pyGOURGS - global optimization of n-ary tree representable problems using uniform random global search

Abstract

Global optimization problems are ubiquitous to engineering and the sciences. Many such problems are not amenable to analytical techniques and examinations of some potential solutions for these problems often suggest that hill-climbing algorithms would be unable to navigate the jagged and confusing terrain. Despite this, genetic programming is often applied to these problems in the hopes that it will be able to identify high-quality solutions. We suspect that genetic programming would perform no better than random search, in agreement with the No Free Lunch Theorems (Wolpert, Macready, & others, 1997), and we devised this software to allow us to perform uniform random global search, also known as pure random search, on these problems. The challenge lies in creating a system that enumerates all possible solutions, such that we are then able to randomly select from this space of solutions, giving each solution the same probability of being selected.