I have data that has been used to fit a Bayesian non-parametric mixture model. One of the things I can do with that is sample conditional posterior cluster assignments. Given a sufficiently large sample, I’ve used that to calculate a similarity matrix–that is, a (n x n) matrix of pairwise probabilities of falling into the same cluster.

Using that as a set of edge weights, I want to label emergent clusters. One thing I was looking at is MST–minimum spanning tree clustering, as an algorithm. But every MST algorithm I see implemented seems to be working in parameter space, and using that to calculate the edge weights. I already have the edge weights calculated, and want to use them to build the tree.

Are there any such algorithms implemented?