PROBI: A Heuristic for the probabilistic k-median problem

Abstract

We develop the heuristic PROBI for the probabilistic Euclidean k-median problem based on a coreset construction by Lammersen et al. Our algorithm computes a summary of the data and then uses an adapted version of k-means++ (Arthur and Vassilvitskii, 2007) to compute a good solution on the summary. The summary is maintained in a data stream, so PROBI can be used in a data stream setting on very large data sets. We experimentally evaluate the quality of the summary and of the computed solution and compare the running time to state of the art data stream clustering algorithms.