[GitHub] [lucene] jtibshirani commented on pull request #641: LUCENE-10391: Reuse data structures across HnswGraph#searchLevel calls

[GitBox <gi...@apache.org>]

jtibshirani commented on pull request #641:
URL: https://github.com/apache/lucene/pull/641#issuecomment-1028617830


   This PR contains two changes:
   * Reuse `visited` bit set across calls. For graph construction, switch to `FixedBitSet` instead of `SparseBitSet`.
   * Reuse `candidates` queue across calls.
   
   They both seem to have a small positive impact on indexing. Here's an example using `KnnGraphTester` with `glove-100-angular`:
   
   **Baseline:** 788966 msec to write vectors
   **Only reuse visited set:** 768152 msec to write vectors
   **PR (reuse both visited and candidates):** 760896 msec to write vectors
   
   The PR also shows a small search improvement:
   
   **Baseline** 
   ```
   Recall    QPS
   0.741     2515.710
   0.814     1589.089
   0.924      459.971
   ```
   
   **PR (reuse both visited and candidates)**
   ```
   Recall    QPS
   0.740     2569.156
   0.813     1626.779
   0.925      468.207
   ```


-- 
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-unsubscribe@lucene.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@lucene.apache.org
For additional commands, e-mail: issues-help@lucene.apache.org