lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Grant Ingersoll <gsing...@apache.org>
Subject Re: Whither Query Norm?
Date Fri, 20 Nov 2009 18:08:23 GMT

On Nov 20, 2009, at 11:19 AM, Jake Mannix wrote:

> I should add in my $0.02 on whether to just get rid of queryNorm() altogether: 
> 
>   -1 from me, even though it's confusing, because having that call there (somewhere,
at least) allows you to actually do compare scores across queries if you do the extra work
of properly normalizing the documents as well (at index time).

Do you have some references on this?  I'm interested in reading more on the subject.  I've
never quite been sold on how it is meaningful to compare scores and would like to read more
opinions.

>   And for people who actually do machine-learning training of their per-field query boosts,
this is pretty critical.
> 
>   -jake
> 
> On Fri, Nov 20, 2009 at 8:15 AM, Jake Mannix <jake.mannix@gmail.com> wrote:
> The fact Lucene Similarity is most decidely *not* cosine similarity, but strongly resembles
it with the queryNorm() in there, means that I personally would certainly like to see this
called out, at least in the documentation.
> 
> As for performance, is the queryNorm() called ever in any loops?  It's all set up in
the construction of the Weight, right?  Which means that by the time you're doing scoring,
all the weighting factors are already factored into one?  What's the performance issue which
would be saved here?
> 
>   -jake
> 
> 
> On Fri, Nov 20, 2009 at 7:56 AM, Grant Ingersoll <gsingers@apache.org> wrote:
> For a long time now, we've been telling people not to compare scores across queries,
yet we maintain the queryNorm() code as an attempt to do this and the javadocs even promote
it.  I'm in the process of researching this some more (references welcomed), but wanted to
hear what people think about it here.  I haven't profiled it just yet, but it seems like a
good chunk of wasted computation to me (loops, divisions and square roots).  At a minimum,
I think we might be able to refactor the callback mechanism for it just as we did for the
collectors, such that we push of the actual calculation of the sum of squares into Similarity,
instead of just doing 1/sqrt(sumSqs).  That way, when people want to override queryNorm()
to return 1, they are saving more than just the 1/sqrt calculation.  I haven't tested it yet,
but wanted to find out what others think.
> 
> Thoughts?
> 
> -Grant
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-dev-unsubscribe@lucene.apache.org
> For additional commands, e-mail: java-dev-help@lucene.apache.org
> 
> 
> 



Mime
View raw message