[Xapian-discuss] Re: (Very)Wrong number in get_matches_estimated

[Olly Betts]

On Mon, Jan 22, 2007 at 02:21:16PM -0800, Rafael SDM Sierra wrote:
> Sorry, it's a stupid mail, I'm using match decider, and have some documents
> that don't pass on it, so, the number will be really different

Actually, it's not so stupid.

It's harder to make a good estimate in the presence of a matchdecider,
but currently we just return the same estimate we would in the absence
of the matchdecider.

I think we ought to scale this estimate by the ratio of documents the
matchdecider liked to the number it tested (or something like that).
It seems that would generally improve the estimate in this case and
we do something similar when collapsing or using a percentage cutoff.

Does this patch help:

http://oligarchy.co.uk/xapian/patches/xapian-matcher-better-mdecider-estimates.patch

Cheers,
    Olly