Skip to content

ResetParameterCurriculum ​

python
ResetParameterCurriculum(configuration: ResetParameterCurriculumConfiguration, rng: random.Random) -> None

Chooses the reset parameters a new episode starts with, and that stage's effective (activation-relative) step.

Parameters:

  • configuration (ResetParameterCurriculumConfiguration) – The curriculum configuration.
  • rng (Random) – Source of randomness for weighted stage selection.

Source

Methods ​

choose ​

python
choose(step: int) -> tuple[dict[str, int], int]

Pick the reset parameters for a new episode at global actor step.

Parameters:

  • step (int) – The current global actor step.

Returns:

  • tuple[dict[str, int], int] – (parameters, effective_step) where effective_step = max(0, step - activate_at_step). Feeding effective_step to the existing step-based exploration schedule gives each stage its own exploration window. Before any stage activates, the earliest stage is used.

Source