SCHEME: Caching subtrees in genetic programming

Phillip Wong, Mengjie Zhang

Abstract: This paper introduces SCHEME (Subtree Caching using a Hashing for Equivalence MEthod), a method of caching program subtrees while taking into consideration algebraic equivalences between these programs. By using hashing in order to estimate algebraic equivalence between subtrees, we develop a hash table based caching mechanism which is easily integrated with the standard GP system. Experiments are performed on two regression and four classification tasks of varying difficulty. The results suggest that using SCHEME significantly reduces the number of node evaluations performed during the GP runs, which in turn leads to a faster GP training process.