Anzeige
Erschienen in:
2020 | OriginalPaper | Buchkapitel
An Experimental Study of Operator Choices in the \((1+(\lambda ,\lambda ))\) Genetic Algorithm
verfasst von : Anton Bassin, Maxim Buzdalov
Erschienen in: Mathematical Optimization Theory and Operations Research
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
Abstract
We study the influence of the particular choice of operators on the running time of the recently proposed \((1+(\lambda ,\lambda ))\) genetic algorithm. In particular, we consider three choices for the mutation operator, six choices of the crossover operator, three strategies for sampling population sizes based on non-integer parameter values, and four choices of what to do when the best mutant is better than the parent.
We test all these 216 configurations on four optimization problems and in three adjustment flavours: the fixed \(\lambda \), the unlimited self-adjustment of \(\lambda \) and the logarithmically capped one. For each of these configurations, we consider both the default values of the hyperparameters and the ones produced by the
irace parameter tuning tool.The result of our experimental analysis showed that there exists a configuration that is robust on linear functions and is roughly two times faster compared to the initially proposed one and 12% faster on the OneMax problem compared to one of the similar previous studies. An even more robust configuration exists, which has a slightly worse performance on OneMax but is better on satisfiability problems. Such configurations can be the default choices for the practical evaluation of the \((1+(\lambda ,\lambda ))\) GA.