from typing import cast
import gymnasium as gym
import numpy as np
import torch
from sensai.util.helper import mark_used
from tianshou.algorithm.algorithm_base import TArrOrActBatch
from tianshou.algorithm.modelfree.dqn import (
DiscreteQLearningPolicy,
QLearningOffPolicyAlgorithm,
)
from tianshou.algorithm.modelfree.reinforce import SimpleLossTrainingStats
from tianshou.algorithm.optim import OptimizerFactory
from tianshou.data import Batch, ReplayBuffer, to_numpy, to_torch, to_torch_as
from tianshou.data.batch import BatchProtocol
from tianshou.data.types import (
ActBatchProtocol,
BatchWithReturnsProtocol,
ModelOutputBatchProtocol,
ObsBatchProtocol,
RolloutBatchProtocol,
)
from tianshou.utils.net.common import BranchingNet
mark_used(ActBatchProtocol)
[docs]
class BDQNPolicy(DiscreteQLearningPolicy[BranchingNet]):
def __init__(
self,
*,
model: BranchingNet,
action_space: gym.spaces.Discrete,
observation_space: gym.Space | None = None,
eps_training: float = 0.0,
eps_inference: float = 0.0,
):
"""
:param model: BranchingNet mapping (obs, state, info) -> action_values_BA.
:param action_space: the environment's action space
:param observation_space: the environment's observation space.
:param eps_training: the epsilon value for epsilon-greedy exploration during training.
When collecting data for training, this is the probability of choosing a random action
instead of the action chosen by the policy.
A value of 0.0 means no exploration (fully greedy) and a value of 1.0 means full
exploration (fully random).
:param eps_inference: the epsilon value for epsilon-greedy exploration during inference,
i.e. non-training cases (such as evaluation during test steps).
The epsilon value is the probability of choosing a random action instead of the action
chosen by the policy.
A value of 0.0 means no exploration (fully greedy) and a value of 1.0 means full
exploration (fully random).
"""
super().__init__(
model=model,
action_space=action_space,
observation_space=observation_space,
eps_training=eps_training,
eps_inference=eps_inference,
)
[docs]
def forward(
self,
batch: ObsBatchProtocol,
state: dict | BatchProtocol | np.ndarray | None = None,
model: torch.nn.Module | None = None,
) -> ModelOutputBatchProtocol:
if model is None:
model = self.model
assert model is not None
obs = batch.obs
# TODO: this is very contrived, see also iqn.py
obs_next_BO = obs.obs if hasattr(obs, "obs") else obs
action_values_BA, hidden_BH = model(obs_next_BO, state=state, info=batch.info)
act_B = to_numpy(action_values_BA.argmax(dim=-1))
result = Batch(logits=action_values_BA, act=act_B, state=hidden_BH)
return cast(ModelOutputBatchProtocol, result)
[docs]
def add_exploration_noise(
self,
act: TArrOrActBatch,
batch: ObsBatchProtocol,
) -> TArrOrActBatch:
eps = self.eps_training if self.is_within_training_step else self.eps_inference
if np.isclose(eps, 0.0):
return act
if isinstance(act, np.ndarray):
bsz = len(act)
rand_mask = np.random.rand(bsz) < eps
rand_act = np.random.randint(
low=0,
high=self.model.action_per_branch,
size=(bsz, act.shape[-1]),
)
if hasattr(batch.obs, "mask"):
rand_act += batch.obs.mask
act[rand_mask] = rand_act[rand_mask]
return act # type: ignore[return-value]
else:
raise NotImplementedError(
f"Currently only numpy arrays are supported, got {type(act)=}."
)
[docs]
class BDQN(QLearningOffPolicyAlgorithm[BDQNPolicy]):
"""Implementation of the Branching Dueling Q-Network (BDQN) algorithm arXiv:1711.08946."""
def __init__(
self,
*,
policy: BDQNPolicy,
optim: OptimizerFactory,
gamma: float = 0.99,
target_update_freq: int = 0,
is_double: bool = True,
) -> None:
"""
:param policy: policy
:param optim: the optimizer factory for the policy's model.
:param gamma: the discount factor in [0, 1] for future rewards.
This determines how much future rewards are valued compared to immediate ones.
