Analysis of genetic diversity through population history

The idea that diversity in the population of a genetic algorithm affects the algorithm’s search efficiency is widely accepted. However, little is known about the amount of node level diversity present in Genetic Programming (GP) runs. In this paper, we introduce several techniques for measuring the diversity of a population based on the genetic history of the individuals. We then apply these measures to the genetic histories of several runs of four different problems. The results of this analysis show that a surprisingly small amount of diversity is present in the final population of a GP run. We conclude by suggesting a variety of other potential applications of these measures.