Queue-based Genetic Programming

1. STACK-BASED GENETIC PROGRAMMING Genetic Programming, or GP, has traditionally used tree-based techniques for representation and reproduction. The most widely used crossover method is subtree crossover, and the majority of the alternatives in published literature are variants thereof. However, if the trees are manipulated in their prefix or postfix form, other approaches exist that preserve the syntactic integrity of the participating parent trees. Stack based Genetic programming, introduced by Perkis in [3], represents programs as lists of nodes of functions or terminals that consume their inputs from a stack and place their outputs on a stack. These implementations, including the early work of Bruce, Stoffel and Spector, [1], [5] and later [4], do not try to preserve the stack correctness of the individuals in the population, but rather rely on the evaluation framework to identify any stack underflow or overflow. In contrast, in GP with stack-correct (Forth) crossover, introduced by Tchernev in [6] and [7], the crossover operators manipulate the post order representation of the program tree. Because the crossover points are chosen to have compatible stack depths, no malformation is possible. If the initial population is stack-correct (no individuals have underflow, and the final stack depth equals the desired number of outputs), it is guaranteed that all individuals produced by using stack-correct crossover will be stack-correct.