The Window Algorithm
Douglas Newlands, Damien Brain · International Conference on Data Mining · 2002
Genetic algorithm (GAs) have long been known as effective search techniques for high dimensional spaces. Function finding attempts to discover a mathematical function that adequately describes a set of data. A standard genetic algorithm searches for all coefficients of a function concurrently, but there are difficulties in simultaneously minimising run times and maximizing accuracy. One recent improvement to GA performance is Delta Coding which searches all coefficients in parallel but increases the accuracy as the search progresses. This paper presents another method of tackling the accuracy and speed trade-offs in GAs. Instead of searching for all coefficients at the same time, a sliding window of coefficients is searched. Experiments are performed into fitting functions to data points and results indicate that good perfon-nance can be obtained with small windows. The results indicate that the window algorithm is a useful modification to the standard GA. Some insight into window sizing is presented.