Leland McInnes

Sorry for the delay; I am travelling and have very limited internet access at times. I agree that this is slower -- we are potentially recomputing quantities repeatedly. It does...

Okay, tests are added, and I'm happy to entertain options with regard to the computation approach. Is there anything more I can do at this point?

I'm unfortunately very busy with other things for a while. I'll try to get back to this in a week or two.

I completely forgot about this to be honest, and I think it has been so long it now has a lot of conflicts to resolve. I don't foresee this happening...

PyNNDescent was built with knn-graph construction in mind as a use-case. That means that the index construction is actually split into two parts, with the second part optional. The first...

I know that kgraph does. My understanding is that the various NGT algorithms do (and cache it). The bottom layer of an HNSW index is an approximation to one, and...

I'll try to have a PR soon. Hopefully it will cover both versions of what the sparse data could/should look like. I'm trying to get NGT working (the C++ has...

I think the new sparse format makes sense. My personal preference would be to use a ``scipy.sparse`` matrix format, as that is pretty standard for sparse data. Notably it also...

In the current implementation it needs to be a distance supported by sklearn's BallTree or KDTree structures, which Levenshtein is not. You can, however, use sparse precomputed distance matrices (using...

I think you would want to use an approximate nearest neighbor library than supports levenshtein distance. I believe hnsw in nmslib meets those criteria.