def __init__(
self,
name,
# config_or_config_dict: (DictConfig, dict),
env: (Env, Wrapper),
algo: Algo,
env_spec: EnvSpec,
sampler: Sampler = None,
noise_adder: AgentActionNoiseWrapper = None,
reset_noise_every_terminal_state=False,
reset_state_every_sample=False,
exploration_strategy: ExplorationStrategy = None,
algo_saving_scheduler: EventScheduler = None):
"""
:param name: the name of the agent instance
:type name: str
:param env: environment that interacts with agent
:type env: Env
:param algo: algorithm of the agent
:type algo: Algo
:param env_spec: environment specifications: action apace and environment space
:type env_spec: EnvSpec
:param sampler: sampler
:type sampler: Sampler
:param reset_noise_every_terminal_state: reset the noise every sampled trajectory
:type reset_noise_every_terminal_state: bool
:param reset_state_every_sample: reset the state everytime perofrm the sample/rollout
:type reset_state_every_sample: bool
:param noise_adder: add action noise for exploration in action space
:type noise_adder: AgentActionNoiseWrapper
:param exploration_strategy: exploration strategy in action space
:type exploration_strategy: ExplorationStrategy
:param algo_saving_scheduler: control the schedule the varying parameters in training process
:type algo_saving_scheduler: EventSchedule
"""
super(Agent, self).__init__(name=name, status=StatusWithSubInfo(self))
self.parameters = Parameters(parameters=dict(
reset_noise_every_terminal_state=reset_noise_every_terminal_state,
reset_state_every_sample=reset_state_every_sample))
self.env = env
self.algo = algo
self._env_step_count = 0
if sampler is None:
sampler = Sampler()
self.sampler = sampler
self.recorder = Recorder(default_obj=self)
self.env_spec = env_spec
if exploration_strategy:
assert isinstance(exploration_strategy, ExplorationStrategy)
self.explorations_strategy = exploration_strategy
else:
self.explorations_strategy = None
self.noise_adder = noise_adder
self.algo_saving_scheduler = algo_saving_scheduler
def __init__(
self,
name,
# config_or_config_dict: (DictConfig, dict),
env: Env,
algo: Algo,
env_spec: EnvSpec,
sampler: Sampler = None,
noise_adder: AgentActionNoiseWrapper = None,
exploration_strategy: ExplorationStrategy = None,
algo_saving_scheduler: EventScheduler = None):
"""
:param name: the name of the agent instance
:type name: str
:param env: environment that interacts with agent
:type env: Env
:param algo: algorithm of the agent
:type algo: Algo
:param env_spec: environment specifications: action apace and environment space
:type env_spec: EnvSpec
:param sampler: sampler
:type sampler: Sampler
:param noise_adder: add action noise for exploration in action space
:type noise_adder: AgentActionNoiseWrapper
:param exploration_strategy: exploration strategy in action space
:type exploration_strategy: ExplorationStrategy
:param algo_saving_scheduler: control the schedule the varying parameters in training process
:type algo_saving_scheduler: EventSchedule
"""
super(Agent, self).__init__(name=name, status=StatusWithSubInfo(self))
self.parameters = Parameters(parameters=dict())
self.total_test_samples = 0
self.total_train_samples = 0
self.env = env
self.algo = algo
self._env_step_count = 0
self.sampler = sampler
self.recorder = Recorder()
self.env_spec = env_spec
if exploration_strategy:
assert isinstance(exploration_strategy, ExplorationStrategy)
self.explorations_strategy = exploration_strategy
else:
self.explorations_strategy = None
self.sampler = sampler if sampler else Sampler(
env_spec=env_spec, name='{}_sampler'.format(name))
self.noise_adder = noise_adder
self.algo_saving_scheduler = algo_saving_scheduler
def __init__(
self,
name,
# config_or_config_dict: (DictConfig, dict),
env: Env,
algo: Algo,
env_spec: EnvSpec,
sampler: Sampler = None,
noise_adder: AgentActionNoiseWrapper = None,
exploration_strategy: ExplorationStrategy = None,
algo_saving_scheduler: EventSchedule = None):
"""
:param name:
:param env:
:param algo:
:param env_spec:
:param sampler:
:param noise_adder:
:param exploration_strategy:
:param algo_saving_scheduler:
"""
super(Agent, self).__init__(name=name, status=StatusWithSubInfo(self))
self.parameters = Parameters(parameters=dict())
self.total_test_samples = 0
self.total_train_samples = 0
self.env = env
self.algo = algo
self._env_step_count = 0
self.sampler = sampler
self.recorder = Recorder()
self.env_spec = env_spec
if exploration_strategy:
assert isinstance(exploration_strategy, ExplorationStrategy)
self.explorations_strategy = exploration_strategy
else:
self.explorations_strategy = None
self.sampler = sampler if sampler else Sampler(
env_spec=env_spec, name='{}_sampler'.format(name))
self.noise_adder = noise_adder
self.algo_saving_scheduler = algo_saving_scheduler
def create_sampler(self, env_spec):
return Sampler(env_spec=env_spec, name='sampler')