Erik Bernhardsson

I don't have super strong feelings about this, but think this is arguably a "feature" not a bug. Vectors that are zero should simply be ignored and never returned. We...

yeah i don't have super strong feelings, might make sense to remove them. feel free to submit a pull request if you want to. otherwise i can do it but...

ok – sounds great! i'll keep an eye out for your other PR. and regarding removing null vectors i may or may not do it in the next couple of...

You mean the data in `results/` right? Or do you mean the granular metrics (i.e. the stuff that's plotted just in numerical form)? Can make that available although I'm a...

Ok, sounds good...as @maumueller pointed out they should be published with #57

Hi @yurymalkov – sorry I forgot about this one. This is really weird, not sure what's going on. I'll take a look. @maumueller any idea?

I have a theory what's going on here. Is nmslib/hnsw using threads under the hood, even if queried with a single thread? The time consumption is measured inside each thread....

Btw one thing we could do to eliminate the Python overhead is to swap out the thread pool for a process pool...however that could cause issues for any algorithm that...

Are you sure it's not using eg a multithreaded linear algebra package or something similar? It's possible that it's the Python overhead then, like GIL locking or something similar. Have...

How do you feel about swapping out the thread pool for a process pool? It's a one line change and will support single/threaded libraries alike