3a1bc18add
This PR adds a new method for getting actions from an env's observation and info. This is useful for standard inference and stands in contrast to batch-based methods that are currently used in training and evaluation. Without this, users have to do some kind of gymnastics to actually perform inference with a trained policy. I have also added a test for the new method. In future PRs, this method should be included in the examples (in the the "watch" section). To add this required improving multiple typing things and, importantly, _simplifying the signature of `forward` in many policies!_ This is a **breaking change**, but it will likely affect no users. The `input` parameter of forward was a rather hacky mechanism, I believe it is good that it's gone now. It will also help with #948 . The main functional change is the addition of `compute_action` to `BasePolicy`. Other minor changes: - improvements in typing - updated PR and Issue templates - Improved handling of `max_action_num` Closes #981
184 lines
6.8 KiB
Python
184 lines
6.8 KiB
Python
from typing import Any, cast
|
|
|
|
import gymnasium as gym
|
|
import numpy as np
|
|
import torch
|
|
from overrides import override
|
|
from torch.distributions import Categorical
|
|
|
|
from tianshou.data import Batch, ReplayBuffer, to_torch
|
|
from tianshou.data.batch import BatchProtocol
|
|
from tianshou.data.types import ObsBatchProtocol, RolloutBatchProtocol
|
|
from tianshou.policy import SACPolicy
|
|
from tianshou.policy.base import TLearningRateScheduler
|
|
|
|
|
|
class DiscreteSACPolicy(SACPolicy):
|
|
"""Implementation of SAC for Discrete Action Settings. arXiv:1910.07207.
|
|
|
|
:param actor: the actor network following the rules in
|
|
:class:`~tianshou.policy.BasePolicy`. (s -> logits)
|
|
:param actor_optim: the optimizer for actor network.
|
|
:param critic: the first critic network. (s, a -> Q(s, a))
|
|
:param critic_optim: the optimizer for the first critic network.
|
|
:param action_space: Env's action space. Should be gym.spaces.Box.
|
|
:param critic2: the second critic network. (s, a -> Q(s, a)).
|
|
If None, use the same network as critic (via deepcopy).
|
|
:param critic2_optim: the optimizer for the second critic network.
|
|
If None, clone critic_optim to use for critic2.parameters().
|
|
:param tau: param for soft update of the target network.
|
|
:param gamma: discount factor, in [0, 1].
|
|
:param alpha: entropy regularization coefficient.
|
|
If a tuple (target_entropy, log_alpha, alpha_optim) is provided,
|
|
then alpha is automatically tuned.
|
|
:param estimation_step: the number of steps to look ahead for calculating
|
|
:param observation_space: Env's observation space.
|
|
:param lr_scheduler: a learning rate scheduler that adjusts the learning rate
|
|
in optimizer in each policy.update()
|
|
|
|
.. seealso::
|
|
|
|
Please refer to :class:`~tianshou.policy.BasePolicy` for more detailed
|
|
explanation.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
*,
|
|
actor: torch.nn.Module,
|
|
actor_optim: torch.optim.Optimizer,
|
|
critic: torch.nn.Module,
|
|
critic_optim: torch.optim.Optimizer,
|
|
action_space: gym.spaces.Discrete,
|
|
critic2: torch.nn.Module | None = None,
|
|
critic2_optim: torch.optim.Optimizer | None = None,
|
|
tau: float = 0.005,
|
|
gamma: float = 0.99,
|
|
alpha: float | tuple[float, torch.Tensor, torch.optim.Optimizer] = 0.2,
|
|
estimation_step: int = 1,
|
|
observation_space: gym.Space | None = None,
|
|
lr_scheduler: TLearningRateScheduler | None = None,
|
|
) -> None:
|
|
super().__init__(
|
|
actor=actor,
|
|
actor_optim=actor_optim,
|
|
critic=critic,
|
|
critic_optim=critic_optim,
|
|
action_space=action_space,
|
|
critic2=critic2,
|
|
critic2_optim=critic2_optim,
|
|
tau=tau,
|
|
gamma=gamma,
|
|
alpha=alpha,
|
|
estimation_step=estimation_step,
|
|
# Note: inheriting from continuous sac reduces code duplication,
|
|
# but continuous stuff has to be disabled
|
|
exploration_noise=None,
|
|
action_scaling=False,
|
|
action_bound_method=None,
|
|
observation_space=observation_space,
|
|
lr_scheduler=lr_scheduler,
|
|
)
|
|
|
|
# TODO: violates Liskov substitution principle, incompatible action space with SAC
|
|
# Not too urgent, but still..
|
|
@override
|
|
def _check_field_validity(self) -> None:
|
|
if not isinstance(self.action_space, gym.spaces.Discrete):
|
|
raise ValueError(
|
|
f"DiscreteSACPolicy only supports gym.spaces.Discrete, but got {self.action_space=}."
