kevindrosendahl commented on issue #12615: URL: https://github.com/apache/lucene/issues/12615#issuecomment-1800243525
> I haven't had a chance to read your branch yet, but hope to soon. Great, thanks! To save you a bit of time, the tl;dr of going from HNSW to vamana is that it's actually fairly simple in the end. For the most part, I just removed all references to levels and changed the [diversity check](https://github.com/kevindrosendahl/lucene/blob/075116396e8af35b0a19a277491be9acab150beb/lucene/core/src/java/org/apache/lucene/util/vamana/VamanaGraphBuilder.java#L376-L404) to incorporate vamana's `alpha` parameter (essentially just adding a second, outer loop). There were a few small other implementation detail changes as well like [pruning down all the way to `M` when passing the overflow threshold instead of just removing one neighbor](https://github.com/kevindrosendahl/lucene/blob/075116396e8af35b0a19a277491be9acab150beb/lucene/core/src/java/org/apache/lucene/util/vamana/VamanaGraphBuilder.java#L360-L369). Then on the storage size, you just [write vectors into the graph instead of into the `.vec` file](https://github.com/kevindrosendahl/lucene/blob/075116396e8af35b0a19a277491be9acab150beb/lucene/core/src/java/org/apache/lucene/codecs/vectorsandbox/VectorSandboxVamanaVectorsWriter.java#L703-L734), and [read from the right offsets to get the vector values as appropriate](https://github.com/kevindrosendahl/lucene/blob/075116396e8af35b0a19a277491be9acab150beb/lucene/core/src/java/org/apache/lucene/codecs/vectorsandbox/VectorSandboxVamanaVectorsReader.java#L598-L684). -- 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
