ray/rllib/agents/ars/ars.py

363 lines
13 KiB
Python

# Code in this file is copied and adapted from
# https://github.com/openai/evolution-strategies-starter and from
# https://github.com/modestyachts/ARS
from collections import namedtuple
import logging
import numpy as np
import time
import ray
from ray.rllib.agents import Trainer, with_common_config
from ray.rllib.agents.ars.ars_tf_policy import ARSTFPolicy
from ray.rllib.agents.es import optimizers, utils
from ray.rllib.agents.es.es import validate_config
from ray.rllib.agents.es.es_tf_policy import rollout
from ray.rllib.env.env_context import EnvContext
from ray.rllib.policy.sample_batch import DEFAULT_POLICY_ID
from ray.rllib.utils.annotations import override
from ray.rllib.utils import FilterManager
logger = logging.getLogger(__name__)
Result = namedtuple("Result", [
"noise_indices", "noisy_returns", "sign_noisy_returns", "noisy_lengths",
"eval_returns", "eval_lengths"
])
# yapf: disable
# __sphinx_doc_begin__
DEFAULT_CONFIG = with_common_config({
"action_noise_std": 0.0,
"noise_stdev": 0.02, # std deviation of parameter noise
"num_rollouts": 32, # number of perturbs to try
"rollouts_used": 32, # number of perturbs to keep in gradient estimate
"num_workers": 2,
"sgd_stepsize": 0.01, # sgd step-size
"observation_filter": "MeanStdFilter",
"noise_size": 250000000,
"eval_prob": 0.03, # probability of evaluating the parameter rewards
"report_length": 10, # how many of the last rewards we average over
"offset": 0,
})
# __sphinx_doc_end__
# yapf: enable
@ray.remote
def create_shared_noise(count):
"""Create a large array of noise to be shared by all workers."""
seed = 123
noise = np.random.RandomState(seed).randn(count).astype(np.float32)
return noise
class SharedNoiseTable:
def __init__(self, noise):
self.noise = noise
assert self.noise.dtype == np.float32
def get(self, i, dim):
return self.noise[i:i + dim]
def sample_index(self, dim):
return np.random.randint(0, len(self.noise) - dim + 1)
def get_delta(self, dim):
idx = self.sample_index(dim)
return idx, self.get(idx, dim)
@ray.remote
class Worker:
def __init__(self,
config,
env_creator,
noise,
worker_index,
min_task_runtime=0.2):
self.min_task_runtime = min_task_runtime
self.config = config
self.config["single_threaded"] = True
self.noise = SharedNoiseTable(noise)
env_context = EnvContext(config["env_config"] or {}, worker_index)
self.env = env_creator(env_context)
from ray.rllib import models
self.preprocessor = models.ModelCatalog.get_preprocessor(self.env)
policy_cls = get_policy_class(config)
self.policy = policy_cls(self.env.observation_space,
self.env.action_space, config)
@property
def filters(self):
return {DEFAULT_POLICY_ID: self.policy.observation_filter}
def sync_filters(self, new_filters):
for k in self.filters:
self.filters[k].sync(new_filters[k])
def get_filters(self, flush_after=False):
return_filters = {}
for k, f in self.filters.items():
return_filters[k] = f.as_serializable()
if flush_after:
f.clear_buffer()
return return_filters
def rollout(self, timestep_limit, add_noise=False):
rollout_rewards, rollout_fragment_length = rollout(
self.policy,
self.env,
timestep_limit=timestep_limit,
add_noise=add_noise,
offset=self.config["offset"])
return rollout_rewards, rollout_fragment_length
def do_rollouts(self, params, timestep_limit=None):
# Set the network weights.
self.policy.set_flat_weights(params)
noise_indices, returns, sign_returns, lengths = [], [], [], []
eval_returns, eval_lengths = [], []
