You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@lucene.apache.org by GitBox <gi...@apache.org> on 2021/09/01 20:58:17 UTC

[GitHub] [lucene] mayya-sharipova edited a comment on pull request #267: LUCENE-10054 Handle hierarchy in graph construction and search

mayya-sharipova edited a comment on pull request #267:
URL: https://github.com/apache/lucene/pull/267#issuecomment-910744639


   @jtibshirani  Thank a lot for your review. I've tried to address in 6a05951772cc72a1530f3fd863d906dc0e3bef88
   
   
   > I wonder if we should really allow ml to be configurable. I think other HNSW implementations choose this using a heuristic and don't allow it to be set. This would let us drop the alternate code paths for graphs with only one level and remove old naming like numSeed.
   
   I took an inspiration from the `nmslib` library where this parameter [is configurable](https://github.com/nmslib/nmslib/blob/master/similarity_search/src/method/hnsw.cc#L203).   Also it allows us to have a flat graph (`sw-graph` from `nmslib`) for cases where we want it.  Not sure though if we still want to keep flat graph, happy to continue the discussion.
   
   ---
   
   
   > ./gradlew test --tests TestHnswGraph.testSearchWithAcceptOrds -Dtests.seed=E7C78E0C128E2469
   
   
   This test fails because it ends up marking many expected results docs (from 0 to 9) as deleted which don't get retrieved.
   I've modified `createRandomAcceptOrds(int startIndex, int length)` to add `startIndex` from which we want to start marking documents deleted.
   
   The reason this test was not failing on the main branch is because there we use `numSeed=5` and return 5 results instead of 5 and their sum always ends up < 75. 
   
   
   


-- 
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