...), Discrete(...), {"gamma": 0.99}),
... "car_policy2":
... (PGTFPolicy, Box(...), Discrete(...), {"gamma": 0.95}),
... # Use a single shared policy for all traffic lights
... "traffic_light_policy":
... (PGTFPolicy, Box(...), Discrete(...), {}),
... },
... policy_mapping_fn=lambda agent_id:
... random.choice(["car_policy1", "car_policy2"])
... if agent_id.startswith("car_") else "traffic_light_policy")
>>> print(worker.sample())
MultiAgentBatch({
"car_policy1": SampleBatch(...),
"car_policy2": SampleBatch(...),
"traffic_light_policy": SampleBatch(...)})
"""
@DeveloperAPI
@classmethod
def as_remote(cls,
num_cpus=None,
num_gpus=None,
memory=None,
object_store_memory=None,
resources=None):
return ray.remote(
num_cpus=num_cpus,
num_gpus=num_gpus,
memory=memory,
object_store_memory=object_store_memory,
resources=resources)(cls)
@DeveloperAPI
def __init__(self,
env_creator,
policy,
policy_mapping_fn=None,
policies_to_train=None,
tf_session_creator=None,
rollout_fragment_length=100,
batch_mode="truncate_episodes",
episode_horizon=None,
preprocessor_pref="deepmind",
sample_async=False,
compress_observations=False,
num_envs=1,
observation_fn=None,
observation_filter="NoFilter",
clip_rewards=None,
clip_actions=True,
env_config=None,
model_config=None,
policy_config=None,
worker_index=0,
num_workers=0,
monitor_path=None,
log_dir=None,
log_level=None,
callbacks=None,
input_creator=lambda ioctx: ioctx.default_sampler_input(),
input_evaluation=frozenset([]),
output_creator=lambda ioctx: NoopOutput(),
remote_worker_envs=False,
remote_env_batch_wait_ms=0,
soft_horizon=False,
no_done_at_end=False,
seed=None,
extra_python_environs=None,
fake_sampler=False):
"""Initialize a rollout worker.
Arguments:
env_creator (func): Function that returns a gym.Env given an
EnvContext wrapped configuration.
policy (class|dict): Either a class implementing
Policy, or a dictionary of policy id strings to
(Policy, obs_space, action_space, config) tuples. If a
dict is specified, then we are in multi-agent mode and a
policy_mapping_fn should also be set.
policy_mapping_fn (func): A function that maps agent ids to
policy ids in multi-agent mode. This function will be called
each time a new agent appears in an episode, to bind that agent
to a policy for the duration of the episode.
policies_to_train (list): Optional whitelist of policies to train,
or None for all policies.
tf_session_creator (func): A function that returns a TF session.
This is optional and only useful with TFPolicy.
rollout_fragment_length (int): The target number of env transitions
to include in each sample batch returned from this worker.
batch_mode (str): One of the following batch modes:
"truncate_episodes": Each call to sample() will return a batch
of at most `rollout_fragment_length * num_envs` in size.
The batch will be exactly
`rollout_fragment_length * num_envs` in size if
postprocessing does not change batch sizes. Episodes may be
truncated in order to meet this size requirement.
"complete_episodes": Each call to sample() will return a batch
of at least `rollout_fragment_length * num_envs` in size.
Episodes will not be truncated, but multiple episodes may
be packed within one batch to meet the batch size. Note
that when `num_envs > 1`, episode steps will be buffered
until the episode completes, and hence batches may contain
significant amounts of off-policy data.
episode_horizon (int): Whether to stop episodes at this horizon.
preprocessor_pref (str): Whether to prefer RLlib preprocessors
("rllib") or deepmind ("deepmind") when applicable.
sample_async (bool): Whether to compute samples asynchronously in
the background, which improves throughput but can cause samples
to be slightly off-policy.
compress_observations (bool): If true, compress the observations.
They can be decompressed with rllib/utils/compression.
num_envs (int): If more than one, will create multiple envs
and vectorize the computation of actions. This has no effect if
if the env already implements VectorEnv.
observation_fn (ObservationFunction): Optional multi-agent
observation function.
observation_filter (str): Name of observation filter to use.
clip_rewards (bool): Whether to clip rewards to [-1, 1] prior to
experience postprocessing. Setting to None means clip for Atari
only.
clip_actions (bool): Whether to clip action values to the range
specified by the policy action space.
env_config (dict): Config to pass to the env creator.
model_config (dict): Config to use when creating the policy model.
policy_config (dict): Config to pass to the policy. In the
multi-agent case, this config will be merged with the
per-policy configs specified by `policy`.
worker_index (int): For remote workers, this should be set to a
non-zero and unique value. This index is passed to created envs
through EnvContext so that envs can be configured per worker.
num_workers (int): For remote workers, how many workers altogether
have been created?
monitor_path (str): Write out episode stats and videos to this
directory if specified.
log_dir (str): Directory where logs can be placed.
log_level (str): Set the root log level on creation.
callbacks (DefaultCallbacks): Custom training callbacks.
input_creator (func): Function that returns an InputReader object
for loading previous generated experiences.
input_evaluation (list): How to evaluate the policy performance.
This only makes sense to set when the input is reading offline
data. The possible values include:
- "is": the step-wise importance sampling estimator.
- "wis": the weighted step-wise is estimator.
- "simulation": run the environment in the background, but
use this data for evaluation only and never for learning.
output_creator (func): Function that returns an OutputWriter object
for saving generated experiences.
remote_worker_envs (bool): If using num_envs > 1, whether to create
those new envs in remote processes instead of in the current
process. This adds overheads, but can make sense if your envs
remote_env_batch_wait_ms (float): Timeout that remote workers
are waiting when polling environments. 0 (continue when at
least one env is ready) is a reasonable default, but optimal
value could be obtained by measuring your environment
step / reset and model inference perf.
soft_horizon (bool): Calculate rewards but don't reset the
environment when the horizon is hit.
no_done_at_end (bool): Ignore the done=True at the end of the
episode and instead record done=False.
seed (int): Set the seed of both np and tf to this value to
to ensure each remote worker has unique exploration behavior.
extra_python_environs (dict): Extra python environments need to
be set.
fake_sampler (bool): Use a fake (inf speed) sampler for testing.
