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?
Affinity propagation seems to be good at resolving the “major battles” between many subsets of data points competing to form good clusters, but when it’s finished leaves a few players strewn about the battle-field without tidying them up properly. Sometimes, putting the output of affinity propagation through a couple iterations of k-centers clustering or another greedy algorithm will “polish up” the solution and since doing this is computationally cheap, we recommend it.
Related Questions
- I ran affinity propagation with the default parameter settings and the plot of the net similarity continued to oscillate until the maximum number of iterations was reached. What should I do?
- 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?
- Can affinity propagation be viewed as just a good way to initialize standard methods, such as k-centers clustering?