Adopting lexicase selection for michigan-style learning classifier systems with continuous-valued inputs
Alexander R. M. Wagner, Anthony Stein · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2021
In this work, two variants of lexicase selection are examined for their performance improvement potential in the context of XCS variants allowing for continuous-valued inputs. We furthermore propose a niche-specific mode of operation for lexicase selection when adopted for utilization in XCS, implemented by a dedicated classifier experience storage. To evaluate the impact of lexicase selection on XCS' classification and regression capabilities, our proposed modified variants are tested across various continuous-valued single-step tasks, including both typical toy problems and real-world datasets related to the agricultural domain as well as regression problems.