RLGP: An Efficient Method to Avoid Code Bloating on Genetic Programming

Code bloating presents a serious problem in scaling GP to larger and more difficult problems. The studies involving code bloating in Genetic Programming (GP) are mainly concerned with preventing bloated individuals from producing on the population. GP using a size or depth limit (LGP) is a common approach to battle bloat, but LGP is not ideal in size control and searching efficiency. In this paper, besides extended the concept of bloated individual in LGP, and the concept of candidate crossover points set is presented. A new variants of LGP, named RLGP, which adds some restrictions in genetic operations (crossover, swap, and mutation), is proposed. RLGP introduces Candidate Crossover Points Set (CCPS) into crossover operations. Finally, in even 3, 4, and 5-parity problem, strongly positive results are reported regarding both size control and searching efficiency.