saealib.AccuracyEvaluator#

class saealib.AccuracyEvaluator(metrics=None)[source]#

Bases: ABC

Abstract base for surrogate accuracy evaluators.

Subclasses implement different evaluation strategies (k-fold CV, LOO, held-out). All share the same evaluate() interface so they can be injected into GlobalSurrogateManager interchangeably.

Parameters:

metrics (list[SurrogateAccuracyMetric] or None) – Metrics to compute. Defaults to SpearmanCorrelation, RMSE, R2Score.

Methods

__init__

evaluate

Compute accuracy metrics for surrogate on the given training data.

Method Details

AccuracyEvaluator.__init__(metrics=None)[source]#
Parameters:

metrics (list[SurrogateAccuracyMetric] | None)

abstractmethod AccuracyEvaluator.evaluate(surrogate, train_x, train_y)[source]#

Compute accuracy metrics for surrogate on the given training data.

Parameters:

  • surrogate (Surrogate) – Surrogate instance to evaluate.

  • train_x (np.ndarray) – Training inputs. shape: (n_samples, n_features)

  • train_y (np.ndarray) – Training targets. shape: (n_samples, n_obj)

Return type:

SurrogateAccuracy