Lexicase selection is one of the most successful parent selection methods in evolutionary computation. However, it has the drawback of being a more computationally involved process and thus taking more time compared to other selection methods, such as tournament selection. Here, we study a version of lexicase selection where test cases are combined into several composite errors, called summed batch lexicase selection; the hope being faster but still reasonable success. Runs on some software synthesis problems show that a larger batch size tends to reduce the success rate of runs, but the results are not very conclusive as the number of software synthesis problems tested was small.
"Summed Batch Lexicase Selection on Software Synthesis Problems,"
Scholarly Horizons: University of Minnesota, Morris Undergraduate Journal: Vol. 7:
1, Article 3.
Available at: https://digitalcommons.morris.umn.edu/horizons/vol7/iss1/3