Adding Program Length Bias to the Lexicase and Tournament Selection Algorithms
Eva Moscovici · 2016
Lexicase selection is a relatively new but promising algorithm for selecting parents to participate in evolving the next generation of programs. Tournament selection is a more established and commonly used algorithm that serves the same purpose. As is the case with many other genetic programming algorithms, the programs generated with involvement of the lexicase and tournament selection algorithms are often large and hard to understand. There has been prior work done related to parsimony and managing bloat in the programs generated using genetic programming. In this paper, I will discuss a way to reduce program size for lexicase and tournament selection algorithms, by incorporating the program length bias into the error for each test case used to test whether an individual is fit enough to participate in the evolution or not.