[Top][All Lists]

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

Re: [gnugo-devel] Non-passing Go?

From: Kenneth Owen Stanley
Subject: Re: [gnugo-devel] Non-passing Go?
Date: Sun, 13 Apr 2003 01:39:46 -0500 (CDT)

> > For the purposes of training a machine learning algorithm
> > against GNUGo, it would be useful to be able to force
> > GNUGo to make the best move available even if it wants
> > to pass.  In other words, the game should be allowed to
> > play out even if GNUGo determines that it is effectively
> > over.  This is particularly a problem on small boards.
> > Is there an easy way to force GNUGo to place a piece in
> > the best available position without passing (perhaps
> > through making appropriate library calls)?  
> Normally GNU Go will fill all liberties before passing.
> Further play beyond this point loses points unless you
> use Chinese rules (area counting). If it passes, it 
> believes that any move it plays is inside either its
> own territory or the opponents, unless there is a seki
> on the board.
> Try starting GNU Go with the parameter --play-out-aftermath,
> which sets the the global variable play_out_aftermath to 1.
> (Or add a library function that you can call from your
> engine to accomplish this.) Then genmove will generate moves
> to remove dead stones even if the game is really over. This
> means that GNU Go will play inside its own territory. If you
> combine this with --chinese-rules that should not affect the
> score.
> > Unfortunately, calling
> > 
> > gnugo_genmove(move_i, move_j, WHITE);
> > 
> > results in many passes, sometimes even after black has
> > only placed one stone.
> This could mean that GNU thinks there is no way to live
> on the board. I don't think this should happen with board
> sizes >= 9. What board size are you using?

Thank you for your advice; it should be very helpful.
I am using a board size of 6 presently.  For some
first moves by black, white will pass immediately.


reply via email to

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