msokolov commented on PR #12191: URL: https://github.com/apache/lucene/pull/12191#issuecomment-1492966278
Thanks for raising these interesting possible directions. I spent some time tinkering with ivf/pq-like solution - implement simple k-means, indexed using normal postings and doc-values and ran a bunch of tests. I found that I was able to get high recall without pq., at much lower indexing cost than hnsw, but with a substantially higher query cost. Then I tried working on pq to improve the speed, but was never able to get decent recall with any configuration that I tried. It seems that others have had good results though. Maybe we should start some kind of sandbox branch for trying these out. I'd be happy to share what I have, although the results are not great -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
