There was a bug in my implementation of tf(float freq). It was always 1, even when the frequency was 0. This caused a discrepancy between the score and the debug info score - it seems like in the debug all the dismax fields got score, while in the "real" score fields with document frequency 0 were eliminated.
-Yuri On Wed, Aug 27, 2008 at 7:56 PM, Chris Hostetter <[EMAIL PROTECTED]>wrote: > > : It seems like the debug information is using the custom similarity as it > : should - the bug isn't there. > : I see in the explain information the right tf value (I modified it to be > 1 > : in my custom similarity). > : The numbers in the explain seem to add up and make sense. > : Is it possible that the score itself is wrong (the one that I get from > fl)? > > the score in the doclist is by definition the correct score - the > debug info follows a differnet code path and sometimes that code path > isn't in sink with the the actual searching/scoring code for differnet > query types (although i was pretty confident that the test i added to > Lucene-Java a whle back tested this for anything you can see in Solr > without getting into crazy contrib Query classes) > > it would help if you could post: > > 1) the full debugQuery output from a query where you see this > disconnect, showing the all query toString info, and the score > explanations > 2) the corrisponding scores you see in the doclist > 3) some more details about how your custom similarity works (can you post > the code) > 4) info on how you've configured dismax and what request params you are > using (the output from using echoParams=all would be good) > > > > > -Hoss > >
