Time analysis of standard evolutionary algorithms as software programs

Abstract

This article presents a study which characterizes the computational efficiency behavior of a standard EA as a software program. The study analyzes the effects of some implementation decisions regarding memory utilization (dynamic vs. static, local vs. global) and the generation of pseudorandom numbers, on the execution time of the resulting EA. The experimental analysis allows us to conclude that significant improvements in efficiency can be gained by applying simple guidelines on how to best program the EA.