Evolving High-Level Imperative Program Trees with Strongly Formed Genetic Programming

Abstract

We present a set of extensions to Montana's popular Strongly Typed Genetic Programming system that introduce constraints on the structure of program trees. It is demonstrated that these constraints can be used to evolve programs with a naturally imperative structure, using common high-level imperative language constructs such as loops. A set of three problems including factorial and the general even-n-parity problem are used to test the system. Experimental results are presented which show success rates and required computational effort that compare favourably against other systems on these problems, while providing support for this imperative structure.

locoGP

“…If we do not encourage diversity, we may end up with a population where the majority of programs are very similar to the seed. New generations of individuals are Approach Representation Improvement Fitness Metric locoGP Java (AST) Performance Bytecode Operations Langdon [17], Petke [28] C++ (Statement) Performance, Specialisation Line Count Arcuri [2], White [41] Java-like (AST) Performance Simulated CPU Cycle Walsh & Ryan (Paragen) [30,31], Parallelisation Instructions Parallel Programs Functionality Chennupati (MCGE) [4] Orlov (FINCH) [25] Java (Byecode) Functionality Error Count Castle [3] Java-like (AST) Functionality Error Count O'Cinnéide [5], Simons [33] Java (Refactoring Patterns) Quality (e.g. elegance) Software Metrics Table 1: Feature Comparison of Improvement Approaches.…”

“…Many GP systems include their own parsers and interpreters. However, a wide range of programs are written in general-purpose imperative languages such as Java and so recent work has turned to improving such programs [2,3,26,37].…”

Event-based graphical monitoring in the EpochX genetic programming framework

“…While it is built sensibly to avoid performance issues, speed is not the main goal. Several papers used EpochX in their experimental work [2,5,6,13,12].…”

Evolving program trees with limited scope variable declarations

“…By supporting the evolution of variable declarations, the aim is to lighten this burden without excessively degrading performance. It has previously been described how the Strongly Formed Genetic Programming (SFGP) variant of GP can be used to enforce a high-level imperative structure on evolved program trees [2]. In this paper, we present a series of modifications to SFGP that allow it to support node-types that can declare new limited scope variables.…”