一种综合的模式聚类方法
A Comprehensive Approach to Mode Clustering
摘要:
模式聚类是一种非参数聚类方法,它使用密度估计器模式的吸引盆来定义聚类。我们提供了几个增强模式聚类:(i)集群分配的软变体,(ii)集群之间的连接性的措施,(iii)选择带宽的技术,(iv)去噪小集群的方法,以及(v)可视化集群的方法。结合所有这些增强给我们一个完整的程序聚类在多元问题。我们还比较了模式聚类与其他聚类方法在几个例子
英文摘要:
Mode clustering is a nonparametric method for clustering that defines clusters using the basins of attraction of a density estimator's modes. We provide several enhancements to mode clustering: (i) a soft variant of cluster assignment, (ii) a measure of connectivity between clusters, (iii) a technique for choosing the bandwidth, (iv) a method for denoising small clusters, and (v) an approach to visualizing the clusters. Combining all these enhancements gives us a complete procedure for clustering in multivariate problems. We also compare mode clustering to other clustering methods in several examples
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。