Crossover in grammatical evolution: The search continues

Abstract

Grammatical Evolution is an evolutionary automatic programming algorithm that can produce code in any language, requiring as inputs a BNF grammar definition describing the output language, and the fitness function. The utility of crossover in GP systems has been hotly debated for some time, and this debate has also arisen with respect to Grammatical Evolution. This paper serves to continue an analysis of the crossover operator in Grammatical Evolution by looking at the result of turning off crossover, and by exchanging randomly generated blocks in a headless chicken-like crossover. Results show that crossover in Grammatical Evolution is essential on the problem domains examined. The mechanism of one-point crossover in Grammatical Evolution is discussed, resulting in the discovery of some interesting properties that could yield an insight into the operator's success.