d3rlpy/algos/qlearning/cql.py
import dataclasses
import math
from ...base import DeviceArg, LearnableConfig, register_learnable
from ...constants import ActionSpace
from ...models.builders import (
create_continuous_q_function,
create_discrete_q_function,
create_normal_policy,
create_parameter,
)
from ...models.encoders import EncoderFactory, make_encoder_field
from ...models.optimizers import OptimizerFactory, make_optimizer_field
from ...models.q_functions import QFunctionFactory, make_q_func_field
from ...types import Shape
from .base import QLearningAlgoBase
from .torch.cql_impl import CQLImpl, CQLModules, DiscreteCQLImpl
from .torch.dqn_impl import DQNModules
__all__ = ["CQLConfig", "CQL", "DiscreteCQLConfig", "DiscreteCQL"]
@dataclasses.dataclass()
class CQLConfig(LearnableConfig):
r"""Config of Conservative Q-Learning algorithm.
CQL is a SAC-based data-driven deep reinforcement learning algorithm, which
achieves state-of-the-art performance in offline RL problems.
CQL mitigates overestimation error by minimizing action-values under the
current policy and maximizing values under data distribution for
underestimation issue.
.. math::
L(\theta_i) = \alpha\, \mathbb{E}_{s_t \sim D}
\left[\log{\sum_a \exp{Q_{\theta_i}(s_t, a)}}
- \mathbb{E}_{a \sim D} \big[Q_{\theta_i}(s_t, a)\big] - \tau\right]
+ L_\mathrm{SAC}(\theta_i)
where :math:`\alpha` is an automatically adjustable value via Lagrangian
dual gradient descent and :math:`\tau` is a threshold value.
If the action-value difference is smaller than :math:`\tau`, the
:math:`\alpha` will become smaller.
Otherwise, the :math:`\alpha` will become larger to aggressively penalize
action-values.
In continuous control, :math:`\log{\sum_a \exp{Q(s, a)}}` is computed as
follows.
.. math::
\log{\sum_a \exp{Q(s, a)}} \approx \log{\left(
\frac{1}{2N} \sum_{a_i \sim \text{Unif}(a)}^N
\left[\frac{\exp{Q(s, a_i)}}{\text{Unif}(a)}\right]
+ \frac{1}{2N} \sum_{a_i \sim \pi_\phi(a|s)}^N
\left[\frac{\exp{Q(s, a_i)}}{\pi_\phi(a_i|s)}\right]\right)}
where :math:`N` is the number of sampled actions.
The rest of optimization is exactly same as :class:`d3rlpy.algos.SAC`.
References:
* `Kumar et al., Conservative Q-Learning for Offline Reinforcement
Learning. <https://arxiv.org/abs/2006.04779>`_
Args:
observation_scaler (d3rlpy.preprocessing.ObservationScaler):
Observation preprocessor.
action_scaler (d3rlpy.preprocessing.ActionScaler): Action preprocessor.
reward_scaler (d3rlpy.preprocessing.RewardScaler): Reward preprocessor.
actor_learning_rate (float): Learning rate for policy function.
critic_learning_rate (float): Learning rate for Q functions.
temp_learning_rate (float):
Learning rate for temperature parameter of SAC.
alpha_learning_rate (float): Learning rate for :math:`\alpha`.
actor_optim_factory (d3rlpy.models.optimizers.OptimizerFactory):
Optimizer factory for the actor.
critic_optim_factory (d3rlpy.models.optimizers.OptimizerFactory):
Optimizer factory for the critic.
temp_optim_factory (d3rlpy.models.optimizers.OptimizerFactory):
Optimizer factory for the temperature.
alpha_optim_factory (d3rlpy.models.optimizers.OptimizerFactory):
Optimizer factory for :math:`\alpha`.
actor_encoder_factory (d3rlpy.models.encoders.EncoderFactory):
Encoder factory for the actor.
critic_encoder_factory (d3rlpy.models.encoders.EncoderFactory):
Encoder factory for the critic.
q_func_factory (d3rlpy.models.q_functions.QFunctionFactory):
Q function factory.
