Abstract
Evolving strategies is one of the fields where GA has gained much acceptance. Games [1], [2] and robotics [3] are rich areas. Fitness of a candidate chromosome is calculated relative to other candidates by allowing it to compete against a certain fitness testbed selected from an infinitely large search space. Two questions arise: How to select the test-bed and how can we tell that an optimum candidate relative to a certain test-bed is "the optimum"? In this paper the results of a solid test performed to study the optimality of evolved strategies using three different test-beds are given. The amount of dynamics of the test-beds varies from totally static to semi-static composed of coevolving members of the generation, and totally dynamic, generated at random for each new generation. Optimality of a test-bed is estimated in terms of efficiency as well as processing effort needed. First results showed that as long as diversity is maintained all three types gave quite similar results w.r.t. fitness scores as well as cross-validation tournaments. However, when performance analysis was conducted the test-bed using changing competitors proved by far better needing remarkably less processing although additional overhead for randomly creating competitors at each generation has been accounted for.