saealib.TrainingSet#

class saealib.TrainingSet[source]#

Bases: ABC

Strategy object that builds (train_x, train_y) for a surrogate.

Inject an instance into LocalSurrogateManager or GlobalSurrogateManager via their training_set constructor argument.

Methods

`__init__`
`build`	Build and return the training data.
`contract`	Return the training-set contract.

Method Details

abstractmethod TrainingSet.build(archive, population, ctx, candidate_x=None)[source]#

Build and return the training data.

Parameters:

archive (Archive) – Evaluated solutions archive.
population (Population | None) – Current population. Not used by all implementations.
ctx (OptimizationState | None) – Optimization context. Provides comparator, problem.eps_cv, etc. Required by ranking- and comparison-based implementations.
candidate_x (np.ndarray | None) – Centre point for k-NN queries (shape (dim,)). Passed by LocalSurrogateManager for each candidate; None when called from GlobalSurrogateManager.

Return type:

TrainingData

TrainingSet.contract()[source]#

Return the training-set contract.