"""
self._original_kwargs = locals().copy()
del self._original_kwargs["self"]
global _global_worker
_global_worker = self
# set extra environs first
if extra_python_environs:
for key, value in extra_python_environs.items():
os.environ[key] = str(value)
def gen_rollouts():
while True:
yield self.sample()
ParallelIteratorWorker.__init__(self, gen_rollouts, False)
policy_config = policy_config or {}
if (tf and policy_config.get("eager")
and not policy_config.get("no_eager_on_workers")
# This eager check is necessary for certain all-framework tests
# that use tf's eager_mode() context generator.
and not tf.executing_eagerly()):
tf.enable_eager_execution()
if log_level:
logging.getLogger("ray.rllib").setLevel(log_level)
if worker_index > 1:
disable_log_once_globally() # only need 1 worker to log
elif log_level == "DEBUG":
enable_periodic_logging()
env_context = EnvContext(env_config or {}, worker_index)
self.policy_config = policy_config
if callbacks:
self.callbacks = callbacks()
else:
from ray.rllib.agents.callbacks import DefaultCallbacks
self.callbacks = DefaultCallbacks()
self.worker_index = worker_index
self.num_workers = num_workers
model_config = model_config or {}
policy_mapping_fn = (policy_mapping_fn
or (lambda agent_id: DEFAULT_POLICY_ID))
if not callable(policy_mapping_fn):
raise ValueError("Policy mapping function not callable?")
self.env_creator = env_creator
self.rollout_fragment_length = rollout_fragment_length * num_envs
self.batch_mode = batch_mode
self.compress_observations = compress_observations
self.preprocessing_enabled = True
self.last_batch = None
self.global_vars = None
self.fake_sampler = fake_sampler
self.env = _validate_env(env_creator(env_context))
if isinstance(self.env, MultiAgentEnv) or \
isinstance(self.env, BaseEnv):
def wrap(env):
return env # we can't auto-wrap these env types
elif is_atari(self.env) and \
not model_config.get("custom_preprocessor") and \
preprocessor_pref == "deepmind":
# Deepmind wrappers already handle all preprocessing
self.preprocessing_enabled = False
if clip_rewards is None:
clip_rewards = True
def wrap(env):
env = wrap_deepmind(
env,
dim=model_config.get("dim"),
framestack=model_config.get("framestack"))
if monitor_path:
from gym import wrappers
env = wrappers.Monitor(env, monitor_path, resume=True)
return env
else:
def wrap(env):
if monitor_path:
from gym import wrappers
env = wrappers.Monitor(env, monitor_path, resume=True)
return env
self.env = wrap(self.env)
def make_env(vector_index):
return wrap(
env_creator(
env_context.copy_with_overrides(
vector_index=vector_index, remote=remote_worker_envs)))
self.tf_sess = None
policy_dict = _validate_and_canonicalize(policy, self.env)
self.policies_to_train = policies_to_train or list(policy_dict.keys())
# set numpy and python seed
if seed is not None:
np.random.seed(seed)
random.seed(seed)
if not hasattr(self.env, "seed"):
raise ValueError("Env doesn't support env.seed(): {}".format(
self.env))
self.env.seed(seed)
try:
assert torch is not None
torch.manual_seed(seed)
except AssertionError:
logger.info("Could not seed torch")
if _has_tensorflow_graph(policy_dict) and not (tf and
tf.executing_eagerly()):
if not tf:
raise ImportError("Could not import tensorflow")
with tf.Graph().as_default():
if tf_session_creator:
self.tf_sess = tf_session_creator()
else:
self.tf_sess = tf.Session(
config=tf.ConfigProto(
gpu_options=tf.GPUOptions(allow_growth=True)))
with self.tf_sess.as_default():
# set graph-level seed
if seed is not None:
tf.set_random_seed(seed)
self.policy_map, self.preprocessors = \
self._build_policy_map(policy_dict, policy_config)
if (ray.is_initialized()
and ray.worker._mode() != ray.worker.LOCAL_MODE):
if not ray.get_gpu_ids():
logger.debug(
"Creating policy evaluation worker {}".format(
worker_index) +
" on CPU (please ignore any CUDA init errors)")
elif not tf.test.is_gpu_available():
raise RuntimeError(
"GPUs were assigned to this worker by Ray, but "
"TensorFlow reports GPU acceleration is disabled. "
"This could be due to a bad CUDA or TF installation.")
else:
self.policy_map, self.preprocessors = self._build_policy_map(
policy_dict, policy_config)
self.multiagent = set(self.policy_map.keys()) != {DEFAULT_POLICY_ID}
if self.multiagent:
if not ((isinstance(self.env, MultiAgentEnv)
or isinstance(self.env, ExternalMultiAgentEnv))
or isinstance(self.env, BaseEnv)):
raise ValueError(
"Have multiple policies {}, but the env ".format(
self.policy_map) +
"{} is not a subclass of BaseEnv, MultiAgentEnv or "
"ExternalMultiAgentEnv?".format(self.env))
self.filters = {
policy_id: get_filter(observation_filter,
policy.observation_space.shape)
for (policy_id, policy) in self.policy_map.items()
}
if self.worker_index == 0:
logger.info("Built filter map: {}".format(self.filters))
# Always use vector env for consistency even if num_envs = 1
self.async_env = BaseEnv.to_base_env(
self.env,
make_env=make_env,
num_envs=num_envs,
remote_envs=remote_worker_envs,
remote_env_batch_wait_ms=remote_env_batch_wait_ms)
self.num_envs = num_envs
if self.batch_mode == "truncate_episodes":
pack_episodes = True
elif self.batch_mode == "complete_episodes":
rollout_fragment_length = float("inf") # never cut episodes
pack_episodes = False # sampler will return 1 episode per poll
else:
raise ValueError("Unsupported batch mode: {}".format(
self.batch_mode))
self.io_context = IOContext(log_dir, policy_config, worker_index, self)
self.reward_estimators = []
for method in input_evaluation:
if method == "simulation":
logger.warning(
"Requested 'simulation' input evaluation method: "
"will discard all sampler outputs and keep only metrics.")
