OK, I see it now. I was thinking I was scoring that individual term
at that position, but if I had read the Scorer documentation better
instead of assuming I knew what it meant I wouldn't have had this
problem, as it clearly says it advances the document.
I will try to fix it by the end of this weekend.
Thanks for the review.
-Grant
On Apr 19, 2007, at 2:27 PM, Michael Busch wrote:
Grant Ingersoll wrote:
OK, I need to take a step back, Michael, b/c I thought I
understood your original comment, but I went to make the change
and am no longer sure.
By "first term position" are you referring to multiple terms per
position or do you mean the same term in different positions?
When I implemented the BTQ (BoostingTermQuery) I modeled it pretty
much off of the SpanTermQuery (STQ) which I felt had very similar
functionality, other than having to load the payload.
Doesn't the next() method on the BoostingSpanScorer take care of
moving through the various positions that the term appears at,
whereupon it loads the payload at the position? Could you write
up a patch to the test to demonstrate?
Thanks,
Grant
Grant,
I mean the case when the same term has multiple positions in a
document. In BoostingSpanScorer.next() you call super.next() (from
SpanScorer), which calls SpanScorer.setFreqCurrentDoc(). This
method iterates through all spans for the current doc via
TermSpans.next(). So TermSpans.next() is the actual method which
calls TermPositions.nextPosition(). This means when SpanScorer.next
() returns it has iterated through all positions of that doc
already. Then you load the payload, which means that you only get
the payload of the first term position of the next (the wrong!)
document in the term's posting list.
Your testcase does not show this behavior, because the term you
search for only appears once at most in each document. And since
all payloads of the term you search for have the same value, the
testcase doesn't fail even though it loads the payloads of the
wrong documents for scoring.
- Michael
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
--------------------------
Grant Ingersoll
Center for Natural Language Processing
http://www.cnlp.org
Read the Lucene Java FAQ at http://wiki.apache.org/jakarta-lucene/
LuceneFAQ
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
