[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kinosearch
Subject:    [KinoSearch] subclassing term scorers
From:       nate () verse ! com (Nathan Kurz)
Date:       2007-07-18 21:12:49
Message-ID: 65d3176c0707182106r3b7fb6f2xb3f118e15a01b463 () mail ! gmail ! com
[Download RAW message or body]

On 7/18/07, Marvin Humphrey <marvin@rectangular.com> wrote:
>
> Well put.  You've persuaded me.
>

Wonderful!  I'm sure there are details left, but I think we can safely
wait until later to discuss them, and concentrate on positions.

I said I'd send you more thoughts tonight, but I'm not going to have
time.  Things came up (happy things:  an opportunity to go out tuna
fishing leaving tomorrow morning at 2:30 am) so I'm not going to get
to it until Friday.

The gist of my thoughts was that:

1) I think we can get away with a single flat array of positions
rather than a complex structure.

2) All the Booleans can should be able to pass positions by reference,
so no memory troubles foreseen there.

3) PhraseScorer and the like end up reducing the number of positions
so shouldn't be a problem.

4) I haven't been able to think of any geometric space issues.

5) Geometric time:  that could be a problem, which is why I'm
concerned with avoiding unnecessary work on unused branches.

I think your test case is a good one to think about.

Possibly disappearing until Friday,

Nathan Kurz
nate@verse.com

[prev in list] [next in list] [prev in thread] [next in thread]