ReverseCurriculumMidpointStartPolicy

ReverseCurriculumMidpointStartPolicy(configuration: ReverseCurriculumMidpointStartConfiguration, view: CurriculumCompetenceView) -> None

Competence-driven recede: the probability falls as the policy learns to win from progressively earlier cells of the best route. Stateless -- recomputed from current competence each call, so there is no curriculum state to checkpoint.

Source

Methods

midpoint_probability

midpoint_probability(step: int) -> float

Source