jpountz opened a new pull request, #12446: URL: https://github.com/apache/lucene/pull/12446
Both MAXSCORE and WAND can easily be tuned to perform rank-unsafe optimizations, by skipping doc IDs that are unlikely to make it to the top-k. The main challenge is how to expose this kind of optimization. One approach could consist of artificially increasing the minimum competitive score as suggested in the original WAND paper. The approach I'm considering here is to configure a target evaluation cost, giving the scorer a budget of documents that it can visit and asking it to compute the best hits it can identify with this budget. This draft PR tries to give an idea of how it could look like. It's currently only implemented for our MAXSCORE implementation but could easily be ported to our WAND scorer too. An interesting follow-up could be to integrate this into the timout mechanism, so that `IndexSearcher` would progressively reduce the target cost as the amount of remaining time reduces. I'm interested in gathering feedback on this approach. -- 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
