A stepwise evolution of functions

A Genotype-Phenotype mapping in most Genetic Programming (GP) systems uses a predefined and rigid grammar definition. This method has been successful in producing the required solution. However, it can only be used to solve a limited set of problems. In this paper, a Teachable GP (TGP) system is proposed. An external GP system evolves a complete computer program, which acceptable solution is then added automatically to the existing grammar definition as a function and made available to the TGP system. This dynamic grammar definition allows for a more complex program to be generated, solving more complex problems. Experiments are performed to compare performances between GP without the added function, GP with a user-defined function and GP with the evolved function and results shows that GP with an evolved function is comparable to the GP with user-defined function and outperformed GP without function.