batch_size (int): Mini-batch size.
gamma (float): Discount factor.
tau (float): Target network synchronization coefficiency.
n_critics (int): Number of Q functions for ensemble.
initial_temperature (float): Initial temperature value.
initial_alpha (float): Initial :math:`\alpha` value.
alpha_threshold (float): Threshold value described as :math:`\tau`.
conservative_weight (float): Constant weight to scale conservative loss.
n_action_samples (int): Number of sampled actions to compute
:math:`\log{\sum_a \exp{Q(s, a)}}`.
soft_q_backup (bool): Flag to use SAC-style backup.
max_q_backup (bool): Flag to sample max Q-values for target.
"""
actor_learning_rate: float = 1e-4
critic_learning_rate: float = 3e-4
temp_learning_rate: float = 1e-4
alpha_learning_rate: float = 1e-4
actor_optim_factory: OptimizerFactory = make_optimizer_field()
critic_optim_factory: OptimizerFactory = make_optimizer_field()
temp_optim_factory: OptimizerFactory = make_optimizer_field()
alpha_optim_factory: OptimizerFactory = make_optimizer_field()
actor_encoder_factory: EncoderFactory = make_encoder_field()
critic_encoder_factory: EncoderFactory = make_encoder_field()
q_func_factory: QFunctionFactory = make_q_func_field()
batch_size: int = 256
gamma: float = 0.99
tau: float = 0.005
n_critics: int = 2
initial_temperature: float = 1.0
initial_alpha: float = 1.0
alpha_threshold: float = 10.0
conservative_weight: float = 5.0
n_action_samples: int = 10
soft_q_backup: bool = False
max_q_backup: bool = False
def create(self, device: DeviceArg = False) -> "CQL":
return CQL(self, device)
@staticmethod
def get_type() -> str:
return "cql"
class CQL(QLearningAlgoBase[CQLImpl, CQLConfig]):
def inner_create_impl(
self, observation_shape: Shape, action_size: int
) -> None:
assert not (
self._config.soft_q_backup and self._config.max_q_backup
), "soft_q_backup and max_q_backup are mutually exclusive."
policy = create_normal_policy(
observation_shape,
action_size,
self._config.actor_encoder_factory,
device=self._device,
)
q_funcs, q_func_fowarder = create_continuous_q_function(
observation_shape,
action_size,
self._config.critic_encoder_factory,
self._config.q_func_factory,
n_ensembles=self._config.n_critics,
device=self._device,
)
targ_q_funcs, targ_q_func_forwarder = create_continuous_q_function(
observation_shape,
action_size,
self._config.critic_encoder_factory,
self._config.q_func_factory,
n_ensembles=self._config.n_critics,
device=self._device,
)
log_temp = create_parameter(
(1, 1),
math.log(self._config.initial_temperature),
device=self._device,
)
log_alpha = create_parameter(
(1, 1), math.log(self._config.initial_alpha), device=self._device
)
actor_optim = self._config.actor_optim_factory.create(
policy.named_modules(), lr=self._config.actor_learning_rate
)
critic_optim = self._config.critic_optim_factory.create(
q_funcs.named_modules(), lr=self._config.critic_learning_rate
)
if self._config.temp_learning_rate > 0:
temp_optim = self._config.temp_optim_factory.create(
log_temp.named_modules(), lr=self._config.temp_learning_rate
)
else:
temp_optim = None
if self._config.alpha_learning_rate > 0:
alpha_optim = self._config.alpha_optim_factory.create(
log_alpha.named_modules(), lr=self._config.alpha_learning_rate
)
else:
alpha_optim = None
modules = CQLModules(
policy=policy,
q_funcs=q_funcs,
targ_q_funcs=targ_q_funcs,
log_temp=log_temp,
log_alpha=log_alpha,
actor_optim=actor_optim,
critic_optim=critic_optim,
temp_optim=temp_optim,
alpha_optim=alpha_optim,
)
self._impl = CQLImpl(
observation_shape=observation_shape,
action_size=action_size,
modules=modules,
q_func_forwarder=q_func_fowarder,
targ_q_func_forwarder=targ_q_func_forwarder,
gamma=self._config.gamma,
tau=self._config.tau,
alpha_threshold=self._config.alpha_threshold,
conservative_weight=self._config.conservative_weight,
n_action_samples=self._config.n_action_samples,
soft_q_backup=self._config.soft_q_backup,
max_q_backup=self._config.max_q_backup,
device=self._device,
)
def get_action_type(self) -> ActionSpace:
return ActionSpace.CONTINUOUS
@dataclasses.dataclass()
class DiscreteCQLConfig(LearnableConfig):
r"""Config of Discrete version of Conservative Q-Learning algorithm.
