On the importance of specialists for lexicase selection

Abstract

Lexicase parent selection filters the population by considering one random training case at a time, eliminating any individual with an error for the current case that is worse than the best error of any individual in the selection pool, until a single individual remains. This process often stops before considering all training cases, meaning that it will ignore the error values on any cases that were not yet considered. Lexicase selection can therefore select specialist individuals that have high errors on some training cases, if they have low errors on others and those errors come near the start of the random list of cases used for the parent selection event in question. We hypothesize here that selecting such specialists, which may have high total error, plays an important role in lexicase selection’s observed performance advantages over error-aggregating parent selection methods such as tournament selection, which select specialists less frequently. We conduct experiments examining this hypothesis, and find that lexicase selection’s performance and diversity maintenance degrade when we deprive it of the ability to select specialists. We also conduct experiments with a form of tournament selection that has been modified to allow for the selection of specialists, and find that it performs better than ordinary tournament selection, but not as well as lexicase selection. These findings, and other data that we present here, help explain the improved performance of lexicase selection compared to tournament selection, and suggest that specialists help drive evolution with lexicase selection toward global solutions.