benwtrent commented on issue #15427: URL: https://github.com/apache/lucene/issues/15427#issuecomment-3528938618
> Anyway, it made me wonder whether a 2nd phase through all the vectors, with the "benefit of hindsight", might somehow yield improvements (smaller graph, faster/better search-time recall curve)? I think this would be an interesting experiment. Basically using its existing connections as "entry points" and doing the regular insertion logic. Very similar to the new fix up logic added in this PR: https://github.com/apache/lucene/pull/15003 All that said, I am not sure it would be beneficial (as you said). However, hnsw graphs are directional, it is possible there are weird cases where a node A would be near new node B, but never gets the back-link because the new B node doesn't consider node A one of its nearest neighbors. This sort of smells like the "connected components" thing, but likely much cheaper. -- 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]
