Learning to assess the quality of genetic programs using cultural algorithms

We explore solution generalizability, bloat, and effective length as examples of software quality issues and measurements that are useful in the analysis of genetic programming (GP) solution programs. The total program size of GP solution programs can be partitioned into effective program size and several types of excess code for which, following Angeline, we use the term “bloat” (P.J. Angeline, 1998). We define several types of bloat: local, global, and representational. We use a cultural algorithm tool called the Metrics Apprentice to explore the relationships between the generalizability of the programmed solution, program size, and the effect of three GP processes purported to reduce bloat