Measuring the ability of lexicase selection to find obscure pathways to optimality
Jose Guadalupe Hernandez, Alexander Lalejini, Charles Ofria · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2022
This Hot-off-the-Press paper summarizes our recently published work, "An Exploration of Exploration: Measuring the Ability of Lexicase Selection to Find Obscure Pathways to Optimality," published as a chapter in Genetic Programming Theory and Practice XVIII [3]. In evolutionary search, selection schemes drive populations through a problem's search space, often trading off exploitation with exploration. Indeed, problem-solving success depends on how a selection scheme balances search space exploitation with exploration. In [3], we introduce an "exploration diagnostic" that measures a selection scheme's ability to explore different pathways in a search space. We use our exploration diagnostic to investigate the exploratory capacity of lexicase selection and several of its variants: epsilon lexicase, down-sampled lexicase, cohort lexicase, and novelty lexicase. We verify that lexicase selection out-explores tournament selection, and we demonstrate that lexicase selection's ability to explore a search space is sensitive to the ratio between population size and the number of test cases used for evaluating candidate solutions. We find that relaxing lexicase selection's elitism with epsilon lexicase can further improve search space exploration. Additionally, we find that both down-sampled and cohort lexicase---two methods of applying random subsampling to test cases---substantially degrade lexicase's exploratory capacity; however, cohort partitioning better preserves exploration than down-sampling. Finally, we find evidence that the addition of novelty-based test cases can degrade lexicase selection's exploratory capacity.