Source code for tianshou.algorithm.imitation.discrete_cql

from dataclasses import dataclass

import numpy as np
import torch
import torch.nn.functional as F

from tianshou.algorithm import QRDQN
from tianshou.algorithm.algorithm_base import OfflineAlgorithm
from tianshou.algorithm.modelfree.qrdqn import QRDQNPolicy
from tianshou.algorithm.modelfree.reinforce import SimpleLossTrainingStats
from tianshou.algorithm.optim import OptimizerFactory
from tianshou.data import to_torch
from tianshou.data.types import RolloutBatchProtocol




@dataclass(kw_only=True)
class DiscreteCQLTrainingStats(SimpleLossTrainingStats):
    cql_loss: float
    qr_loss: float



# NOTE: This uses diamond inheritance to convert from off-policy to offline


class DiscreteCQL(OfflineAlgorithm[QRDQNPolicy], QRDQN[QRDQNPolicy]):  # type: ignore[misc]
    """Implementation of discrete Conservative Q-Learning algorithm. arXiv:2006.04779."""

    def __init__(
        self,
        *,
        policy: QRDQNPolicy,
        optim: OptimizerFactory,
        min_q_weight: float = 10.0,
        gamma: float = 0.99,
        num_quantiles: int = 200,
        n_step_return_horizon: int = 1,
        target_update_freq: int = 0,
    ) -> None:
        """
        :param policy: the policy
        :param optim: the optimizer factory for the policy's model.
        :param min_q_weight: the weight for the cql loss.
        :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 num_quantiles: the number of quantile midpoints in the inverse
            cumulative distribution function of the value.
        :param n_step_return_horizon: the number of future steps (> 0) to consider when computing temporal
            difference (TD) targets. Controls the balance between TD learning and Monte Carlo methods:
            higher values reduce bias (by relying less on potentially inaccurate value estimates)
            but increase variance (by incorporating more environmental stochasticity and reducing
            the averaging effect). A value of 1 corresponds to standard TD learning with immediate
            bootstrapping, while very large values approach Monte Carlo-like estimation that uses
            complete episode returns.
        :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.
        """
        QRDQN.__init__(
            self,
            policy=policy,
            optim=optim,
            gamma=gamma,
            num_quantiles=num_quantiles,
            n_step_return_horizon=n_step_return_horizon,
            target_update_freq=target_update_freq,
        )
        self.min_q_weight = min_q_weight

    def _update_with_batch(
        self,
        batch: RolloutBatchProtocol,
    ) -> DiscreteCQLTrainingStats:
        self._periodically_update_lagged_network_weights()
        weight = batch.pop("weight", 1.0)
        all_dist = self.policy(batch).logits
        act = to_torch(batch.act, dtype=torch.long, device=all_dist.device)
        curr_dist = all_dist[np.arange(len(act)), act, :].unsqueeze(2)
        target_dist = batch.returns.unsqueeze(1)
        # calculate each element's difference between curr_dist and target_dist
        dist_diff = F.smooth_l1_loss(target_dist, curr_dist, reduction="none")
        huber_loss = (
            (dist_diff * (self.tau_hat - (target_dist - curr_dist).detach().le(0.0).float()).abs())
            .sum(-1)
            .mean(1)
        )
        qr_loss = (huber_loss * weight).mean()
        # ref: https://github.com/ku2482/fqf-iqn-qrdqn.pytorch/
        # blob/master/fqf_iqn_qrdqn/agent/qrdqn_agent.py L130
        batch.weight = dist_diff.detach().abs().sum(-1).mean(1)  # prio-buffer
        # add CQL loss
        q = self.policy.compute_q_value(all_dist, None)
        dataset_expec = q.gather(1, act.unsqueeze(1)).mean()
        negative_sampling = q.logsumexp(1).mean()
        min_q_loss = negative_sampling - dataset_expec
        loss = qr_loss + min_q_loss * self.min_q_weight
        self.optim.step(loss)

        return DiscreteCQLTrainingStats(
            loss=loss.item(),
            qr_loss=qr_loss.item(),
            cql_loss=min_q_loss.item(),
        )