[FEA] Take advantage of MST in hierarchical clustering #2727
Labels
? - Needs Triage
Need team to review and classify
feature request
New feature or request
inactive-90d
Minimum spanning trees come up in hierarchical clustering is to enable a single-linkage clustering and the ability to draw dendrograms like Scipy's hierarchy package, which also enables HDBSCAN.
The kernels are being implemented in RAFT since it could also be used un cuGraph in the long run : rapidsai/raft#52
Solution
Parallel Baruvka algorithm supporting disconnected components. It is similar to Louvain in the sense it starts from all vertices as seeds and aggregate vertices into super vertices based on edge weights.
relevant papers :
Alternatives considered
Specific solutions for KNN-graphs. While this allows some optimizations, a more generic solution would have more benefits for the whole RAPIDS platform
The text was updated successfully, but these errors were encountered: