A Rank Transformation Can Improve Sequential Parameter Optimization

Wessing, S.1, a; Wagner, T.2, b

Fakultät für Informatik, Lehrstuhl für Algorithm Engineering (LS 11), Technische Universität Dortmund, Otto-Hahn-Straße 14, 44227 Dortmund
Institut für Spanende Fertigung, Technische Universität Dortmund, Baroper Str. 301, 44227 Dortmund

a) simon.wessing@tu-dortmund.de; b) wagner@isf.de


Over recent years, parameter tuning of Evolutionary Algorithms (EAs) is attracting more and more interest. In particular, the sequential parameter optimization (SPO) framework for model-assisted tuning procedures resulted in established parameter tuning algorithms. Most variants of SPO apply Kriging for modeling the fitness landscape and, thereby, finding an optimal parameter configuration on a limited budget of EA runs. In this work, we enhance the SPO framework by introducing transformation steps before the aggregation and before the modeling. We empirically show that a rank transformation of the data improves the mean performance of SPO and is superior to other transformations, such as the Box-Cox and the logarithmic transformation.


Data Transformation, Design and Analysis of Computer Experiments, Kriging, Sequential Parameter Optimization (SPO)


In: Proceedings of the Workshop on Experimental Methods for the Assessment of Computational Systems (WEMACS 2010), 11.9. 2010, Krakau, Poland, Bartz-Beielstein, T.; Chiarandini, M.; Paquete, L.; Preuss, M. (Hrsg.), S. 32-46