An analysis of exchanging fitness cases with population size in symbolic regression genetic programming with respect to the computational model

Symbolic regression using genetic programming is an ideal algorithm for automatically determining an otherwise unknown functional relationship between a set of inputs and outputs. More complex problems in this area typically require a larger amount of training epochs to exemplify the relationship. Previous work has shown that using a strategy of trading off higher population sizes with lower data sample sizes in the early generations yields better results. In this paper we take a closer look at this tradeoff policy and how it applies to the computation model, as well as examine some of the parameter settings.