sample_async = True
elif method == "is":
ise = ImportanceSamplingEstimator.create(self.io_context)
self.reward_estimators.append(ise)
elif method == "wis":
wise = WeightedImportanceSamplingEstimator.create(
self.io_context)
self.reward_estimators.append(wise)
else:
raise ValueError(
"Unknown evaluation method: {}".format(method))
if sample_async:
self.sampler = AsyncSampler(
self,
self.async_env,
self.policy_map,
policy_mapping_fn,
self.preprocessors,
self.filters,
clip_rewards,
rollout_fragment_length,
self.callbacks,
horizon=episode_horizon,
pack=pack_episodes,
tf_sess=self.tf_sess,
clip_actions=clip_actions,
blackhole_outputs="simulation" in input_evaluation,
soft_horizon=soft_horizon,
no_done_at_end=no_done_at_end,
observation_fn=observation_fn)
self.sampler.start()
else:
self.sampler = SyncSampler(
self,
self.async_env,
self.policy_map,
policy_mapping_fn,
self.preprocessors,
self.filters,
clip_rewards,
rollout_fragment_length,
self.callbacks,
horizon=episode_horizon,
pack=pack_episodes,
tf_sess=self.tf_sess,
clip_actions=clip_actions,
soft_horizon=soft_horizon,
no_done_at_end=no_done_at_end,
observation_fn=observation_fn)
self.input_reader = input_creator(self.io_context)
assert isinstance(self.input_reader, InputReader), self.input_reader
self.output_writer = output_creator(self.io_context)
assert isinstance(self.output_writer, OutputWriter), self.output_writer
logger.debug(
"Created rollout worker with env {} ({}), policies {}".format(
self.async_env, self.env, self.policy_map))
@override(EvaluatorInterface)
def sample(self):
"""Evaluate the current policies and return a batch of experiences.
Return:
SampleBatch|MultiAgentBatch from evaluating the current policies.
"""
if self.fake_sampler and self.last_batch is not None:
return self.last_batch
if log_once("sample_start"):
logger.info("Generating sample batch of size {}".format(
self.rollout_fragment_length))
batches = [self.input_reader.next()]
steps_so_far = batches[0].count
# In truncate_episodes mode, never pull more than 1 batch per env.
# This avoids over-running the target batch size.
if self.batch_mode == "truncate_episodes":
max_batches = self.num_envs
else:
max_batches = float("inf")
while (steps_so_far < self.rollout_fragment_length
and len(batches) < max_batches):
batch = self.input_reader.next()
steps_so_far += batch.count
batches.append(batch)
batch = batches[0].concat_samples(batches)
self.callbacks.on_sample_end(worker=self, samples=batch)
# Always do writes prior to compression for consistency and to allow
# for better compression inside the writer.
self.output_writer.write(batch)
# Do off-policy estimation if needed
if self.reward_estimators:
for sub_batch in batch.split_by_episode():
for estimator in self.reward_estimators:
estimator.process(sub_batch)
if log_once("sample_end"):
logger.info("Completed sample batch:\n\n{}\n".format(
summarize(batch)))
if self.compress_observations == "bulk":
batch.compress(bulk=True)
elif self.compress_observations:
batch.compress()
if self.fake_sampler:
self.last_batch = batch
return batch
@DeveloperAPI
@ray.method(num_return_vals=2)
def sample_with_count(self):
"""Same as sample() but returns the count as a separate future."""
batch = self.sample()
return batch, batch.count
@override(EvaluatorInterface)
def get_weights(self, policies=None):
if policies is None:
policies = self.policy_map.keys()
return {
pid: policy.get_weights()
for pid, policy in self.policy_map.items() if pid in policies
}
@override(EvaluatorInterface)
def set_weights(self, weights, global_vars=None):
for pid, w in weights.items():
self.policy_map[pid].set_weights(w)
if global_vars:
self.set_global_vars(global_vars)
@override(EvaluatorInterface)
def compute_gradients(self, samples):
if log_once("compute_gradients"):
logger.info("Compute gradients on:\n\n{}\n".format(
summarize(samples)))
if isinstance(samples, MultiAgentBatch):
grad_out, info_out = {}, {}
if self.tf_sess is not None:
builder = TFRunBuilder(self.tf_sess, "compute_gradients")
for pid, batch in samples.policy_batches.items():
if pid not in self.policies_to_train:
continue
grad_out[pid], info_out[pid] = (
self.policy_map[pid]._build_compute_gradients(
builder, batch))
grad_out = {k: builder.get(v) for k, v in grad_out.items()}
info_out = {k: builder.get(v) for k, v in info_out.items()}
else:
for pid, batch in samples.policy_batches.items():
if pid not in self.policies_to_train:
continue
grad_out[pid], info_out[pid] = (
self.policy_map[pid].compute_gradients(batch))
else:
grad_out, info_out = (
self.policy_map[DEFAULT_POLICY_ID].compute_gradients(samples))
info_out["batch_count"] = samples.count
if log_once("grad_out"):
logger.info("Compute grad info:\n\n{}\n".format(
summarize(info_out)))
return grad_out, info_out
@override(EvaluatorInterface)
def apply_gradients(self, grads):
if log_once("apply_gradients"):
logger.info("Apply gradients:\n\n{}\n".format(summarize(grads)))
if isinstance(grads, dict):
if self.tf_sess is not None:
builder = TFRunBuilder(self.tf_sess, "apply_gradients")
outputs = {
pid: self.policy_map[pid]._build_apply_gradients(
builder, grad)
for pid, grad in grads.items()
}
return {k: builder.get(v) for k, v in outputs.items()}
else:
return {
pid: self.policy_map[pid].apply_gradients(g)
for pid, g in grads.items()
}
else:
return self.policy_map[DEFAULT_POLICY_ID].apply_gradients(grads)
@override(EvaluatorInterface)
def learn_on_batch(self, samples):
if log_once("learn_on_batch"):
logger.info(
"Training on concatenated sample batches:\n\n{}\n".format(
summarize(samples)))
if isinstance(samples, MultiAgentBatch):
info_out = {}
to_fetch = {}
if self.tf_sess is not None:
builder = TFRunBuilder(self.tf_sess, "learn_on_batch")
else:
builder = None
for pid, batch in samples.policy_batches.items():
if pid not in self.policies_to_train:
continue
policy = self.policy_map[pid]
if builder and hasattr(policy, "_build_learn_on_batch"):
to_fetch[pid] = policy._build_learn_on_batch(
builder, batch)
else:
info_out[pid] = policy.learn_on_batch(batch)
info_out.update({k: builder.get(v) for k, v in to_fetch.items()})
else:
info_out = self.policy_map[DEFAULT_POLICY_ID].learn_on_batch(
samples)
if log_once("learn_out"):
logger.debug("Training out:\n\n{}\n".format(summarize(info_out)))
return info_out
def sample_and_learn(self, expected_batch_size, num_sgd_iter,
sgd_minibatch_size, standardize_fields):
"""Sample and batch and learn on it.
