This Article Full Text (PDF) References Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Adenso-Díaz, B. Articles by Laguna, M. Search for Related Content

Fine-Tuning of Algorithms Using Fractional Experimental Designs and Local Search

Escuela Superior de Ingenieros Industriales, Campus de Viesques, Universidad de Oviedo, 33204-Gijón, Spain
Leeds School of Business, University of Colorado, Boulder, Colorado 80309-0419
adenso{at}epsig.uniovi.es
laguna{at}colorado.edu

Researchers and practitioners frequently spend more time fine-tuning algorithms than designing and implementing them. This is particularly true when developing heuristics and metaheuristics, where the "right" choice of values for search parameters has a considerable effect on the performance of the procedure. When testing metaheuristics, performance typically is measured considering both the quality of the solutions obtained and the time needed to find them. In this paper, we describe the development of CALIBRA, a procedure that attempts to find the best values for up to five search parameters associated with a procedure under study. Because CALIBRA uses Taguchi’s fractional factorial experimental designs coupled with a local search procedure, the best values found are not guaranteed to be optimal. We test CALIBRA on six existing heuristic-based procedures. These experiments show that CALIBRA is able to find parameter values that either match or improve the performance of the procedures resulting from using the parameter values suggested by their developers. The latest version of CALIBRA can be downloaded for free from the website that appears in the online supplement of this paper at http://or.pubs.informs.org/Pages.collect.html.

Subject classifications: parameter setting; Taguchi design of experiments; heuristic search.
History: Received January 1999; revision received May 2000; revision received February 2001; revision received December 2002; revision received June 2004; revision received December 2004; accepted December 2004.