[computer-go] New version of Crazy Stone

[Don Dailey]

Ok,  I blundered.  I see your point.   If it's a win for either color a
2 ply search will find it, otherwise it must be a draw.

- Don

  



On Sun, 2006-06-04 at 22:34 +0200, Erik van der Werf wrote:
> On 6/4/06, Don Dailey <drd at mit.edu> wrote:
> >
> > On Sun, 2006-06-04 at 16:51 +0200, Erik van der Werf wrote:
> > > On 6/4/06, Don Dailey <drd at mit.edu> wrote:
> > > > So if you wanted to build a 32 man endgame database it would be nice
> > > > if you could do it 1 bit per position.   But you could only represent 2
> > > > states.   Which 2 states would you keep?   Would you play to win, or
> > > > NOT to lose?
> > >
> > > Does it matter? If this hypothetical database is complete, couldn't
> > > you just infer 'the other bit' by using symmetry and colour reversal?
> >
> > Not unless you doubled the database size.  I'm assuming that the
> > database is 1 bit per "configuration" - for instance only giving a
> > result for WHITE to MOVE.
> 
> No need to shout... Your example seems to be simply a matter of
> compression, which one should probably do anyway. Why would that make
> the database incomplete?
> 
> 
> > So a better rephrasing of the question: if you had a database that
> > contained either win or loss information for the color your computer
> > is playing - which would you choose?
> 
> If you have a complete database then you can use symmetry of the board
> to obtain the information for the (colour+player-to-move) reversed
> situation, right?
> 
> I only see your point if the database is somehow incomplete, because
> it would, e.g., not contain irrelevant positions that cannot be forced
> from the starting position...
> 
> 
> > I think with either database, you could probably still approach (but not
> > reach) perfect play with the clever use of null move tricks and deep
> > searches.   Because tempo is so critical in chess, a simple thing might
> > be to find positions that don't lose even if you play a null move.  This
> > might work early in the game.  At the very least you could make a tree
> > search much more focused.
> 
> Maybe my chess-knowledge is wrong, but why *not* perfect with a simple
> one ply search? You only need to distinguish draw from win, if both
> sides can reach draw or win (which will show up from the database with
> a one ply search) you know it's a draw, otherwise it's a win...
> 
> E.