ray.rllib.core.rl_module.rl_module.RLModule.get_train_action_dist_cls#

RLModule.get_train_action_dist_cls() → Type[Distribution][source]#

Returns the action distribution class for this RLModule used for training.

This class is used to create action distributions from outputs of the forward_train method. If the case that no action distribution class is needed, this method can return None.

Note that RLlib’s distribution classes all implement the Distribution interface. This requires two special methods: Distribution.from_logits() and Distribution.to_deterministic(). See the documentation for Distribution for more detail.