jbellis commented on PR #12421: URL: https://github.com/apache/lucene/pull/12421#issuecomment-1646504742
Parallel code with one thread will build the same graph as serial code. (This is validated by unit tests.) With multiple threads, parallel code errs on the side of potentially adding "extra" connections. So recall should be equal or greater and that is what I see with the 1M SIFT dataset. Serial code: 92.0% recall over 5 runs Parallel code built with 8 threads: 92.4% over 5 runs On Mon, Jul 10, 2023 at 3:54 PM Mayya Sharipova ***@***.***> wrote: > Great work! > Have you compared the recall of the parallel graph with the serially built > graph (for example using ann-benchmarks)? > > — > Reply to this email directly, view it on GitHub > <https://github.com/apache/lucene/pull/12421#issuecomment-1629016815>, or > unsubscribe > <https://github.com/notifications/unsubscribe-auth/AAAKJLUSCMPNEWFIOJLGPALXPQCQJANCNFSM6AAAAAA2ATDZEA> > . > You are receiving this because you authored the thread.Message ID: > ***@***.***> > -- Jonathan Ellis co-founder, http://www.datastax.com @spyced -- 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
