How well does affinity propagation perform when finding two clusters?
Partitioning N data points into two clusters can be done exactly in time N(N-1)(N-2) by trying out all N(N-1) pairs of possible exemplars and for each pair assigning all other (N-2) data points to their most similar exemplars. Affinity propagation would take NxN time per iteration, but since it is an approximate algorithm, you may be better off using an exact method for two clusters. Alternatively, using the best of a large number of k-centers clustering runs initialized randomly should also lead to a good result. For more than just a few clusters, affinity propagation is more appropriate.
Related Questions
- When the clusters found by affinity propagation are fed into the standard k-centers clustering method, the net similarity sometimes increases (albeit only slightly). Why?
- Is affinity propagation only good at finding a large number of quite small clusters?
- How well does affinity propagation perform when finding two clusters?