This is typically used in combination with distributed allreduce.
Arguments:
expected_batch_size (int): Expected number of samples to learn on.
num_sgd_iter (int): Number of SGD iterations.
sgd_minibatch_size (int): SGD minibatch size.
standardize_fields (list): List of sample fields to normalize.
Returns:
info: dictionary of extra metadata from learn_on_batch().
count: number of samples learned on.
"""
batch = self.sample()
assert batch.count == expected_batch_size, \
("Batch size possibly out of sync between workers, expected:",
expected_batch_size, "got:", batch.count)
logger.info("Executing distributed minibatch SGD "
"with epoch size {}, minibatch size {}".format(
batch.count, sgd_minibatch_size))
info = do_minibatch_sgd(batch, self.policy_map, self, num_sgd_iter,
sgd_minibatch_size, standardize_fields)
return info, batch.count
@DeveloperAPI
def get_metrics(self):
"""Returns a list of new RolloutMetric objects from evaluation."""
out = self.sampler.get_metrics()
for m in self.reward_estimators:
out.extend(m.get_metrics())
return out
@DeveloperAPI
def foreach_env(self, func):
"""Apply the given function to each underlying env instance."""
envs = self.async_env.get_unwrapped()
if not envs:
return [func(self.async_env)]
else:
return [func(e) for e in envs]
@DeveloperAPI
def get_policy(self, policy_id=DEFAULT_POLICY_ID):
"""Return policy for the specified id, or None.
Arguments:
policy_id (str): id of policy to return.
"""
return self.policy_map.get(policy_id)
@DeveloperAPI
def for_policy(self, func, policy_id=DEFAULT_POLICY_ID):
"""Apply the given function to the specified policy."""
return func(self.policy_map[policy_id])
@DeveloperAPI
def foreach_policy(self, func):
"""Apply the given function to each (policy, policy_id) tuple."""
return [func(policy, pid) for pid, policy in self.policy_map.items()]
@DeveloperAPI
def foreach_trainable_policy(self, func):
"""
Applies the given function to each (policy, policy_id) tuple, which
can be found in `self.policies_to_train`.
Args:
func (callable): A function - taking a Policy and its ID - that is
called on all Policies within `self.policies_to_train`.
Returns:
List[any]: The list of n return values of all
`func([policy], [ID])`-calls.
"""
return [
func(policy, pid) for pid, policy in self.policy_map.items()
if pid in self.policies_to_train
]
@DeveloperAPI
def sync_filters(self, new_filters):
"""Changes self's filter to given and rebases any accumulated delta.
Args:
new_filters (dict): Filters with new state to update local copy.
"""
assert all(k in new_filters for k in self.filters)
for k in self.filters:
self.filters[k].sync(new_filters[k])
@DeveloperAPI
def get_filters(self, flush_after=False):
"""Returns a snapshot of filters.
Args:
flush_after (bool): Clears the filter buffer state.
Returns:
return_filters (dict): Dict for serializable filters
"""
return_filters = {}
for k, f in self.filters.items():
return_filters[k] = f.as_serializable()
if flush_after:
f.clear_buffer()
return return_filters
@DeveloperAPI
def save(self):
filters = self.get_filters(flush_after=True)
state = {
pid: self.policy_map[pid].get_state()
for pid in self.policy_map
}
return pickle.dumps({"filters": filters, "state": state})
@DeveloperAPI
def restore(self, objs):
objs = pickle.loads(objs)
self.sync_filters(objs["filters"])
for pid, state in objs["state"].items():
self.policy_map[pid].set_state(state)
@DeveloperAPI
def set_global_vars(self, global_vars):
self.foreach_policy(lambda p, _: p.on_global_var_update(global_vars))
self.global_vars = global_vars
@DeveloperAPI
def get_global_vars(self):
return self.global_vars
@DeveloperAPI
def export_policy_model(self, export_dir, policy_id=DEFAULT_POLICY_ID):
self.policy_map[policy_id].export_model(export_dir)
@DeveloperAPI
def import_policy_model_from_h5(self,
import_file,
policy_id=DEFAULT_POLICY_ID):
self.policy_map[policy_id].import_model_from_h5(import_file)
@DeveloperAPI
def export_policy_checkpoint(self,
export_dir,
filename_prefix="model",
policy_id=DEFAULT_POLICY_ID):
self.policy_map[policy_id].export_checkpoint(export_dir,
filename_prefix)
@DeveloperAPI
def stop(self):
self.async_env.stop()
@DeveloperAPI
def creation_args(self):
"""Returns the args used to create this worker."""
