A Parallel Joinless Algorithm for Co-location Pattern Mining Based on Group-Dependent Shard

摘要

Spatial co-location patterns, whose instances are frequently located together in geography, are particularly valuable for discovering spatial dependencies. Since its inception, lots of co-location pattern mining algorithms have been developed, but the computational cost remains prohibitively expensive with large data size. In this work, we propose to parallelize joinless algorithm on MapReduce framework. Our approach partitions computation in such a way that each machine independently executes joinless algorithm to finish a group of mining tasks. Such partitioning eliminates computational dependencies and reduces communication cost between machines. Moreover, a novel pruning technique is suggested to improve mining performance. The experimental results on synthetic and real-world data sets show that the parallel joinless algorithm is efficient and scalable.