msokolov commented on PR #14226:
URL: https://github.com/apache/lucene/pull/14226#issuecomment-2660180526

   I don't believe 16 is "special" except in the sense that it happens to be a 
sweet spot is this context.  We expect that as we increase that per-segment 
factor we will get increased recall because it is the equivalent of the old 
"fanout" parameter we used to have - it expands the beam of the search while 
keeping the number of results returned fixed.  This is different from "fanout" 
because that was a global setting, and this one scales with the relative size 
of the segment.
   
   What I'd like to understand is whether there is a fixed value for this sweet 
spot, or whether it changes with {data, graph construction parameters, 
something else}. It seems clear that my idea that it is a probabilistic thing 
is incorrect since 16 standard deviations is just crazy.  But we can at least 
look at it empirically.


-- 
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