msokolov commented on PR #15208: URL: https://github.com/apache/lucene/pull/15208#issuecomment-3572268374
Looks like good progress! Excellent speedup in merge times. I do worry a bit about the recall drop, and I wonder if we can somehow have a benchmark that lets us compare merge time results for similar recalls? EG maybe we increase M slightly for candidate? Also "I modified the algorithm a bit so that it does not depend on previous node being added to the graph when adding the rest of nodes that are outside of join set." makes me wonder if this might be part of the problem with recall? I'm not sure exactly what you meant (haven't reviewed the code), but it sounds as if we don't wait for all of the "covering set" to be indexed before beginning to add the remainder of the nodes? If we're still relying on seeding the query with nodes in the existing graph, but some of them are not in the existing graph ... not sure what that will do! -- 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]
