from typing import Any, Dict, Optional, Sequence
from ..argument_utility import (
ActionScalerArg,
EncoderArg,
QFuncArg,
RewardScalerArg,
ScalerArg,
UseGPUArg,
check_encoder,
check_q_func,
check_use_gpu,
)
from ..constants import IMPL_NOT_INITIALIZED_ERROR, ActionSpace
from ..dataset import TransitionMiniBatch
from ..gpu import Device
from ..models.encoders import EncoderFactory
from ..models.optimizers import AdamFactory, OptimizerFactory
from ..models.q_functions import QFunctionFactory
from .base import AlgoBase
from .torch.awac_impl import AWACImpl
[docs]class AWAC(AlgoBase):
r"""Advantage Weighted Actor-Critic algorithm.
AWAC is a TD3-based actor-critic algorithm that enables efficient
fine-tuning where the policy is trained with offline datasets and is
deployed to online training.
The policy is trained as a supervised regression.
.. math::
J(\phi) = \mathbb{E}_{s_t, a_t \sim D}
[\log \pi_\phi(a_t|s_t)
\exp(\frac{1}{\lambda} A^\pi (s_t, a_t))]
where :math:`A^\pi (s_t, a_t) = Q_\theta(s_t, a_t) -
Q_\theta(s_t, a'_t)` and :math:`a'_t \sim \pi_\phi(\cdot|s_t)`
The key difference from AWR is that AWAC uses Q-function trained via TD
learning for the better sample-efficiency.
References:
* `Nair et al., Accelerating Online Reinforcement Learning with Offline
Datasets. <https://arxiv.org/abs/2006.09359>`_
Args:
actor_learning_rate (float): learning rate for policy function.
critic_learning_rate (float): learning rate for Q functions.
actor_optim_factory (d3rlpy.models.optimizers.OptimizerFactory):
optimizer factory for the actor.
critic_optim_factory (d3rlpy.models.optimizers.OptimizerFactory):
optimizer factory for the critic.
actor_encoder_factory (d3rlpy.models.encoders.EncoderFactory or str):
encoder factory for the actor.
critic_encoder_factory (d3rlpy.models.encoders.EncoderFactory or str):
encoder factory for the critic.
q_func_factory (d3rlpy.models.q_functions.QFunctionFactory or str):
Q function factory.
batch_size (int): mini-batch size.
n_frames (int): the number of frames to stack for image observation.
n_steps (int): N-step TD calculation.
gamma (float): discount factor.
tau (float): target network synchronization coefficiency.
lam (float): :math:`\lambda` for weight calculation.
n_action_samples (int): the number of sampled actions to calculate
:math:`A^\pi(s_t, a_t)`.
max_weight (float): maximum weight for cross-entropy loss.
n_critics (int): the number of Q functions for ensemble.
target_reduction_type (str): ensemble reduction method at target value
estimation. The available options are
``['min', 'max', 'mean', 'mix', 'none']``.
update_actor_interval (int): interval to update policy function.
use_gpu (bool, int or d3rlpy.gpu.Device):
flag to use GPU, device ID or device.
scaler (d3rlpy.preprocessing.Scaler or str): preprocessor.
The available options are `['pixel', 'min_max', 'standard']`
action_scaler (d3rlpy.preprocessing.ActionScaler or str):
action preprocessor. The available options are ``['min_max']``.
reward_scaler (d3rlpy.preprocessing.RewardScaler or str):
reward preprocessor. The available options are
``['clip', 'min_max', 'standard']``.
impl (d3rlpy.algos.torch.awac_impl.AWACImpl): algorithm implementation.
"""
_actor_learning_rate: float
_critic_learning_rate: float
_actor_optim_factory: OptimizerFactory
_critic_optim_factory: OptimizerFactory
_actor_encoder_factory: EncoderFactory
_critic_encoder_factory: EncoderFactory
_q_func_factory: QFunctionFactory
_tau: float
_lam: float
_n_action_samples: int
_max_weight: float
_n_critics: int
_target_reduction_type: str
_update_actor_interval: int
_use_gpu: Optional[Device]
_impl: Optional[AWACImpl]
def __init__(
self,
*,
actor_learning_rate: float = 3e-4,
critic_learning_rate: float = 3e-4,
actor_optim_factory: OptimizerFactory = AdamFactory(weight_decay=1e-4),
critic_optim_factory: OptimizerFactory = AdamFactory(),
actor_encoder_factory: EncoderArg = "default",
critic_encoder_factory: EncoderArg = "default",
q_func_factory: QFuncArg = "mean",
batch_size: int = 1024,
n_frames: int = 1,
n_steps: int = 1,
gamma: float = 0.99,
tau: float = 0.005,
lam: float = 1.0,
n_action_samples: int = 1,
max_weight: float = 20.0,
n_critics: int = 2,
target_reduction_type: str = "min",
update_actor_interval: int = 1,
use_gpu: UseGPUArg = False,
scaler: ScalerArg = None,
action_scaler: ActionScalerArg = None,
reward_scaler: RewardScalerArg = None,
impl: Optional[AWACImpl] = None,
**kwargs: Any
):
super().__init__(
batch_size=batch_size,
n_frames=n_frames,
n_steps=n_steps,
gamma=gamma,
scaler=scaler,
action_scaler=action_scaler,
reward_scaler=reward_scaler,
kwargs=kwargs,
)
self._actor_learning_rate = actor_learning_rate
self._critic_learning_rate = critic_learning_rate
self._actor_optim_factory = actor_optim_factory
self._critic_optim_factory = critic_optim_factory
self._actor_encoder_factory = check_encoder(actor_encoder_factory)
self._critic_encoder_factory = check_encoder(critic_encoder_factory)
self._q_func_factory = check_q_func(q_func_factory)
self._tau = tau
self._lam = lam
self._n_action_samples = n_action_samples
self._max_weight = max_weight
self._n_critics = n_critics
self._target_reduction_type = target_reduction_type
self._update_actor_interval = update_actor_interval
self._use_gpu = check_use_gpu(use_gpu)
self._impl = impl
def _create_impl(
self, observation_shape: Sequence[int], action_size: int
) -> None:
self._impl = AWACImpl(
observation_shape=observation_shape,
action_size=action_size,
actor_learning_rate=self._actor_learning_rate,
critic_learning_rate=self._critic_learning_rate,
actor_optim_factory=self._actor_optim_factory,
critic_optim_factory=self._critic_optim_factory,
actor_encoder_factory=self._actor_encoder_factory,
critic_encoder_factory=self._critic_encoder_factory,
q_func_factory=self._q_func_factory,
gamma=self._gamma,
tau=self._tau,
lam=self._lam,
n_action_samples=self._n_action_samples,
max_weight=self._max_weight,
n_critics=self._n_critics,
target_reduction_type=self._target_reduction_type,
use_gpu=self._use_gpu,
scaler=self._scaler,
action_scaler=self._action_scaler,
reward_scaler=self._reward_scaler,
)
self._impl.build()
def _update(self, batch: TransitionMiniBatch) -> Dict[str, float]:
assert self._impl is not None, IMPL_NOT_INITIALIZED_ERROR
metrics = {}
critic_loss = self._impl.update_critic(batch)
metrics.update({"critic_loss": critic_loss})
# delayed policy update
if self._grad_step % self._update_actor_interval == 0:
actor_loss, mean_std = self._impl.update_actor(batch)
metrics.update({"actor_loss": actor_loss, "mean_std": mean_std})
self._impl.update_critic_target()
self._impl.update_actor_target()
return metrics
[docs] def get_action_type(self) -> ActionSpace:
return ActionSpace.CONTINUOUS