Package description

GLIS - Global and preference-based optimization of expensive black-box functions using inverse distance weighting and radial basis function surrogates.

glis

GLIS is a package for finding the global (GL) minimum of a function that is expensive to evaluate, possibly under constraints, using inverse (I) distance weighting and surrogate (S) radial basis functions.

The package implements two algorithms:

GLIS: Global optimization based on inverse distance weighting and radial basis function surrogates, a method for global optimization of a function f that is possibly expensive to evaluate. The algorithm is based on the following paper:

[1] A. Bemporad, "Global optimization via inverse weighting," 2019. Submitted for publication. Available on arXiv:1906.06498

GLISp: Global optimization of a function f whose value cannot be evaluated but, given two points x,y, it is possible to query whether f(x) is better than f(y) (preference-based optimization). The algorithm is based on the following paper:

[2] A. Bemporad, D. Piga, "Active preference learning based on radial basis functions," 2019. Submitted for publication. Available on arXiv:1909.13049.

This software is distributed without any warranty. Please cite the above papers if you use this software.

Download

Latest version:

GLIS V2.2 - January 22, 2020
(Added support for Genetic Algorithm method from MATLAB Global Optimization Toolbox. Minor change in GLISp demo - MATLAB version only)


Previous versions:

GLIS V2.1 - October 18, 2019
(file and function names changed to GLIS/GLISp)

GLIS V2.0.1 - September 29, 2019

GLIS V2.0 - September 28, 2019
(added preference-based optimization functions. Preference-based Bayesian optimization by D. Piga)

GLIS V1.1.1 - September 2, 2019

GLIS V1.1 - August 3, 2019
(Python version improved by M. Forgione)

GLIS V1.0.2 - July 6, 2019

GLIS V1.0.1 - July 4, 2019

GLIS V1.0 - June 15, 2019




Last update: October 18, 2019