Simple Implementation of Genetic Programming by Column Tables

Abstract

Simple implementation of genetic programming by makinguse of the column tables is discussed. Implementationsof Koza's genetic programming in compiled languages areusually not most efficient when crossover is applied.If chromosomes are directed acyclic graphs, moreefficient than rooted trees both in memory requirementas well as in evaluation time of chromosome, thencrossover requires traversing the data structures andtheir preliminary analysis. Column tables inherentlycode directed acyclic graphs, the implementation ofcrossover is simple and needs neither traversing norchecking of integrity of resulting data structures andshould be therefore more efficient. Stochastictransformation operation mutation is also easilydefined. Column tables can represent graphs withseveral output nodes and may be used e.g. foroptimization of feed-forward neural networks. Simpleillustrative examples of symbolic regression based onthe column tables are presented.