Clustering Mixtures of Discrete Distributions: A Note on Mitra's Algorithm