I’m pretty sure this has been discussed or even exists somewhere in the roadmap, nonetheless, having a performant 3D clustering engine built in TD would unlock so many very interesting workflows. I know it is no small feat and that many algorithms choke with relatively light datasets, however, here’s a few resources I’ve found in case some will be of help:
Reference implementations of various clustering algorithms:
High-performance HDBSCAN implementation:
Paper discussing a performant ‘G-DBSCAN’ variant:
A benchmark of various methods:
I picture a decimation step happening before clustering in larger datasets.