Leonid Boytsov
Leonid Boytsov
@maumueller not sure. Which parameter is it mapped to?
@erikbern @yurymalkov the new bindings seem to release the GIL lock. However, there might be some subtle bug that is not obvious on a first look. I will double check.
Hi @erikbern a somewhat unrelated question. Is the new version of Annoy using 2-means to build trees? The previous stable one (I checked a month ago) was still based on...
Thank you. I am currently fine with it, but if you add more structure in the future, this will be great.
Let me chime in on this, though it's Yura's package. It seems to me that copying data and even indexing it has a negligible cost nowadays compared to the effort...
But Annoy would still be slow if index doesn't fit into RAM?
@yurymalkov if I remember correctly, there was an issue in NMSLIB that complained about it too. However, some experiments showed that fixing the bug didn't help the overall performance. :-)
This works mostly fine for SW-graph: https://github.com/nmslib/nmslib/blob/master/test_batch_app/test_batch_mod.cc Doing something similar for HNSW is a bit more work, but not too much. My understanding is that HNSW will actually be much...
@phdowling sometime, it will. SW-graph in NMSLIB has an experimental deletion feature, but it's not exposed to Python.
No reduction is necessary. In fact, it can be detrimental: Morozov, Stanislav, and Artem Babenko. "Non-metric similarity graphs for maximum inner product search." Advances in Neural Information Processing Systems. 2018.