Grammar-Guided Genetic Programming and Automatically Defined Functions

Abstract

Genetic Programming (GP) is a powerful software induction technique that has been recently applied for solving a wide variety of problems. Attempts to extend GP have focussed on applying type restrictions to the language to control genetic operators and to ensure that only valid programs are created. In this sense, the use of context free grammar (CFG) was proposed. This work studies the use of a CFG to define the structure of the initial population and direct crossover and mutation operators. Chameleon, a Grammar-Guided Genetic Programming system (GGGP) is also presented. On a suite of experiments composed of even-parity problems, the performance of Chameleon is compared to traditional GP. Furthermore, the automatic discovery of sub-functions, one of the most important research areas in GP, is also explored. We describe how to use ADFs with GGGP and, using Chameleon, we demonstrate that GGGP has similar results to Koza’s Automatically Defined Functions (ADF) approach.