return self._original_kwargs
def _build_policy_map(self, policy_dict, policy_config):
policy_map = {}
preprocessors = {}
for name, (cls, obs_space, act_space,
conf) in sorted(policy_dict.items()):
logger.debug("Creating policy for {}".format(name))
merged_conf = merge_dicts(policy_config, conf)
merged_conf["num_workers"] = self.num_workers
merged_conf["worker_index"] = self.worker_index
if self.preprocessing_enabled:
preprocessor = ModelCatalog.get_preprocessor_for_space(
obs_space, merged_conf.get("model"))
preprocessors[name] = preprocessor
obs_space = preprocessor.observation_space
else:
preprocessors[name] = NoPreprocessor(obs_space)
if isinstance(obs_space, gym.spaces.Dict) or \
isinstance(obs_space, gym.spaces.Tuple):
raise ValueError(
"Found raw Tuple|Dict space as input to policy. "
"Please preprocess these observations with a "
"Tuple|DictFlatteningPreprocessor.")
if tf and tf.executing_eagerly():
if hasattr(cls, "as_eager"):
cls = cls.as_eager()
if policy_config["eager_tracing"]:
cls = cls.with_tracing()
elif not issubclass(cls, TFPolicy):
pass # could be some other type of policy
else:
raise ValueError("This policy does not support eager "
"execution: {}".format(cls))
if tf:
with tf.variable_scope(name):
policy_map[name] = cls(obs_space, act_space, merged_conf)
else:
policy_map[name] = cls(obs_space, act_space, merged_conf)
if self.worker_index == 0:
logger.info("Built policy map: {}".format(policy_map))
logger.info("Built preprocessor map: {}".format(preprocessors))
return policy_map, preprocessors
def setup_torch_data_parallel(self, url, world_rank, world_size, backend):
"""Join a torch process group for distributed SGD."""
logger.info("Joining process group, url={}, world_rank={}, "
"world_size={}, backend={}".format(url, world_rank,
world_size, backend))
torch.distributed.init_process_group(
backend=backend,
init_method=url,
rank=world_rank,
world_size=world_size)
for pid, policy in self.policy_map.items():
if not isinstance(policy, TorchPolicy):
raise ValueError(
"This policy does not support torch distributed", policy)
policy.distributed_world_size = world_size
def get_node_ip(self):
"""Returns the IP address of the current node."""
return ray.services.get_node_ip_address()
def find_free_port(self):
"""Finds a free port on the current node."""
from ray.util.sgd import utils
return utils.find_free_port()
def __del__(self):
if hasattr(self, "sampler") and isinstance(self.sampler, AsyncSampler):
self.sampler.shutdown = True
class PolicyEvaluator(EvaluatorInterface):
"""Common ``PolicyEvaluator`` implementation that wraps a ``PolicyGraph``.
This class wraps a policy graph instance and an environment class to
collect experiences from the environment. You can create many replicas of
this class as Ray actors to scale RL training.
This class supports vectorized and multi-agent policy evaluation (e.g.,
VectorEnv, MultiAgentEnv, etc.)
Examples:
>>> # Create a policy evaluator and using it to collect experiences.
>>> evaluator = PolicyEvaluator(
... env_creator=lambda _: gym.make("CartPole-v0"),
... policy_graph=PGPolicyGraph)
>>> print(evaluator.sample())
SampleBatch({
"obs": [[...]], "actions": [[...]], "rewards": [[...]],
"dones": [[...]], "new_obs": [[...]]})
>>> # Creating policy evaluators using optimizer_cls.make().
>>> optimizer = SyncSamplesOptimizer.make(
... evaluator_cls=PolicyEvaluator,
... evaluator_args={
... "env_creator": lambda _: gym.make("CartPole-v0"),
... "policy_graph": PGPolicyGraph,
... },
... num_workers=10)
>>> for _ in range(10): optimizer.step()
>>> # Creating a multi-agent policy evaluator
>>> evaluator = PolicyEvaluator(
... env_creator=lambda _: MultiAgentTrafficGrid(num_cars=25),
... policy_graphs={
... # Use an ensemble of two policies for car agents
... "car_policy1":
... (PGPolicyGraph, Box(...), Discrete(...), {"gamma": 0.99}),
... "car_policy2":
... (PGPolicyGraph, Box(...), Discrete(...), {"gamma": 0.95}),
... # Use a single shared policy for all traffic lights
... "traffic_light_policy":
... (PGPolicyGraph, Box(...), Discrete(...), {}),
... },
... policy_mapping_fn=lambda agent_id:
... random.choice(["car_policy1", "car_policy2"])
... if agent_id.startswith("car_") else "traffic_light_policy")
>>> print(evaluator.sample())
MultiAgentBatch({
"car_policy1": SampleBatch(...),
"car_policy2": SampleBatch(...),
"traffic_light_policy": SampleBatch(...)})
"""
@DeveloperAPI
@classmethod
def as_remote(cls, num_cpus=None, num_gpus=None, resources=None):
return ray.remote(
num_cpus=num_cpus, num_gpus=num_gpus, resources=resources)(cls)
@DeveloperAPI
def __init__(self,
env_creator,
policy_graph,
policy_mapping_fn=None,
policies_to_train=None,
tf_session_creator=None,
batch_steps=100,
batch_mode="truncate_episodes",
episode_horizon=None,
preprocessor_pref="deepmind",
sample_async=False,
compress_observations=False,
num_envs=1,
observation_filter="NoFilter",
clip_rewards=None,
clip_actions=True,
env_config=None,
model_config=None,
policy_config=None,
worker_index=0,
monitor_path=None,
log_dir=None,
log_level=None,
callbacks=None,
input_creator=lambda ioctx: ioctx.default_sampler_input(),
input_evaluation=frozenset([]),
output_creator=lambda ioctx: NoopOutput(),
remote_worker_envs=False,
async_remote_worker_envs=False):
"""Initialize a policy evaluator.
