对于基于划分的聚类算法随机选取初始聚类中心导致初始中心敏感,聚类结果不稳定、集群效率低等问题,提出一种基于MapReduce框架和改进的密度峰值的划分聚类算法(based on MapReduce framework and improved density peak partition clus...对于基于划分的聚类算法随机选取初始聚类中心导致初始中心敏感,聚类结果不稳定、集群效率低等问题,提出一种基于MapReduce框架和改进的密度峰值的划分聚类算法(based on MapReduce framework and improved density peak partition clustering algorithm,MR-IDPACA)。首先,通过自然最近邻定义新的局部密度计算方式,将搜索样本密度峰值点作为划分聚类算法的初始聚类中心;其次针对算法在大规模数据下运行时间复杂,提出基于E2LSH(exact Euclidean locality sensitive hashing)的一种分区方法,即KLSH(K of locality sensitive hashing)。通过该方法对数据分区后结合MapReduce框架并行搜寻初始聚类中心,有效减少了算法在搜索初始聚类中心时的运行时间;对于MapReduce框架中的数据倾斜问题,提出ME(multistage equilibrium)策略对中间数据进行多段均衡分区,以提升算法运行效率;在MapReduce框架下并行聚类,得到最终聚类结果。实验得出MR-IDPACA算法在单机环境下有着较高的准确率和较强的稳定性,集群性能上也有着较好的加速比和运行时间,聚类效果有所提升。展开更多
Binary digit representation of partial sums for random variables has been investigated, and a good upper bound of moments of maximum partial sums for random variables has been reduced by using this representation. As ...Binary digit representation of partial sums for random variables has been investigated, and a good upper bound of moments of maximum partial sums for random variables has been reduced by using this representation. As an applications, stability and strong law of large numbers have been discussed. Many known classical results have been refined.展开更多
文摘对于基于划分的聚类算法随机选取初始聚类中心导致初始中心敏感,聚类结果不稳定、集群效率低等问题,提出一种基于MapReduce框架和改进的密度峰值的划分聚类算法(based on MapReduce framework and improved density peak partition clustering algorithm,MR-IDPACA)。首先,通过自然最近邻定义新的局部密度计算方式,将搜索样本密度峰值点作为划分聚类算法的初始聚类中心;其次针对算法在大规模数据下运行时间复杂,提出基于E2LSH(exact Euclidean locality sensitive hashing)的一种分区方法,即KLSH(K of locality sensitive hashing)。通过该方法对数据分区后结合MapReduce框架并行搜寻初始聚类中心,有效减少了算法在搜索初始聚类中心时的运行时间;对于MapReduce框架中的数据倾斜问题,提出ME(multistage equilibrium)策略对中间数据进行多段均衡分区,以提升算法运行效率;在MapReduce框架下并行聚类,得到最终聚类结果。实验得出MR-IDPACA算法在单机环境下有着较高的准确率和较强的稳定性,集群性能上也有着较好的加速比和运行时间,聚类效果有所提升。
文摘Binary digit representation of partial sums for random variables has been investigated, and a good upper bound of moments of maximum partial sums for random variables has been reduced by using this representation. As an applications, stability and strong law of large numbers have been discussed. Many known classical results have been refined.
