Algorithms#
Pages summarizing how named algorithms from the literature are reproduced as combinations of saealib components.
When a configuration does not exactly match the theoretical definition, each page states the implementation differences, such as acquisition-function constraints, selection or archive handling, and evaluation order.
Sequential model-based optimization combining Gaussian Process regression with the Expected Improvement acquisition function.
Sequential model-based optimization combining Gaussian Process regression with the Upper Confidence Bound acquisition function.
Exploration-only sequential model-based optimization based solely on the predictive uncertainty of Gaussian Process regression.
Sequential model-based optimization combining an RBF-interpolation surrogate model with a distance constraint from existing evaluated points.
Multi-objective genetic algorithm based on non-dominated sorting and crowding distance. The basis for saealib’s multi-objective comparators.
Multi-objective genetic algorithm combining dominance-based strength and density into a fitness value, with a fixed-size archive.
Multi-objective genetic algorithm that uses reference-direction niche preservation. Primarily targets many-objective optimization, while exposing reference-direction behavior with three objectives.
Steady-state multi-objective evolutionary algorithm that incorporates dominated hypervolume directly into the selection criterion.