Introducing a Perl genetic programming system - and can meta-evolution solve the bloat problem?

Abstract

An open source Perl package for genetic programming, called PerlGP, is presented. The supplied algorithm is strongly typed tree-based GP with homologous crossover. User-defined grammars allow any valid Perl to be evolved, including object oriented code and parameters of the PerlGP system itself. Time trials indicate that PerlGP is around 10 times slower than a C based system on a numerical problem, but this is compensated by the speed and ease of implementing new problems, particularly string-based ones. The effect of per-node, fixed and self-adapting crossover and mutation rates on code growth and fitness is studied. On a pi estimation problem, self-adapting rates give both optimal and compact solutions. The source code and manual can be found at http://perlgp.org. © Springer-Verlag Berlin Heidelberg 2003.