Arguments:
env_creator (func): Function that returns a gym.Env given an
EnvContext wrapped configuration.
policy_graph (class|dict): Either a class implementing
PolicyGraph, or a dictionary of policy id strings to
(PolicyGraph, obs_space, action_space, config) tuples. If a
dict is specified, then we are in multi-agent mode and a
policy_mapping_fn should also be set.
policy_mapping_fn (func): A function that maps agent ids to
policy ids in multi-agent mode. This function will be called
each time a new agent appears in an episode, to bind that agent
to a policy for the duration of the episode.
policies_to_train (list): Optional whitelist of policies to train,
or None for all policies.
tf_session_creator (func): A function that returns a TF session.
This is optional and only useful with TFPolicyGraph.
batch_steps (int): The target number of env transitions to include
in each sample batch returned from this evaluator.
batch_mode (str): One of the following batch modes:
"truncate_episodes": Each call to sample() will return a batch
of at most `batch_steps * num_envs` in size. The batch will
be exactly `batch_steps * num_envs` in size if
postprocessing does not change batch sizes. Episodes may be
truncated in order to meet this size requirement.
"complete_episodes": Each call to sample() will return a batch
of at least `batch_steps * num_envs` in size. Episodes will
not be truncated, but multiple episodes may be packed
within one batch to meet the batch size. Note that when
`num_envs > 1`, episode steps will be buffered until the
episode completes, and hence batches may contain
significant amounts of off-policy data.
episode_horizon (int): Whether to stop episodes at this horizon.
preprocessor_pref (str): Whether to prefer RLlib preprocessors
("rllib") or deepmind ("deepmind") when applicable.
sample_async (bool): Whether to compute samples asynchronously in
the background, which improves throughput but can cause samples
to be slightly off-policy.
compress_observations (bool): If true, compress the observations.
They can be decompressed with rllib/utils/compression.
num_envs (int): If more than one, will create multiple envs
and vectorize the computation of actions. This has no effect if
if the env already implements VectorEnv.
observation_filter (str): Name of observation filter to use.
clip_rewards (bool): Whether to clip rewards to [-1, 1] prior to
experience postprocessing. Setting to None means clip for Atari
only.
clip_actions (bool): Whether to clip action values to the range
specified by the policy action space.
env_config (dict): Config to pass to the env creator.
model_config (dict): Config to use when creating the policy model.
policy_config (dict): Config to pass to the policy. In the
multi-agent case, this config will be merged with the
per-policy configs specified by `policy_graph`.
worker_index (int): For remote evaluators, this should be set to a
non-zero and unique value. This index is passed to created envs
through EnvContext so that envs can be configured per worker.
monitor_path (str): Write out episode stats and videos to this
directory if specified.
log_dir (str): Directory where logs can be placed.
log_level (str): Set the root log level on creation.
callbacks (dict): Dict of custom debug callbacks.
input_creator (func): Function that returns an InputReader object
for loading previous generated experiences.
input_evaluation (list): How to evaluate the policy performance.
This only makes sense to set when the input is reading offline
data. The possible values include:
- "is": the step-wise importance sampling estimator.
- "wis": the weighted step-wise is estimator.
- "simulation": run the environment in the background, but
use this data for evaluation only and never for learning.
output_creator (func): Function that returns an OutputWriter object
for saving generated experiences.
remote_worker_envs (bool): If using num_envs > 1, whether to create
those new envs in remote processes instead of in the current
process. This adds overheads, but can make sense if your envs
are very CPU intensive (e.g., for StarCraft).
async_remote_worker_envs (bool): Similar to remote_worker_envs,
but runs the envs asynchronously in the background.
"""
if log_level:
logging.getLogger("ray.rllib").setLevel(log_level)
env_context = EnvContext(env_config or {}, worker_index)
policy_config = policy_config or {}
self.policy_config = policy_config
self.callbacks = callbacks or {}
model_config = model_config or {}
policy_mapping_fn = (policy_mapping_fn
or (lambda agent_id: DEFAULT_POLICY_ID))
if not callable(policy_mapping_fn):
raise ValueError(
"Policy mapping function not callable. If you're using Tune, "
"make sure to escape the function with tune.function() "
"to prevent it from being evaluated as an expression.")
