Gaussian process surrogate models for the CMA-ES

Additional results

authors: Lukáš Bajer, Zbyněk Pitra, Jakub Repický, Martin Holeňa

BBOB output data for the S-CMA-ES (GP model, 5 generations life-length) [tgz]
BBOB output data for the fixed DTS-CMA-ES (alpha = 0.05, double population) [tgz]
BBOB output data for the self-adaptive DTS-CMA-ES (beta = 0.3) [tgz]
BBOB post-processed results -- BBOB postprocessing graphs of 13 algorithms for budget equal to best f-value for 100 FE/D of the best2009 algorithm (the default for --expensive setting is 50 FE/D)

Contents

Black-box optimizers comparison

Medians of the distances to the optima of 24 COCO/BBOB benchmarks in dimensions 2,3,5,10,20 for 13 black-box optimizers. Medianswere calculated across 15 independent instances for each algorithm and are shown in log_10 scale

Aggregated algorithm comparison

Median distances to the benchmarks optima averaged over all 24 BBOB benchmark functions in dimensions 2,3,5,10,20. The log_10 of the median distances were for each function linearly scaled to [-8, 0] before calculating the averages.