OPTISIA: An Evolutionary Approach to Parameter Optimisation in a Family of Point-Set Pattern-Discovery Algorithms

Abstract

We propose a genetic algorithm (GA), OPTISIA, for efficiently finding optimal parameter combinations when running OMNISIA [15], a program that implements a family of analysis and compression algorithms based on the SIA point-set pattern discovery algorithm [20]. The GA, when given a point-set representation of a piece of music as input, runs OMNISIA multiple times, attempting to evolve a combination of parameter values that achieves the highest compression factor on the input piece. When evaluated on two musicological tasks, the system consistently selected well-performing parameters for Forth’s algorithm [6] compared to combinations found in published evaluations on the same musicological tasks.