コールバック

Layer 2: Compose

イベント

saealib.Event	Base class for all callback events.
saealib.RunStartEvent	Fired once when the optimization run starts.
saealib.RunEndEvent	Fired once when the optimization run ends.
saealib.InitialEvaluationStartEvent	Fired after initial sampling, before initial archive evaluation.
saealib.InitialEvaluationEndEvent	Fired after initial archive evaluation, before archive sorting.
saealib.GenerationStartEvent	Fired at the beginning of each generation.
saealib.GenerationEndEvent	Fired at the end of each generation, before yielding the context.
saealib.SurrogateStartEvent	Fired before surrogate-based candidate scoring.
saealib.SurrogateEndEvent	Fired after surrogate-based candidate scoring.
saealib.AcquisitionStartEvent	Fired before acquisition scoring, inside AcquisitionStage.
saealib.AcquisitionEndEvent	Fired after acquisition scoring succeeds, inside AcquisitionStage.
saealib.PostCrossoverEvent	Fired after crossover and repair.
saealib.PostMutationEvent	Fired after mutation and repair.
saealib.PostAskEvent	Fired after the full ask step (post crossover and mutation).
saealib.PostSurrogateFitEvent	Fired after the surrogate model is fitted.
saealib.PostEvaluationEvent	Fired after true evaluation of selected candidates.

コールバックマネージャー

saealib.CallbackManager

Manages event handlers.

組み込みハンドラー

saealib.logging_generation(event)

Log the state at the start of each generation.

For single-objective problems, logs the best objective value determined by the comparator. For multi-objective problems, logs the size of the first Pareto front and the per-objective value ranges of that front.

パラメータ:

event (GenerationStartEvent) -- The generation-start event.

戻り値の型:

None

saealib.logging_generation_hv(reference_point)

Return a handler that logs the hypervolume per generation.

Computes the hypervolume of the first Pareto front in the archive with respect to the given reference point (minimization convention).

パラメータ:

reference_point (np.ndarray) -- Reference (nadir) point, shape (n_obj,). Each component should be strictly greater than the best achievable value per objective.

戻り値:

A handler compatible with CallbackManager.register.

戻り値の型:

Callable[[GenerationStartEvent], None]

サンプル

>>> optimizer.cbmanager.register(
...     GenerationStartEvent,
...     logging_generation_hv(np.array([1.1, 1.1]))
... )

チェックポイント保存

saealib.CheckpointCallback

Callback that saves checkpoints at regular generation intervals.