torchrl.trainers.algorithms.configs.objectives.TransitiveQLossConfig#
- class torchrl.trainers.algorithms.configs.objectives.TransitiveQLossConfig(_partial_: bool = False, actor_network: Any = '???', qvalue_network: Any = '???', oracle_qvalue_network: Any = None, num_qvalue_nets: int = 2, discount: float = 0.99, expectile: float = 0.7, distance_weight: float = 0.0, alpha: float = 1.0, actor_action: str = 'mode', _target_: str = 'torchrl.objectives.TransitiveQLoss', _convert_: str = 'object')[source]#
Hydra configuration for
TransitiveQLoss.The actor, critic and oracle critic networks are supplied as instantiated modules (or lists of modules for ensembles).