# Perform some rollouts with noise.
while (len(noise_indices) == 0):
if np.random.uniform() < self.config["eval_prob"]:
# Do an evaluation run with no perturbation.
self.policy.set_flat_weights(params)
rewards, length = self.rollout(timestep_limit, add_noise=False)
eval_returns.append(rewards.sum())
eval_lengths.append(length)
else:
# Do a regular run with parameter perturbations.
noise_index = self.noise.sample_index(self.policy.num_params)
perturbation = self.config["noise_stdev"] * self.noise.get(
noise_index, self.policy.num_params)
# These two sampling steps could be done in parallel on
# different actors letting us update twice as frequently.
self.policy.set_flat_weights(params + perturbation)
rewards_pos, lengths_pos = self.rollout(timestep_limit)
self.policy.set_flat_weights(params - perturbation)
rewards_neg, lengths_neg = self.rollout(timestep_limit)
noise_indices.append(noise_index)
returns.append([rewards_pos.sum(), rewards_neg.sum()])
sign_returns.append(
[np.sign(rewards_pos).sum(),
np.sign(rewards_neg).sum()])
lengths.append([lengths_pos, lengths_neg])
return Result(
noise_indices=noise_indices,
noisy_returns=returns,
sign_noisy_returns=sign_returns,
noisy_lengths=lengths,
eval_returns=eval_returns,
eval_lengths=eval_lengths)
def get_policy_class(config):
if config["framework"] == "torch":
from ray.rllib.agents.ars.ars_torch_policy import ARSTorchPolicy
policy_cls = ARSTorchPolicy
else:
policy_cls = ARSTFPolicy
return policy_cls
class ARSTrainer(Trainer):
"""Large-scale implementation of Augmented Random Search in Ray."""
_name = "ARS"
_default_config = DEFAULT_CONFIG
@override(Trainer)
def _init(self, config, env_creator):
validate_config(config)
env_context = EnvContext(config["env_config"] or {}, worker_index=0)
env = env_creator(env_context)
policy_cls = get_policy_class(config)
self.policy = policy_cls(env.observation_space, env.action_space,
config)
self.optimizer = optimizers.SGD(self.policy, config["sgd_stepsize"])
self.rollouts_used = config["rollouts_used"]
self.num_rollouts = config["num_rollouts"]
self.report_length = config["report_length"]
# Create the shared noise table.
logger.info("Creating shared noise table.")
noise_id = create_shared_noise.remote(config["noise_size"])
self.noise = SharedNoiseTable(ray.get(noise_id))
# Create the actors.
logger.info("Creating actors.")
self.workers = [
Worker.remote(config, env_creator, noise_id, idx + 1)
for idx in range(config["num_workers"])
]
self.episodes_so_far = 0
self.reward_list = []
self.tstart = time.time()
@override(Trainer)
def get_policy(self, policy=DEFAULT_POLICY_ID):
if policy != DEFAULT_POLICY_ID:
raise ValueError("ARS has no policy '{}'! Use {} "
"instead.".format(policy, DEFAULT_POLICY_ID))
return self.policy
@override(Trainer)
def _train(self):
config = self.config
theta = self.policy.get_flat_weights()
assert theta.dtype == np.float32
assert len(theta.shape) == 1
# Put the current policy weights in the object store.
theta_id = ray.put(theta)
# Use the actors to do rollouts, note that we pass in the ID of the
# policy weights.
results, num_episodes, num_timesteps = self._collect_results(
theta_id, config["num_rollouts"])
all_noise_indices = []
all_training_returns = []
all_training_lengths = []
all_eval_returns = []
all_eval_lengths = []
# Loop over the results.
for result in results:
all_eval_returns += result.eval_returns
all_eval_lengths += result.eval_lengths
all_noise_indices += result.noise_indices
all_training_returns += result.noisy_returns
all_training_lengths += result.noisy_lengths
assert len(all_eval_returns) == len(all_eval_lengths)
assert (len(all_noise_indices) == len(all_training_returns) ==
len(all_training_lengths))
