Abstract

This paper described edit distance which is so far unexploited by the GP community despite its suitability for investigating the "step size" of GP operators. The paper has not answered the ultimate question. In fact, the data from crossover distance, BI-POOL distance and BI-POP distance is disappointing with respect to revealing clear relationships between distance and run performance. Instead, it simply offers insight and some degree of confirmation of intuitive interpretations of what happens during a GP run. My conjecture is that edit distance has much more to offer. This attempt did not live up to expectation possibly because it tried to relate macroscopic data such as average population size, average fitness and mean crossover distance. Instead, microscopic data, for example parental fitness, parental average size and edit distance may prove more insightful because of its specificity. Correlations within the microscopic scope potentially will prove more predictive of run performance. runs. References

Citations