|
|
f"Please use SACPolicy for continuous action spaces.",
|
|
)
|
|
|
|
def forward( # type: ignore
|
|
self,
|
|
batch: ObsBatchProtocol,
|
|
state: dict | Batch | np.ndarray | None = None,
|
|
**kwargs: Any,
|
|
) -> Batch:
|
|
logits, hidden = self.actor(batch.obs, state=state, info=batch.info)
|
|
dist = Categorical(logits=logits)
|
|
if self.deterministic_eval and not self.training:
|
|
act = logits.argmax(axis=-1)
|
|
else:
|
|
act = dist.sample()
|
|
return Batch(logits=logits, act=act, state=hidden, dist=dist)
|
|
|
|
def _target_q(self, buffer: ReplayBuffer, indices: np.ndarray) -> torch.Tensor:
|
|
obs_next_batch = Batch(
|
|
obs=buffer[indices].obs_next,
|
|
info=[None] * len(indices),
|
|
) # obs_next: s_{t+n}
|
|
obs_next_result = self(obs_next_batch)
|
|
dist = obs_next_result.dist
|
|
target_q = dist.probs * torch.min(
|
|
self.critic_old(obs_next_batch.obs),
|
|
self.critic2_old(obs_next_batch.obs),
|
|
)
|
|
return target_q.sum(dim=-1) + self.alpha * dist.entropy()
|
|
|
|
def learn(self, batch: RolloutBatchProtocol, *args: Any, **kwargs: Any) -> dict[str, float]:
|
|
weight = batch.pop("weight", 1.0)
|
|
target_q = batch.returns.flatten()
|
|
act = to_torch(batch.act[:, np.newaxis], device=target_q.device, dtype=torch.long)
|
|
|
|
# critic 1
|
|
current_q1 = self.critic(batch.obs).gather(1, act).flatten()
|
|
td1 = current_q1 - target_q
|
|
critic1_loss = (td1.pow(2) * weight).mean()
|
|
|
|
self.critic_optim.zero_grad()
|
|
critic1_loss.backward()
|
|
self.critic_optim.step()
|
|
|
|
# critic 2
|
|
current_q2 = self.critic2(batch.obs).gather(1, act).flatten()
|
|
td2 = current_q2 - target_q
|
|
critic2_loss = (td2.pow(2) * weight).mean()
|
|
|
|
self.critic2_optim.zero_grad()
|
|
critic2_loss.backward()
|
|
self.critic2_optim.step()
|
|
batch.weight = (td1 + td2) / 2.0 # prio-buffer
|
|
|
|
# actor
|
|
dist = self(batch).dist
|
|
entropy = dist.entropy()
|
|
with torch.no_grad():
|
|
current_q1a = self.critic(batch.obs)
|
|
current_q2a = self.critic2(batch.obs)
|
|
q = torch.min(current_q1a, current_q2a)
|
|
actor_loss = -(self.alpha * entropy + (dist.probs * q).sum(dim=-1)).mean()
|
|
self.actor_optim.zero_grad()
|
|
actor_loss.backward()
|
|
self.actor_optim.step()
|
|
|
|
if self.is_auto_alpha:
|
|
log_prob = -entropy.detach() + self.target_entropy
|
|
alpha_loss = -(self.log_alpha * log_prob).mean()
|
|
self.alpha_optim.zero_grad()
|
|
alpha_loss.backward()
|
|
self.alpha_optim.step()
|
|
self.alpha = self.log_alpha.detach().exp()
|
|
|
|
self.sync_weight()
|
|
|
|
result = {
|
|
"loss/actor": actor_loss.item(),
|
|
"loss/critic1": critic1_loss.item(),
|
|
"loss/critic2": critic2_loss.item(),
|
|
}
|
|
if self.is_auto_alpha:
|
|
self.alpha = cast(torch.Tensor, self.alpha)
|
|
result["loss/alpha"] = alpha_loss.item()
|
|
result["alpha"] = self.alpha.item()
|
|
|
|
return result
|
|
|
|
def exploration_noise(
|
|
self,
|
|
act: np.ndarray | BatchProtocol,
|
|
batch: RolloutBatchProtocol,
|
|
) -> np.ndarray | BatchProtocol:
|
|
return act
|