Analysis of a restricted test case set for a sorting network genetic algorithm

Sorting networks have been of interest to the computer science and evolutionary computing communities for several decades. In this paper we present a measure of test case quality that may have application to other problems in which fitness is based on a large number of test cases. We use this measure to reduce the number of tests used in fitness evaluation as a means of optimizing the speed of a genetic algorithm, although the quality measure itself is expensive. We report the results of several genetic algorithm experiments evolving sorting networks using a restricted number of test cases, and compare those results to similar experiments in the literature.