saealib.MeanPrediction

class saealib.MeanPrediction(weights=None, reference=None)[source]

Bases: AcquisitionFunction

Acquisition function based on predicted mean value (exploitation).

For single-objective problems, returns the predicted mean directly. For multi-objective problems, returns a weighted scalarization of the predicted mean.

A higher score indicates a more promising candidate. The sign convention follows the weight: use a negative weight for minimization (e.g., weights=np.array([-1.0])) so that lower objective values yield higher scores.

Parameters:

  • weights (np.ndarray or None) – Weights for scalarizing multi-objective predictions. shape: (n_obj,). If None, uses the first objective only.

  • reference (Any)

Methods

__init__

compute_reference

Return fixed reference if set, otherwise None.

score

Compute scores based on predicted mean.

Method Details

MeanPrediction.__init__(weights=None, reference=None)[source]
Parameters:

  • weights (ndarray | None)

  • reference (Any)

MeanPrediction.compute_reference(archive)[source]

Return fixed reference if set, otherwise None.

Parameters:

archive (Archive)

Return type:

Any

MeanPrediction.score(prediction, reference=None)[source]

Compute scores based on predicted mean.

Parameters:

  • prediction (SurrogatePrediction) – Surrogate predictions. prediction.mean shape: (n_samples, n_obj)

  • reference (Any) – Not used. Accepted for interface compatibility.

Returns:

Scores. shape: (n_samples,)

Return type:

np.ndarray