loading page

Cross Validated Racing : a new approach to tuning of Metaheuristics
  • Ago Lemos
Ago Lemos
Universidade Federal do Maranhão - UFMA
Author Profile

Abstract

The process of tuning a metaheuristic can be responsible for its success or failure in the face of the problem which it is intended to solve, for which reason great efforts are made to the elaboration of techniques that allow the fine adjust of these algorithms. However, the absence of a formal tuning definition, as well as a variety of implementations and types of metaheuristics, results in tuning dependent on a specialist or by trial and error, which is inefficient and makes it difficult to validate results. This article presents a tuning method named Cross-Validated Racing (CVR) that has problem and set of instance independence, making tuning results more reliable and generalist. For a validation of the CVR, the BRKeCS was developed, a hybrid algorithm that is the advantage of BRKGA ( biased random-key genetic algorithm) with ECS (Evolutionary Clustering Search) and applied to solve Permutacional Flow Shop Scheduling problems. Continue...

Keywords: Tuning, Irace, Permutation Flow Shop Scheduling, BRKeCS, Evolutionary Clustering Search, Cross Validated Racing Approach