Re: [Bug-gnubg] include 'MET used' in rollout results (export) ?

[Robert-Jan Veldhuizen]

At 23:32 8/14/2003 +0200, you wrote:
> On Thu, 14 Aug 2003 06:43:39 +0000, Joern Thyssen wrote:
> > On Wed, Aug 13, 2003 at 11:20:14PM +0200, Robert-Jan Veldhuizen wrote
> > > Just a short, simple suggestion: would it be a good idea to let gnubg 
> also
> > > include the MET being used for a rollout result (export) ? Sometimes 
> this
> > > can make a significant difference in the results one gets. And maybe 
> also
> > > include this for evaluation results?
> > >
> > > I was also wondering whether the MET being used is stored 
> 'internally' and
> > > in f.i. a SGF file, and if so, for the file (match, session) as a 
> whole or
> > > seperately for each "move"? The latter would be the most complete and
> > > versatile, but perhaps it's overkill? I have no idea how gnubg is 
> handling
> > > this at the moment.
> >
> > This is somewhat difficult due to the flexibility gnubg allows. If you
> > could only choose among a fixed set of METs then we just need to store
> > the number of the one used, but gnubg allows you to choose an
> > arbitrary MET. So in order to store which one used, we need to score the
> > entire MET! Each MET is 1-10kb, so adding it to all moves will require
> > either lots of work or lots of memory.
>
> What about hashing the known ones and saving a checksum only?

Something like that seems a good idea to me. Especially when comparing 
rollout results with others, knowing which MET was used is sometimes 
essential and I really miss that feature right now with GNUBG.

> And
> maybe it doesn't have to be for every single move.

Personally I'd be happy if the MET used for analysis was stored on a match 
or session level, but maybe a good compromise would be to do it on the game 
level. Perhaps some users sometimes like to switch MET's at certain match 
scores, f.i.

I hope this can somehow be implemented oin GNUBG.

Cheers,


--
Robert-Jan (Zorba on FIBS)