Clustering Geometric Data Streams

Using recent knowledge in data stream clustering we present a modified approach to the facility location problem in the context of geometric data streams. We give insight to the existing algorithm from a less mathematical point of view, focusing on understanding and practical use, namely by computer graphics experts. We propose a modification of the original data stream k-median clustering to solve facility location which is the case when we a priori do not know the number of clusters in the input data. Like the original, the modified version is capable of processing millions of points while using rather small amount of memory. Based on our experiments with clustering geometric data we present suggestions on how to set processing parameters. We also describe how the algorithm handles various distributions of input data within the stream. These findings may be applied back to the original algorithm. CR Categories: I.5.3 [Computing Methodologies]: Pattern Recognition—Clustering; I.3.5 [Computing Methodologies]: Computer Graphics—Computational Geometry and Object Modeling