Discrete version of CQL is a DoubleDQN-based data-driven deep reinforcement
learning algorithm (the original paper uses DQN), which achieves
state-of-the-art performance in offline RL problems.
CQL mitigates overestimation error by minimizing action-values under the
current policy and maximizing values under data distribution for
underestimation issue.
.. math::
L(\theta) = \alpha \mathbb{E}_{s_t \sim D}
[\log{\sum_a \exp{Q_{\theta}(s_t, a)}}
- \mathbb{E}_{a \sim D} [Q_{\theta}(s, a)]]
+ L_{DoubleDQN}(\theta)
References:
* `Kumar et al., Conservative Q-Learning for Offline Reinforcement
Learning. <https://arxiv.org/abs/2006.04779>`_
Args:
observation_scaler (d3rlpy.preprocessing.ObservationScaler):
Observation preprocessor.
reward_scaler (d3rlpy.preprocessing.RewardScaler): Reward preprocessor.
learning_rate (float): Learning rate.
optim_factory (d3rlpy.models.optimizers.OptimizerFactory):
Optimizer factory.
encoder_factory (d3rlpy.models.encoders.EncoderFactory):
Encoder factory.
q_func_factory (d3rlpy.models.q_functions.QFunctionFactory):
Q function factory.
batch_size (int): Mini-batch size.
gamma (float): Discount factor.
n_critics (int): Number of Q functions for ensemble.
target_update_interval (int): Interval to synchronize the target
network.
alpha (float): math:`\alpha` value above.
"""
learning_rate: float = 6.25e-5
optim_factory: OptimizerFactory = make_optimizer_field()
encoder_factory: EncoderFactory = make_encoder_field()
q_func_factory: QFunctionFactory = make_q_func_field()
batch_size: int = 32
gamma: float = 0.99
n_critics: int = 1
target_update_interval: int = 8000
alpha: float = 1.0
def create(self, device: DeviceArg = False) -> "DiscreteCQL":
return DiscreteCQL(self, device)
@staticmethod
def get_type() -> str:
return "discrete_cql"
class DiscreteCQL(QLearningAlgoBase[DiscreteCQLImpl, DiscreteCQLConfig]):
def inner_create_impl(
self, observation_shape: Shape, action_size: int
) -> None:
q_funcs, q_func_forwarder = create_discrete_q_function(
observation_shape,
action_size,
self._config.encoder_factory,
self._config.q_func_factory,
n_ensembles=self._config.n_critics,
device=self._device,
)
targ_q_funcs, targ_q_func_forwarder = create_discrete_q_function(
observation_shape,
action_size,
self._config.encoder_factory,
self._config.q_func_factory,
n_ensembles=self._config.n_critics,
device=self._device,
)
optim = self._config.optim_factory.create(
q_funcs.named_modules(), lr=self._config.learning_rate
)
modules = DQNModules(
q_funcs=q_funcs,
targ_q_funcs=targ_q_funcs,
optim=optim,
)
self._impl = DiscreteCQLImpl(
observation_shape=observation_shape,
action_size=action_size,
modules=modules,
q_func_forwarder=q_func_forwarder,
targ_q_func_forwarder=targ_q_func_forwarder,
target_update_interval=self._config.target_update_interval,
gamma=self._config.gamma,
alpha=self._config.alpha,
device=self._device,
)
def get_action_type(self) -> ActionSpace:
return ActionSpace.DISCRETE
register_learnable(CQLConfig)
register_learnable(DiscreteCQLConfig)