Lucene: Status=Open

2.6 hours ago 20 comments 0 votes 0 watches HoustonPutman, alessandrobenedetti, benwtrent, david-sitsky, hpvd, krickert, msokolov, uschindler, vigyasharma

Description It would be nice to support multiple values in a Knn vector field. This must be compatible with both the Exact and Approximate Nearest Neighbor search. There are two sides to the coin: 1) Index time support - allowing to add in the indexing data ...

benwtrent 2.6 hours ago:	> Not sure I understand option 3. Are you thinking that graph has different types of edges b/w documents based on diff. similarity ... So if you were using max similarity you would follow one path, but if you wanted avg. similarity, ...
vigyasharma 2.8 hours ago:	@benwtrent I like the idea of having documents be vertices in the graph, with an API that let's ... It would have an indexing time similarity used during graph build (e.g. max-sim), and could allow ... I suppose this is option 2 you describe above?

#13003: Exploring GPU based kNN vector search

2.8 hours ago 2 comments 0 votes 0 watches chatman, yupeng9

Description Through this issue, I wish to explore integrating NVIDIA's kNN indexing and search ... Through our initial benchmarks/prototypes, we found it a lot faster than our HNSW based search. Along the way, we shall add more details of our experiments through this issue.

yupeng9 2.8 hours ago:	This is very interesting work. We saw Milvus published article on how GPU accelerates vector search, which looks like a game ... .
chatman 1.8 months ago:	As an initial proof of concept integration to evaluate performance, we put together a repository. ... Will add more benchmarks with multithreaded search on CPU. Comments, thoughts and questions welcome.

#13358 PR: Backport to 9x: Reduce duplication in taxonomy facets; always do counts #12966

3.2 hours ago 2 comments 0 votes 0 watches mikemccand, stefanvodita

Annoying things I had to do to backport 12966: 1. TaxonomyFacetCounts stores the counts twice because it extends IntTaxonomyFacets. The correct way would be to extend TaxonomyFacets, but I couldn't make that change, like I could ...

mikemccand 3.2 hours ago:	Thanks @stefanvodita! Make sure to update the main lucene/CHANGES.txt also to move the entry down to 9.11.0 section.
stefanvodita 2.9 days ago:	Despite the "annoying" bits in the description, I don't expect this backport to be controversial, ... I plan to wait over the weekend and then merge.

#13363 PR: Make IndexInput#prefetch take an offset.

3.2 hours ago 1 comments 0 votes 0 watches jpountz, rmuir

This makes IndexInput#prefetch take an offset instead of being relative to the current position. This avoids requiring callers to seek only to call prefetch().

rmuir 3.3 hours ago:

I like this much better from api perspective: it closer maps to madvise() and to me is more ... Especially as the current PRs out there to use it (terms and postings) are using absolute FPs and ...

#13360 PR: Call ArrayUtil.copyArray instead of ArrayUtil.copySubArray for full array copy.

8.5 hours ago 1 comments 0 votes 0 watches bruno-roustant, dsmiley

ArrayUtil provides copySubArray() as replacement of Array.copyOf and Array.copyOfRange (with better ... Unfortunately ArrayUtil does not provide copyArray(), so it is weird for callers to have to call a ... This PR adds ArrayUtil.copyArray(), and replaces the calls to ArrayUtil.copySubArray(array, 0, ...