Leland McInnes

Hmm, not sure why HDBSCAN is so slow then. I would not imagine it would take that long. You may want to try using fast_hdbscan instead -- on a machine...

I think GPU UMAP and fast_hdbscan should work well together. I'm not sure Bayesian optimization will be any better than a well planned grid-search strategy, but it depends on how...

If you really need smaller clusters then ``cluster_extraction_method`="leaf"`` should work for you. You may also want to look at using datashader for plotting/visualization since there is a lot of overplotting...

I'm not sure how to achieve all your desiderata here. You can just run K-Means with k=20 and get something that meets your requirements; that may not exactly be a...

It is difficult to accurately predict runtime internally to the algorithm, so I can't see any obvious way to provide a progress report without at least some non-trivial work. Sorry....

I think, right now, that isn't really going to be possible. The prediction data relies on fast nearest neighbors which (for now) relies on sklearn's KDTrees and BallTrees, and you...

Thanks. I can potentially add this to the FAQ if you like.

One option is the support for sparse precomputed metrics, so you can set ``metric=precomputed`` and supply a scipy sparse matrix of just the distances for the graph you want to...

Exporting from networkX to a sparse graph just gives the adjacency matrix, so each node will only see it's neighbours. If you actually want to consider distances on the graph...

What you want is certainly not available by default. I would suggest you reach out to @JelmerBot, the author of the BranchDetector, to see if you can work out how...