bug-gnubg
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Bug-gnubg] Re: Naw..panic or teach me what I am doing wrong !!


From: Jim Segrave
Subject: Re: [Bug-gnubg] Re: Naw..panic or teach me what I am doing wrong !!
Date: Fri, 10 Jan 2003 17:21:12 +0100
User-agent: Mutt/1.4i

On Fri 10 Jan 2003 (16:09 +0000), Joern Thyssen wrote:
> On Fri, Jan 10, 2003 at 01:31:20PM +0100, Jim Segrave wrote
> > On Thu 09 Jan 2003 (17:10 -0500), Gary Wong wrote:
> > I was idly looking at what it would take to write .sgf format files of
> > rollouts. I think it's not too hard to do. 
> 
> I don't think you need to write the .sgf file while doing the rollout. 
> 
> I think it can be accomplished by having a local matchstate and plGame in
> BasicCubefulRollout, and then call AddMoveRecord[*] for each move and
> cube decision (check play.c where there is lot's of examples
> around). At the end, you just call SaveGame from sgf.c.
> You have to set board, cube position etc first. See CommandSavePosition
> for an example.
> 
> This approach should be very generic.
> 
> We can add new moverecords for truncation (either bearoff or just normal
> truncation), and a new moverecord for the rollout result (win, lose +
> variance reduction for this game).

That's a much nicer idea. I like it a lot better than hand
coding. But...
If I do a 1296 game rollout of a mid-match cube decision (I did one
recently, it took 3 days and told me that the experts on gammonline,
not surprisingly, are a lot better at cube handling than I am :-)
you'll end up with a *huge* movelist. If you're writing it on the fly,
you don't have to save things. So you'd probably want to write at
least individual games of a rollout as each one completes or risk
starting to swap heavily while running. Depending on how malloc is
implemented, the move list might not get paged out and left in swap.


-- 
Jim Segrave           address@hidden




reply via email to

[Prev in Thread] Current Thread [Next in Thread]