harishankar-gopalan commented on issue #11354: URL: https://github.com/apache/lucene/issues/11354#issuecomment-1290668547
> Update: Sorry for delay, I am still working on this but got a little side tracked with other work. > > > > Hi @harishankar-gopalan, yes what currently happens is the graph gets reconstructed from scratch. In https://github.com/apache/lucene/pull/11719, I am working on selecting the largest graph from a segment and using that to initialize the newly created segment's graph. Posted above are my initial benchmark results. However, I am running into some issues where the recall is slightly lower with the test setup and the merge time is higher. I have been debugging a little bit why this is happening, but have not yet make progress. I am going to take another try at it this week or next week. Hi @jmazanec15 thanks for the update. Are there any public stats available for the current segment merges for HNSW based graph indexes in Lucene ? To be more clear any performance benchmarks to compare the Lucene segment merges for Documents with and without KnnVectorFields indexed as a HNSW Graph. If you are aware of any initial benchmarks that you are using as reference, would be great full if you could share links to those if possible. -- 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: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
