Module sac 

Soft Actor-Critic (SAC) trainer for continuous control.

This module is the integration capstone of the SAC decomposition (#136): it assembles the four building blocks — SacActor, ContinuousQNetwork, ContinuousReplayBuffer, and the PendulumSwingUp env — into a working SAC trainer following Haarnoja et al. 2018 v2 (arXiv:1812.05905).

• SacConfig — hyperparameters (builder + validate()), mirroring crate::train::dqn::DQNConfig.
• SacTrainer — twin critics + targets, stochastic actor, automatic entropy-temperature tuning, replay buffer, and Polyak target updates, driven by three independent Adam optimizers.

Structs

LogAlpha

Tiny single-parameter Module holding the log of the entropy temperature alpha.

SacConfig

SAC configuration parameters.

SacStepStats

Per-step training statistics for the SAC trainer.

SacTrainer

Burn-backend Soft Actor-Critic trainer.