Oversearching and layered search in empirical learning

摘要: When learning classifiers, more extensive search for rules is shown to lead lower predictive accuracy on many of the leal-world domains investigated. This counter-intuitive re suit particularly relevant recent system methods that use risk-free pruning achieve same outcome as exhaustive search. We propose an iterated method commences with greedy extending its scope at each Iteration until a stopping criterion satisfied. layered often found produce theories are accurate than those obtained either or moderately, beam