class PGAgent(Agent):
"""Simple policy gradient agent.
This is an example agent to show how to implement algorithms in RLlib.
In most cases, you will probably want to use the PPO agent instead.
"""
_agent_name = "PG"
_default_config = DEFAULT_CONFIG
_policy_graph = PGPolicyGraph
@classmethod
def default_resource_request(cls, config):
cf = merge_dicts(cls._default_config, config)
return Resources(cpu=1, gpu=0, extra_cpu=cf["num_workers"])
def _init(self):
self.local_evaluator = self.make_local_evaluator(
self.env_creator, self._policy_graph)
self.remote_evaluators = self.make_remote_evaluators(
self.env_creator, self._policy_graph, self.config["num_workers"],
{})
self.optimizer = SyncSamplesOptimizer(self.local_evaluator,
self.remote_evaluators,
self.config["optimizer"])
def _train(self):
prev_steps = self.optimizer.num_steps_sampled
self.optimizer.step()
result = self.optimizer.collect_metrics()
result.update(timesteps_this_iter=self.optimizer.num_steps_sampled -
prev_steps)
return result
class PGAgent(Agent):
"""Simple policy gradient agent.
This is an example agent to show how to implement algorithms in RLlib.
In most cases, you will probably want to use the PPO agent instead.
"""
_agent_name = "PG"
_default_config = DEFAULT_CONFIG
_policy_graph = PGPolicyGraph
@override(Agent)
def _init(self):
self.local_evaluator = self.make_local_evaluator(
self.env_creator, self._policy_graph)
self.remote_evaluators = self.make_remote_evaluators(
self.env_creator, self._policy_graph, self.config["num_workers"])
optimizer_config = dict(
self.config["optimizer"],
**{"train_batch_size": self.config["train_batch_size"]})
self.optimizer = SyncSamplesOptimizer(
self.local_evaluator, self.remote_evaluators, optimizer_config)
@override(Agent)
def _train(self):
prev_steps = self.optimizer.num_steps_sampled
self.optimizer.step()
result = self.optimizer.collect_metrics(
self.config["collect_metrics_timeout"])
result.update(timesteps_this_iter=self.optimizer.num_steps_sampled -
prev_steps)
return result
class PGAgent(Agent):
"""Simple policy gradient agent.
This is an example agent to show how to implement algorithms in RLlib.
In most cases, you will probably want to use the PPO agent instead.
"""
_agent_name = "PG"
_default_config = DEFAULT_CONFIG
_policy_graph = PGPolicyGraph
@override(Agent)
def _init(self):
self.local_evaluator = self.make_local_evaluator(
self.env_creator, self._policy_graph)
self.remote_evaluators = self.make_remote_evaluators(
self.env_creator, self._policy_graph, self.config["num_workers"])
optimizer_config = dict(
self.config["optimizer"],
**{"train_batch_size": self.config["train_batch_size"]})
self.optimizer = SyncSamplesOptimizer(self.local_evaluator,
self.remote_evaluators,
optimizer_config)
@override(Agent)
def _train(self):
prev_steps = self.optimizer.num_steps_sampled
self.optimizer.step()
result = self.optimizer.collect_metrics(
self.config["collect_metrics_timeout"])
result.update(timesteps_this_iter=self.optimizer.num_steps_sampled -
prev_steps)
return result