Analyzing Module Usage in Grammatical Evolution

Being able to exploit modularity in genetic programming (GP) is an open issue and a promising vein of research. Previous work has identified a variety of methods of finding and using modules, but little is reported on how the modules are being used in order to yield the observed performance gains. In this work, multiple methods for identifying modules are applied to some common, dynamic benchmark problems. Results show there is little difference in the performance of the approaches. However, trends in how modules are used and how "good" individuals use these modules are seen. These trends indicate that discovered modules can be used frequently and by good individuals. Further examination of the modules uncovers that useful as well as unhelpful modules are discovered and used frequently. The results suggest directions for future work in improving module manipulation via crossover and mutation and module usage in the population.