[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-gnubg] The match and game data structure
From: |
Jim Segrave |
Subject: |
Re: [Bug-gnubg] The match and game data structure |
Date: |
Mon, 3 Jul 2006 13:05:56 +0200 |
User-agent: |
Mutt/1.4.2.1i |
On Mon 03 Jul 2006 (11:38 +0100), Ian Shaw wrote:
>
> Jim Segrave Sent: 30 June 2006 16:11
> >
> > Along with this, move the analysis and rollout contexts, if
> > any analysis or rollout is done to the match header. I think
> > few analysed matches have more than a single analysis context
> > or rollout context used,
>
> I don't think this is correct (but maybe I misunderstand your point). I know
> people who will perform a quick 0-ply rollout, and then maybe follow it up
> with a 2-ply rollout on the most likely candidates.
>
> Similarly, I often analyse a whole match on 0-ply, then look at some plays at
> 2-ply or higher.
>
> Tutor mode is another example where one might use the tutor at 0-ply, but you
> might sometimes ask for higher ply analysis.
Sure, but then all the analysis is either 0 ply evalutation, 2 ply
evaluation or one of two rollout contexts. In all, any data on a
particular move uses only one of 4 possible contexts. In a 21 point
match, we currently store perhaps 800 or so analysis contexts and
output those 800 contexts to the .sgf file. I'm proposing you store
and output only 4 contexts, then 800 moves with a simple index to the
analysis context which is applicable.
--
Jim Segrave address@hidden