In this paper we describe experimental results using Winnow and Weighted-Majority based algorithms (two algorithms highly studied in the theoretical machine learning literature) on a calendar scheduling domain. We show that these algorithms can be quite competitive practically, outperforming the ID3-based approach currently in use by the Calendar Apprentice system in terms of both accuracy and speed, on a large dataset. In addition we show how Winnow can be applied to achieve a good accuracy/coverage tradeoff and we explore issues that arise such as concept drift. We also provide a theoretical analysis of the Winnow variant that we use (which is one especially suited to conditions with string-valued classifications) and an analysis of a policy for discarding predictors in Weighted-Majority that allows it to speed up as it learns.
展开▼
