SAC¶
-
class
SAC
(model, gamma=None, tau=None, alpha=None, actor_lr=None, critic_lr=None)[source]¶ Bases:
parl.core.paddle.algorithm.Algorithm
-
__init__
(model, gamma=None, tau=None, alpha=None, actor_lr=None, critic_lr=None)[source]¶ SAC algorithm
Parameters: - model (parl.Model) – forward network of actor and critic.
- gamma (float) – discounted factor for reward computation
- tau (float) – decay coefficient when updating the weights of self.target_model with self.model
- alpha (float) – temperature parameter determines the relative importance of the entropy against the reward
- actor_lr (float) – learning rate of the actor model
- critic_lr (float) – learning rate of the critic model
-