On sampling error in genetic programming

Abstract The initial population in genetic programming (GP) should form a representative sample of all possible solutions (the search space). While large populations accurately approximate the distribution solutions, small tend to incorporate sampling error. This paper analyzes how size GP affects error and contributes answering question populations. First, we present probabilistic model expected number subtrees for initialized with full, grow, or ramped half-and-half. Second, based on our frequency model, that estimates given size. We validate models empirically show that, compared smaller sizes, recommended sizes largely reduce measured fitness values. Increasing even more, however, does not considerably Last, recommend some widely used benchmark problem instances result low A at initialization is necessary (but sufficient) reliable since lowering means overall random variations are reduced. Our results indicate severe GP, making obtain allows practitioners determine minimum so lower than threshold, confidence level.