Lower values (closer to 0) make the agent focus on immediate rewards, creating "myopic"
behavior. Higher values (closer to 1) make the agent value long-term rewards more,
potentially improving performance in tasks where delayed rewards are important but
increasing training variance by incorporating more environmental stochasticity.
Typically set between 0.9 and 0.99 for most reinforcement learning tasks
:param target_update_freq: the number of training iterations between each complete update of
the target network.
Controls how frequently the target Q-network parameters are updated with the current
Q-network values.
A value of 0 disables the target network entirely, using only a single network for both
action selection and bootstrap targets.
Higher values provide more stable learning targets but slow down the propagation of new
value estimates. Lower positive values allow faster learning but may lead to instability
due to rapidly changing targets.
Typically set between 100-10000 for DQN variants, with exact values depending on environment
complexity.
:param is_double: flag indicating whether to use Double Q-learning for target value calculation.
If True, the algorithm uses the online network to select actions and the target network to evaluate their Q-values.
This decoupling helps reduce the overestimation bias that standard Q-learning is prone to.
If False, the algorithm selects actions by directly taking the maximum Q-value from the target network.
Note: This parameter is most effective when used with a target network (target_update_freq > 0).
"""
super().__init__(
policy=policy,
optim=optim,
gamma=gamma,
# BDQN implements its own returns computation (below), which supports only 1-step returns
n_step_return_horizon=1,
target_update_freq=target_update_freq,
)
self.is_double = is_double
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}
result = self.policy(obs_next_batch)
if self.use_target_network:
# target_Q = Q_old(s_, argmax(Q_new(s_, *)))
target_q = self.policy(obs_next_batch, model=self.model_old).logits
else:
target_q = result.logits
if self.is_double:
act = np.expand_dims(self.policy(obs_next_batch).act, -1)
act = to_torch(act, dtype=torch.long, device=target_q.device)
else:
act = target_q.max(-1).indices.unsqueeze(-1)
return torch.gather(target_q, -1, act).squeeze()
def _compute_return(
self,
batch: RolloutBatchProtocol,
buffer: ReplayBuffer,
indice: np.ndarray,
gamma: float = 0.99,
) -> BatchWithReturnsProtocol:
rew = batch.rew
with torch.no_grad():
target_q_torch = self._target_q(buffer, indice) # (bsz, ?)
target_q = to_numpy(target_q_torch)
end_flag = buffer.done.copy()
end_flag[buffer.unfinished_index()] = True
end_flag = end_flag[indice]
mean_target_q = np.mean(target_q, -1) if len(target_q.shape) > 1 else target_q
_target_q = rew + gamma * mean_target_q * (1 - end_flag)
target_q = np.repeat(_target_q[..., None], self.policy.model.num_branches, axis=-1)
target_q = np.repeat(target_q[..., None], self.policy.model.action_per_branch, axis=-1)
batch.returns = to_torch_as(target_q, target_q_torch)
if hasattr(batch, "weight"): # prio buffer update
batch.weight = to_torch_as(batch.weight, target_q_torch)
return cast(BatchWithReturnsProtocol, batch)
def _preprocess_batch(
self,
batch: RolloutBatchProtocol,
buffer: ReplayBuffer,
indices: np.ndarray,
) -> BatchWithReturnsProtocol:
"""Compute the 1-step return for BDQ targets."""
return self._compute_return(batch, buffer, indices)
def _update_with_batch( # type: ignore[override]
self,
batch: BatchWithReturnsProtocol,
) -> SimpleLossTrainingStats:
self._periodically_update_lagged_network_weights()
weight = batch.pop("weight", 1.0)
act = to_torch(batch.act, dtype=torch.long, device=batch.returns.device)
q = self.policy(batch).logits
act_mask = torch.zeros_like(q)
act_mask = act_mask.scatter_(-1, act.unsqueeze(-1), 1)
act_q = q * act_mask
returns = batch.returns
returns = returns * act_mask
td_error = returns - act_q
loss = (td_error.pow(2).sum(-1).mean(-1) * weight).mean()
batch.weight = td_error.sum(-1).sum(-1) # prio-buffer
self.optim.step(loss)
return SimpleLossTrainingStats(loss=loss.item())
