Re: Maxscore for document pruning in case of custom scorer and collector

--000000000000895350064a90f856
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Thanks Michael for your reply, as i mentioned i have already implemented
all the api calls you mentioned in my custom term scorer , like
getMaxScore, and also i am setting minimumCompScore on my top level scorer
instance from my custom collector, i debugged and i see all these working.
The only problem for which i am seeking help here is about part of my
scoring is dependent on doc itself , like its popularity, rating etc stored
in docValues, also part of the score also depends on query context , these
are scored separately from term score and exactly for these i am facing
issue in setting correct max score , right now only term score contributes
to maxscore api call but i somehow need to also include other scoring
components as well.


*Thanks and Regards,*
*Vimal Jain*

On Fri, 6 Feb, 2026, 7:23=E2=80=AFpm Michael Sokolov, <[email protected]> =
wrote:

> There are a handful of optional methods your Scorer must implement in
> order to effectively participate in these skipahead algorithms. I
> forget the names - something like maxScore(int upTo)? Oh yes and
> advanceShallow(). Hopefully the javadocs for these methods give some
> guidance? Also if your Scorers have child scorers they are responsible
> for propagation of these calls.  Beyond that, it can be tricky to tell
> how well your skipping is working, but you could try instrumenting
> your calls to know how many docs they are able to skip due to max
> score?
>
> Regarding doc-values there is no equivalent scorer-skipping that I
> know of, although I think there has been some discussion of it.  One
> possible alternative would be to index your doc-scores as custom term
> frequencies on a known token
>
> On Fri, Feb 6, 2026 at 3:15=E2=80=AFAM Vimal Jain <[email protected]> wrot=
e:
> >
> > Hi Guys,
> > I am looking for advice on how to make best use of maxscore/wand etc fo=
r
> > document pruning ( so as to speed up query execution overall). I have
> > custom scorers and collectors. Also my scoring consists of 2 parts ,
> > termScore * documentBoost. I can provide safe UB in getMaxScore() in my
> > scorer but not sure how to do the same for document level boost which
> > involves reading docValues.
> > Any advice or reference to common practice would be appreciated.
> > Also is there any general documentation around maxscore/wand/block-max =
?
> >
> > *Thanks and Regards,*
> > *Vimal Jain*
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [email protected]
> For additional commands, e-mail: [email protected]
>
>

--000000000000895350064a90f856--