msokolov commented on issue #12256: URL: https://github.com/apache/lucene/issues/12256#issuecomment-1529152279
There's no requirement that these scores have any particular values; the only requirement is that they sort the same as the dot-product score. There is no reason to expect that floating point dot product scores would equal dot product scores for bytes. The reason the scaling is the way it is is to avoid clipping the max values. if both vectors a, b are all -128 (say), then a*b = 128^2 * dim = 2^14 * dim. We want [min, max] score to scale to [-0.5, 0.5] which is why this is the way it is. It also says this is in the comments. Maybe you could suggest better wording to make that clearer? -- 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
