Morph-Fitting. An Effective Technique of Approximation

摘要

An algorithm has been developed and programmed which fits a minimum number of basic patterns, morphs, to a sequence of datapoints. The dependent variable is given as a scalar value; the independent variable is 'distance-like' or 'time-like' and can represent continuous or discrete values or an event-counter. The morphs fitted are an indefinite number of occurrences of trends (straight lines), step functions, and sudden changes (peaks of short duration). Delay functions span over periods characterized by uninterpretable events. A useful by-product of the investigation is a set of optimum decision rules concerning the boundary points between sequential regression functions. (Author)