Examples using precomputed distance matrix for clustering

[celiafish]

Hi, I want to use clustering methods with precomputed distance matrix (NxN). I found DBSCAN has "metric" attribute but can't find examples to follow. Can you please help. Examples for other clustering methods are also very helpful. Thanks.

[amueller]

what part are you unclear about?
use metric="precomputed" and provide a as X a precomputed similarity matrix.
See http://scikit-learn.org/dev/auto_examples/manifold/plot_mds.html for an example using MDS.

[jnothman]

no, it's not feasible algorithmically for k means

[isukrit]

Can we change the type of distance to cosine similarity? Thanks!

[jnothman]

no. We have a PR in the works for K medoid which is a related algorithm that can take an arbitrary distance metric. Try it out: #7694. K means needs to repeatedly calculate Euclidean distance from each point to an arbitrary vector, and requires the mean to be meaningful; it cannot work with a metric of your choice.