New options for hoeffding trees

摘要: Hoeffding trees are state-of-the-art for processing high-speed data streams. Their ingenuity stems from updating sufficient statistics, only addressing growth when decisions can be made that guaranteed to almost identical those would by conventional batch learning methods. Despite this guarantee, still subject limited lookahead and stability issues. In paper we explore Option Trees, a regular tree containing additional option nodes allow several tests applied, leading multiple as separate paths. We show how control in order generate mixture of paths, empirically determine reasonable number then evaluate spectrum variations: single trees, bagged trees. Finally, investigate pruning.We on some datasets pruned smaller more accurate than tree.