Sensible initialisation in chorus

  • Authors:

  • Conor Ryan;R. Muhammad Atif Azad

  • Affiliations:

  • Department of Computer Science And Information Systems, University of Limerick, Ireland;Department of Computer Science And Information Systems, University of Limerick, Ireland

  • Venue:
  • EuroGP'03 Proceedings of the 6th European conference on Genetic programming
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

One of the key characteristics of Evolutionary Algorithms is the manner in which solutions are evolved from a primordial soup. The way this soup, or initial generation, is created can have major implications for the eventual quality of the search, as, if there is not enough diversity, the population may become stuck on a local optimum. This paper reports an initial investigation using a position independent evolutionary algorithm, Chorus, where the usual random initialisation has been compared to an approach modelled on the GP ramped half and half method. Three standard benchmark problems have been chosen from the GP literature for this study. It is shown that the new initialisation method, termed sensible initialisation maintains populations with higher average fitness especially earlier on in evolution than with random initialisation. Only one of the benchmarks fails to show an improvement in a probability of success measure, and we demonstrate that this is more likely a symptom of issues with that benchmark than with the idea of sensible initialisation. Performance seems to be unaffected by the different derivation tree depths used, and having a wider pool of individuals, regardless of their average size, seems enough to improve the performance of the system.