self.env_creator = env_creator
self.sample_batch_size = batch_steps * num_envs
self.batch_mode = batch_mode
self.compress_observations = compress_observations
self.preprocessing_enabled = True
self.env = _validate_env(env_creator(env_context))
if isinstance(self.env, MultiAgentEnv) or \
isinstance(self.env, BaseEnv):
def wrap(env):
return env # we can't auto-wrap these env types
elif is_atari(self.env) and \
not model_config.get("custom_preprocessor") and \
preprocessor_pref == "deepmind":
# Deepmind wrappers already handle all preprocessing
self.preprocessing_enabled = False
if clip_rewards is None:
clip_rewards = True
def wrap(env):
env = wrap_deepmind(
env,
dim=model_config.get("dim"),
framestack=model_config.get("framestack"))
if monitor_path:
env = _monitor(env, monitor_path)
return env
else:
def wrap(env):
if monitor_path:
env = _monitor(env, monitor_path)
return env
self.env = wrap(self.env)
def make_env(vector_index):
return wrap(
env_creator(
env_context.copy_with_overrides(
vector_index=vector_index, remote=remote_worker_envs)))
self.tf_sess = None
policy_dict = _validate_and_canonicalize(policy_graph, self.env)
self.policies_to_train = policies_to_train or list(policy_dict.keys())
if _has_tensorflow_graph(policy_dict):
if (ray.is_initialized()
and ray.worker._mode() != ray.worker.LOCAL_MODE
and not ray.get_gpu_ids()):
logger.info("Creating policy evaluation worker {}".format(
worker_index) +
" on CPU (please ignore any CUDA init errors)")
with tf.Graph().as_default():
if tf_session_creator:
self.tf_sess = tf_session_creator()
else:
self.tf_sess = tf.Session(
config=tf.ConfigProto(
gpu_options=tf.GPUOptions(allow_growth=True)))
with self.tf_sess.as_default():
self.policy_map, self.preprocessors = \
self._build_policy_map(policy_dict, policy_config)
else:
self.policy_map, self.preprocessors = self._build_policy_map(
policy_dict, policy_config)
self.multiagent = set(self.policy_map.keys()) != {DEFAULT_POLICY_ID}
if self.multiagent:
if not (isinstance(self.env, MultiAgentEnv)
or isinstance(self.env, BaseEnv)):
raise ValueError(
"Have multiple policy graphs {}, but the env ".format(
self.policy_map) +
"{} is not a subclass of MultiAgentEnv?".format(self.env))
self.filters = {
policy_id: get_filter(observation_filter,
policy.observation_space.shape)
for (policy_id, policy) in self.policy_map.items()
}
# Always use vector env for consistency even if num_envs = 1
self.async_env = BaseEnv.to_base_env(
self.env,
make_env=make_env,
num_envs=num_envs,
remote_envs=remote_worker_envs,
async_remote_envs=async_remote_worker_envs)
self.num_envs = num_envs
if self.batch_mode == "truncate_episodes":
unroll_length = batch_steps
pack_episodes = True
elif self.batch_mode == "complete_episodes":
unroll_length = float("inf") # never cut episodes
pack_episodes = False # sampler will return 1 episode per poll
else:
raise ValueError("Unsupported batch mode: {}".format(
self.batch_mode))
self.io_context = IOContext(log_dir, policy_config, worker_index, self)
self.reward_estimators = []
for method in input_evaluation:
if method == "simulation":
logger.warning(
"Requested 'simulation' input evaluation method: "
"will discard all sampler outputs and keep only metrics.")
sample_async = True
elif method == "is":
ise = ImportanceSamplingEstimator.create(self.io_context)
self.reward_estimators.append(ise)
elif method == "wis":
wise = WeightedImportanceSamplingEstimator.create(
self.io_context)
self.reward_estimators.append(wise)
else:
raise ValueError(
"Unknown evaluation method: {}".format(method))
if sample_async:
self.sampler = AsyncSampler(
self.async_env,
self.policy_map,
policy_mapping_fn,
self.preprocessors,
self.filters,
clip_rewards,
unroll_length,
self.callbacks,
horizon=episode_horizon,
pack=pack_episodes,
tf_sess=self.tf_sess,
clip_actions=clip_actions,
blackhole_outputs="simulation" in input_evaluation)
self.sampler.start()
else:
self.sampler = SyncSampler(
self.async_env,
self.policy_map,
policy_mapping_fn,
self.preprocessors,
self.filters,
clip_rewards,
unroll_length,
self.callbacks,
horizon=episode_horizon,
pack=pack_episodes,
tf_sess=self.tf_sess,
clip_actions=clip_actions)
self.input_reader = input_creator(self.io_context)
assert isinstance(self.input_reader, InputReader), self.input_reader
self.output_writer = output_creator(self.io_context)
assert isinstance(self.output_writer, OutputWriter), self.output_writer
logger.debug("Created evaluator with env {} ({}), policies {}".format(
self.async_env, self.env, self.policy_map))
@override(EvaluatorInterface)
def sample(self):
"""Evaluate the current policies and return a batch of experiences.
Return:
SampleBatch|MultiAgentBatch from evaluating the current policies.
"""
batches = [self.input_reader.next()]
steps_so_far = batches[0].count
# In truncate_episodes mode, never pull more than 1 batch per env.
# This avoids over-running the target batch size.
if self.batch_mode == "truncate_episodes":
max_batches = self.num_envs
else:
max_batches = float("inf")
while steps_so_far < self.sample_batch_size and len(
batches) < max_batches:
batch = self.input_reader.next()
steps_so_far += batch.count
batches.append(batch)
batch = batches[0].concat_samples(batches)
if self.callbacks.get("on_sample_end"):
self.callbacks["on_sample_end"]({
"evaluator": self,
"samples": batch
})
# Always do writes prior to compression for consistency and to allow
# for better compression inside the writer.
self.output_writer.write(batch)
# Do off-policy estimation if needed
if self.reward_estimators:
for sub_batch in batch.split_by_episode():
for estimator in self.reward_estimators:
estimator.process(sub_batch)
if self.compress_observations:
if isinstance(batch, MultiAgentBatch):
for data in batch.policy_batches.values():
data["obs"] = [pack(o) for o in data["obs"]]
data["new_obs"] = [pack(o) for o in data["new_obs"]]
else:
batch["obs"] = [pack(o) for o in batch["obs"]]
batch["new_obs"] = [pack(o) for o in batch["new_obs"]]
return batch
@DeveloperAPI
@ray.method(num_return_vals=2)
def sample_with_count(self):
"""Same as sample() but returns the count as a separate future."""
