Fig. 3

Average clustering errors for Gaussian means and inverse-Wishart covariances model. The first row corresponds to n=20, and the errors are shown for different missing probabilities. The second row corresponds to n=70 and missing probability of 0.15, where the errors are plotted vs. the Hamming distance threshold used to define the reference partitions in Pseed. a n1=10,n2=10, b n1=12,n2=8, c n1=35,n2=35,miss. prob.=0.15, d n1=42,n2=28,miss. prob.=0.15