kaivalnp commented on code in PR #15836: URL: https://github.com/apache/lucene/pull/15836#discussion_r2957407084
########## lucene/core/src/java/org/apache/lucene/search/AbstractVectorSimilarityQuery.java: ########## Review Comment: Actually, I wonder if we can go one step further and come up with an "expected nodes visited" (like [KNN does](https://github.com/apache/lucene/blob/3ba9a0736636e976677c4c29c351945b23dfee6c/lucene/core/src/java/org/apache/lucene/codecs/lucene99/Lucene99HnswVectorsReader.java#L345)) to bypass approximate search? (something tighter than `-Infinity`) In most cases, would the scores b/w a query vector and document vectors be a normal distribution centered at 0? (at least, for `DOT_PRODUCT` and `MAXIMUM_INNER_PRODUCT`) Then, we could "guess" the `traversalSimilarity` that corresponds to number of docs filtered, and directly use "exact search" if the passed value is <= that number? -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