batch = self.sample()
return batch, batch.count
@override(EvaluatorInterface)
def get_weights(self, policies=None):
if policies is None:
policies = self.policy_map.keys()
return {
pid: policy.get_weights()
for pid, policy in self.policy_map.items() if pid in policies
}
@override(EvaluatorInterface)
def set_weights(self, weights):
for pid, w in weights.items():
self.policy_map[pid].set_weights(w)
@override(EvaluatorInterface)
def compute_gradients(self, samples):
if isinstance(samples, MultiAgentBatch):
grad_out, info_out = {}, {}
if self.tf_sess is not None:
builder = TFRunBuilder(self.tf_sess, "compute_gradients")
for pid, batch in samples.policy_batches.items():
if pid not in self.policies_to_train:
continue
grad_out[pid], info_out[pid] = (
self.policy_map[pid]._build_compute_gradients(
builder, batch))
grad_out = {k: builder.get(v) for k, v in grad_out.items()}
info_out = {k: builder.get(v) for k, v in info_out.items()}
else:
for pid, batch in samples.policy_batches.items():
if pid not in self.policies_to_train:
continue
grad_out[pid], info_out[pid] = (
self.policy_map[pid].compute_gradients(batch))
else:
grad_out, info_out = (
self.policy_map[DEFAULT_POLICY_ID].compute_gradients(samples))
info_out["batch_count"] = samples.count
return grad_out, info_out
@override(EvaluatorInterface)
def apply_gradients(self, grads):
if isinstance(grads, dict):
if self.tf_sess is not None:
builder = TFRunBuilder(self.tf_sess, "apply_gradients")
outputs = {
pid: self.policy_map[pid]._build_apply_gradients(
builder, grad)
for pid, grad in grads.items()
}
return {k: builder.get(v) for k, v in outputs.items()}
else:
return {
pid: self.policy_map[pid].apply_gradients(g)
for pid, g in grads.items()
}
else:
return self.policy_map[DEFAULT_POLICY_ID].apply_gradients(grads)
@override(EvaluatorInterface)
def learn_on_batch(self, samples):
if isinstance(samples, MultiAgentBatch):
info_out = {}
if self.tf_sess is not None:
builder = TFRunBuilder(self.tf_sess, "learn_on_batch")
for pid, batch in samples.policy_batches.items():
if pid not in self.policies_to_train:
continue
info_out[pid], _ = (
self.policy_map[pid]._build_learn_on_batch(
builder, batch))
info_out = {k: builder.get(v) for k, v in info_out.items()}
else:
for pid, batch in samples.policy_batches.items():
if pid not in self.policies_to_train:
continue
info_out[pid], _ = (
self.policy_map[pid].learn_on_batch(batch))
return info_out
else:
grad_fetch, apply_fetch = (
self.policy_map[DEFAULT_POLICY_ID].learn_on_batch(samples))
return grad_fetch
@DeveloperAPI
def get_metrics(self):
"""Returns a list of new RolloutMetric objects from evaluation."""
out = self.sampler.get_metrics()
for m in self.reward_estimators:
out.extend(m.get_metrics())
return out
@DeveloperAPI
def foreach_env(self, func):
"""Apply the given function to each underlying env instance."""
envs = self.async_env.get_unwrapped()
if not envs:
return [func(self.async_env)]
else:
return [func(e) for e in envs]
@DeveloperAPI
def get_policy(self, policy_id=DEFAULT_POLICY_ID):
"""Return policy graph for the specified id, or None.
Arguments:
policy_id (str): id of policy graph to return.
"""
return self.policy_map.get(policy_id)
@DeveloperAPI
def for_policy(self, func, policy_id=DEFAULT_POLICY_ID):
"""Apply the given function to the specified policy graph."""
return func(self.policy_map[policy_id])
@DeveloperAPI
def foreach_policy(self, func):
"""Apply the given function to each (policy, policy_id) tuple."""
return [func(policy, pid) for pid, policy in self.policy_map.items()]
@DeveloperAPI
def foreach_trainable_policy(self, func):
"""Apply the given function to each (policy, policy_id) tuple.
This only applies func to policies in `self.policies_to_train`."""
return [
func(policy, pid) for pid, policy in self.policy_map.items()
if pid in self.policies_to_train
]
@DeveloperAPI
def sync_filters(self, new_filters):
"""Changes self's filter to given and rebases any accumulated delta.
Args:
new_filters (dict): Filters with new state to update local copy.
"""
assert all(k in new_filters for k in self.filters)
for k in self.filters:
self.filters[k].sync(new_filters[k])
@DeveloperAPI
def get_filters(self, flush_after=False):
"""Returns a snapshot of filters.
Args:
flush_after (bool): Clears the filter buffer state.
Returns:
return_filters (dict): Dict for serializable filters
"""
return_filters = {}
for k, f in self.filters.items():
return_filters[k] = f.as_serializable()
if flush_after:
f.clear_buffer()
return return_filters
@DeveloperAPI
def save(self):
filters = self.get_filters(flush_after=True)
state = {
pid: self.policy_map[pid].get_state()
for pid in self.policy_map
}
return pickle.dumps({"filters": filters, "state": state})
@DeveloperAPI
def restore(self, objs):
objs = pickle.loads(objs)
self.sync_filters(objs["filters"])
for pid, state in objs["state"].items():
self.policy_map[pid].set_state(state)
@DeveloperAPI
def set_global_vars(self, global_vars):
self.foreach_policy(lambda p, _: p.on_global_var_update(global_vars))
@DeveloperAPI
def export_policy_model(self, export_dir, policy_id=DEFAULT_POLICY_ID):
self.policy_map[policy_id].export_model(export_dir)
@DeveloperAPI
def export_policy_checkpoint(self,
export_dir,
filename_prefix="model",
policy_id=DEFAULT_POLICY_ID):
self.policy_map[policy_id].export_checkpoint(export_dir,
filename_prefix)
def _build_policy_map(self, policy_dict, policy_config):
policy_map = {}
preprocessors = {}
for name, (cls, obs_space, act_space,
conf) in sorted(policy_dict.items()):
logger.debug("Creating policy graph for {}".format(name))
merged_conf = merge_dicts(policy_config, conf)
if self.preprocessing_enabled:
preprocessor = ModelCatalog.get_preprocessor_for_space(
obs_space, merged_conf.get("model"))
preprocessors[name] = preprocessor
obs_space = preprocessor.observation_space
else:
preprocessors[name] = NoPreprocessor(obs_space)
if isinstance(obs_space, gym.spaces.Dict) or \
isinstance(obs_space, gym.spaces.Tuple):
raise ValueError(
"Found raw Tuple|Dict space as input to policy graph. "
"Please preprocess these observations with a "
"Tuple|DictFlatteningPreprocessor.")
with tf.variable_scope(name):
policy_map[name] = cls(obs_space, act_space, merged_conf)
return policy_map, preprocessors
def __del__(self):
if hasattr(self, "sampler") and isinstance(self.sampler, AsyncSampler):
self.sampler.shutdown = True