self.episodes_so_far += num_episodes
# Assemble the results.
eval_returns = np.array(all_eval_returns)
eval_lengths = np.array(all_eval_lengths)
noise_indices = np.array(all_noise_indices)
noisy_returns = np.array(all_training_returns)
noisy_lengths = np.array(all_training_lengths)
# keep only the best returns
# select top performing directions if rollouts_used < num_rollouts
max_rewards = np.max(noisy_returns, axis=1)
if self.rollouts_used > self.num_rollouts:
self.rollouts_used = self.num_rollouts
percentile = 100 * (1 - (self.rollouts_used / self.num_rollouts))
idx = np.arange(max_rewards.size)[
max_rewards >= np.percentile(max_rewards, percentile)]
noise_idx = noise_indices[idx]
noisy_returns = noisy_returns[idx, :]
# Compute and take a step.
g, count = utils.batched_weighted_sum(
noisy_returns[:, 0] - noisy_returns[:, 1],
(self.noise.get(index, self.policy.num_params)
for index in noise_idx),
batch_size=min(500, noisy_returns[:, 0].size))
g /= noise_idx.size
# scale the returns by their standard deviation
if not np.isclose(np.std(noisy_returns), 0.0):
g /= np.std(noisy_returns)
assert (g.shape == (self.policy.num_params, )
and g.dtype == np.float32)
# Compute the new weights theta.
theta, update_ratio = self.optimizer.update(-g)
# Set the new weights in the local copy of the policy.
self.policy.set_flat_weights(theta)
# update the reward list
if len(all_eval_returns) > 0:
self.reward_list.append(eval_returns.mean())
# Now sync the filters
FilterManager.synchronize({
DEFAULT_POLICY_ID: self.policy.observation_filter
}, self.workers)
info = {
"weights_norm": np.square(theta).sum(),
"weights_std": np.std(theta),
"grad_norm": np.square(g).sum(),
"update_ratio": update_ratio,
"episodes_this_iter": noisy_lengths.size,
"episodes_so_far": self.episodes_so_far,
}
result = dict(
episode_reward_mean=np.mean(
self.reward_list[-self.report_length:]),
episode_len_mean=eval_lengths.mean(),
timesteps_this_iter=noisy_lengths.sum(),
info=info)
return result
@override(Trainer)
def _stop(self):
# workaround for https://github.com/ray-project/ray/issues/1516
for w in self.workers:
w.__ray_terminate__.remote()
@override(Trainer)
def compute_action(self, observation, *args, **kwargs):
action = self.policy.compute_actions(observation, update=True)[0]
if kwargs.get("full_fetch"):
return action, [], {}
return action
def _collect_results(self, theta_id, min_episodes):
num_episodes, num_timesteps = 0, 0
results = []
while num_episodes < min_episodes:
logger.debug(
"Collected {} episodes {} timesteps so far this iter".format(
num_episodes, num_timesteps))
rollout_ids = [
worker.do_rollouts.remote(theta_id) for worker in self.workers
]
# Get the results of the rollouts.
for result in ray.get(rollout_ids):
results.append(result)
# Update the number of episodes and the number of timesteps
# keeping in mind that result.noisy_lengths is a list of lists,
# where the inner lists have length 2.
num_episodes += sum(len(pair) for pair in result.noisy_lengths)
num_timesteps += sum(
sum(pair) for pair in result.noisy_lengths)
return results, num_episodes, num_timesteps
def __getstate__(self):
return {
"weights": self.policy.get_flat_weights(),
"filter": self.policy.observation_filter,
"episodes_so_far": self.episodes_so_far,
}
def __setstate__(self, state):
self.episodes_so_far = state["episodes_so_far"]
self.policy.set_flat_weights(state["weights"])
self.policy.observation_filter = state["filter"]
FilterManager.synchronize({
DEFAULT_POLICY_ID: self.policy.observation_filter
}, self.workers)