Control structures in linear and stack-based Genetic Programming

Genetic Programming, or GP, has traditionally used prefix trees for representation and reproduction, with implicit flow control. The different clauses (the evaluation condition, the if and else sections, etc.) are all subtrees of the flowcontrol node. Linear and stack-based representations, however, require explicit nodes to define the extent of the control structures. This paper introduces a stack-based technique for correct control structure creation and crossover, and discusses its implementation issues in linear and stack-based GP. A set of flowcontrol nodes is presented, and examples given for evolving an artificial ant on the Santa Fe and Los Altos Trails, with and without looping constructs.