关键词: MapReduce, 粗糙集, 知识约简, 数据并行, 任务并行

Abstract:

Knowledge reduction for massive datasets has attracted many research interests in rough set theory. Classical knowledge reduction algorithms assume that all datasets can be loaded into the main memory of a single machine, which are infeasible for large-scale data. Firstly, this paper analyzes the parallel computations among classical knowledge reduction algorithms. Then, in order to compute the equivalence classes and attribute significance on different candidate attribute sets, it designs and implements the Map and Reduce functions using data and task parallelism. Finally, it constructs the parallel algorithm framework model for knowledge reduction using MapReduce, which can be used to compute a reduct for the algorithms based on positive region, discernibility matrix or information entropy. The experimental results demonstrate that the proposed parallel knowledge reduction algorithms can efficiently process massive datasets on Hadoop platform.

Key words: MapReduce, rough set, knowledge reduction, data parallel, task parallel