James Melville

This indeed doesn't have a big effect on any dataset I have tried.

`optimize_graph_layout` was added to `uwot` which will do this.

@ratheraarif a few months ago at https://github.com/lmcinnes/umap/issues/58#issuecomment-1253795090 you asked for the ability to combine the fuzzy simplicial sets/graphs in R. The development version of uwot now has `simplicial_set_intersect` and `simplicial_set_union`...

This got added so I am going to close this

M1 Macs have been added to the CRAN checks and these tests are currently failing in the same way. See (although this will change if/when the next version of uwot...

Hi @jianshu93, HNSW is faster than Annoy for queries, but the benchmarks don't usually include the cost of building the index, which is usually slower than for Annoy. This makes...

Thank you for the pointer to granne, which I hadn't seen. Unfortunately, I don't program in Rust currently, and don't know what the state of interfacing such code with R...

> For hnsw, as long as the small world graph is constructed, nearest neighbors for each datapoint in database is already there and just extract them (e.g., a small world...

Whoever attempts this (and it's not likely to be me), this seems like something that should be pursued at the C++ level in hnswlib itself. Probably not something I would...

@jianshu93 can you open a new issue for this? In that issue, please provide both the command you ran and the output (by adding `verbose = TRUE`) if necessary. In...