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.td3_impl import TD3Impl
[docs]class TD3(AlgoBase):
r"""Twin Delayed Deep Deterministic Policy Gradients algorithm.
TD3 is an improved DDPG-based algorithm.
Major differences from DDPG are as follows.
* TD3 has twin Q functions to reduce overestimation bias at TD learning.
The number of Q functions can be designated by `n_critics`.
* TD3 adds noise to target value estimation to avoid overfitting with the
deterministic policy.
* TD3 updates the policy function after several Q function updates in order
to reduce variance of action-value estimation. The interval of the policy
function update can be designated by `update_actor_interval`.
.. math::
L(\theta_i) = \mathbb{E}_{s_t, a_t, r_{t+1}, s_{t+1} \sim D} [(r_{t+1}
+ \gamma \min_j Q_{\theta_j'}(s_{t+1}, \pi_{\phi'}(s_{t+1}) +
\epsilon) - Q_{\theta_i}(s_t, a_t))^2]
.. math::
J(\phi) = \mathbb{E}_{s_t \sim D}
[\min_i Q_{\theta_i}(s_t, \pi_\phi(s_t))]
where :math:`\epsilon \sim clip (N(0, \sigma), -c, c)`
References:
* `Fujimoto et al., Addressing Function Approximation Error in
Actor-Critic Methods. <https://arxiv.org/abs/1802.09477>`_
Args:
actor_learning_rate (float): learning rate for a 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.
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']``.
target_smoothing_sigma (float): standard deviation for target noise.
target_smoothing_clip (float): clipping range for target noise.
update_actor_interval (int): interval to update policy function
described as `delayed policy update` in the paper.
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.td3_impl.TD3Impl): 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
_n_critics: int
_target_reduction_type: str
_target_smoothing_sigma: float
_target_smoothing_clip: float
_update_actor_interval: int
_use_gpu: Optional[Device]
_impl: Optional[TD3Impl]
def __init__(
self,
*,
actor_learning_rate: float = 3e-4,
critic_learning_rate: float = 3e-4,
actor_optim_factory: OptimizerFactory = AdamFactory(),
critic_optim_factory: OptimizerFactory = AdamFactory(),
actor_encoder_factory: EncoderArg = "default",
critic_encoder_factory: EncoderArg = "default",
q_func_factory: QFuncArg = "mean",
batch_size: int = 100,
n_frames: int = 1,
n_steps: int = 1,
gamma: float = 0.99,
tau: float = 0.005,
n_critics: int = 2,
target_reduction_type: str = "min",
target_smoothing_sigma: float = 0.2,
target_smoothing_clip: float = 0.5,
update_actor_interval: int = 2,
use_gpu: UseGPUArg = False,
scaler: ScalerArg = None,
action_scaler: ActionScalerArg = None,
reward_scaler: RewardScalerArg = None,
impl: Optional[TD3Impl] = 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._n_critics = n_critics
self._target_reduction_type = target_reduction_type
self._target_smoothing_sigma = target_smoothing_sigma
self._target_smoothing_clip = target_smoothing_clip
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 = TD3Impl(
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,
n_critics=self._n_critics,
target_reduction_type=self._target_reduction_type,
target_smoothing_sigma=self._target_smoothing_sigma,
target_smoothing_clip=self._target_smoothing_clip,
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 = self._impl.update_actor(batch)
metrics.update({"actor_loss": actor_loss})
self._impl.update_critic_target()
self._impl.update_actor_target()
return metrics
[docs] def get_action_type(self) -> ActionSpace:
return ActionSpace.CONTINUOUS