kevindrosendahl commented on issue #12615:
URL: https://github.com/apache/lucene/issues/12615#issuecomment-1793186041

   Hey @benwtrent and all, just wanted to let you know that I'm experimenting 
some with different index structures for larger than memory indexes.
   
   I have a working implementation of Vamana based off the existing HNSW 
implementation (with vectors colocated with the adjacency lists in storage) in 
[this branch](https://github.com/kevindrosendahl/lucene/tree/vamana2), which 
I'm currently working on integrating scalar quantization with, and a 
benchmarking framework 
[here](https://github.com/kevindrosendahl/java-ann-bench) which can run various 
Lucene and JVector configurations.
   
   I don't have many numbers to share yet besides the fact that the Vamana 
graph implementation in a single segment seems competitive while in memory with 
Lucene HNSW (single segment) and JVector on small data sets. For 
glove-100-angular from ann-benchmarks (k=10):
   ```
   lucene_hnsw_maxConn:16-beamWidth:100_numCandidates:150
        average recall 0.8227400000000001
        average duration PT0.000968358S
           index size: 529M
   
   
jvector_vamana_M:16-beamWidth:100-neighborOverflow:2.0-alpha:1.2_pqFactor:0-numCandidates:100
        average recall 0.8242200000000001
        average duration PT0.00124232S
           index size: 703M
   
   lucene_sandbox-vamana_maxConn:32-beamWidth:100-alpha:1.2-_numCandidates:100
        average recall 0.82553
           average duration PT0.000940756S
           index size: 554M
   ```
   
   I plan on testing vamana without PQ, vamana with PQ (a la DiskANN), as well 
as SPANN. Happy to collaborate with anyone interested.


-- 
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

Reply via email to