Automatic program generation with genetic network programming using subroutines

Genetic Network Programming with Automatic Program Generation (GNP-APG) is an evolutionary algorithm to generate programs. Genotype-phenotype mapping technology is introduced in this algorithm to create legal programs. With the help of graph-based structures of Genetic Network Programming (GNP), GNP-APG can efficiently generate robust programs to cope with problems. In this paper, the extended algorithm of GNP-APG is proposed which can create a hierarchy program, in other words, a program which contains a main function and subroutines. The proposed method works like Automatic Defined Functions (ADFs) in Genetic Programming (GP). By using subroutines, a complex program can be decomposed to several simple programs which are obtained more easily. Moreover, these subroutines might be called many times, which results in reducing the size of the program significantly. In simulations, different tile-worlds between the training phase and testing phase are used for performance evaluations and the results shows that GNP-APG with subroutines (GNP-APGsr) could have better performances than GNP-APG.