rosella icon indicating copy to clipboard operation
rosella copied to clipboard

UMAP and HDBSCAN alternatives

Open jianshu93 opened this issue 1 year ago • 7 comments

Hello Rosella team,

I notice that hdbscan also replies on python but actually there is a beautiful rust implementation (https://github.com/petabi/petal-clustering) and is paralleled when necessary. I am wondering whether it can be used (petal-decomposition also provides PCA).

Thanks,

Jianshu

jianshu93 avatar Aug 19 '22 13:08 jianshu93

Hi Jianshu,

This would be great to use within Rosella, however there is no UMAP rust implementation yet. The rosella algorithm heavily relies on UMAP and HDBSCAN working together back and forth, thus it is currently easier to just keep both the UMAP and HDBSCAN within the separate python component flight. If a rust UMAP implementation appears with a fully fleshed out API then it would be worth re-writing that part of Rosella, but until then we'll have to leave it as it is.

Cheers, Rhys

rhysnewell avatar Aug 23 '22 03:08 rhysnewell