benwtrent commented on PR #15003: URL: https://github.com/apache/lucene/pull/15003#issuecomment-3502101118
Thank you for all the work @Pulkitg64 OK, I think your plan from this comment (50% check) is the way to go for now. My intuition is that the relative threshold is likely better, but we would need to make the percentage lower (like 25%, or 20%, like your other benchmarks show). However, the 50% shows really nice improvements already, so lets move forward with that. As for "Reconstruct completely", yes, lets focus on 40% or fewer deleted total docs? Does that align with your benchmarking results? If so, I think we have our two thresholds and we can move forward with this PR. Thank you for this practical, and powerful improvement for vector indexing! -- 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]
