[GitHub] [lucene] jtibshirani commented on pull request #872: LUCENE-10527 Use 2*maxConn for last layer in HNSW

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

jtibshirani commented on PR #872:
URL: https://github.com/apache/lucene/pull/872#issuecomment-1120061467

   @mayya-sharipova what is the ann-benchmarks set-up that you're using? I found that in order to get comparable results to hnswlib, I had to make sure to adjust `efSearch` to these parameters:
   ```
         run-groups:
           M-4:
             arg-groups:
               - {"M": 32,  "efConstruction": 100}
             query-args: [[1000, 0, 40, 70, 90, 490, 790, 990]]
   ```
   
   First, there is a warm-up run that we throw away. Then we subtract 10 from the other `efSearch` since `KnnGraphTester` adds `k=10` to this value:
   
   ```
     private static TopDocs doKnnVectorQuery(
         IndexSearcher searcher, String field, float[] vector, int k, int fanout) throws IOException {
       return searcher.search(new KnnVectorQuery(field, vector, k + fanout), k);
     }
   ```


-- 
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