Population diversity and inheritance in genetic programming for symbolic regression

Abstract In this work we aim to empirically characterize two important dynamical aspects of GP search: the evolution diversity and propagation inheritance patterns. Diversity is calculated at genotypic phenotypic levels using efficient similarity metrics. Inheritance information obtained via a full genealogical record as directed acyclic graph set methods for extracting relevant Advances in processing power enable our approach handle previously infeasible sizes millions arcs vertices. To more comprehensive analysis employ three closely-related but different evolutionary models: canonical GP, offspring selection age-layered population structure. Our reveals that relatively small number ancestors are responsible producing majority descendants later generations, leading loss. We show across five benchmark problems each configuration characterized by rates loss patterns, support idea new problem may require unique solve optimally.