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

Re: [computer-go] Computer Go tournament at EGF



In message <b2b836a0502150359495dba4b@xxxxxxxxxxxxxxxxx>, Mark Boon <tesujisoftware@xxxxxxxxxxxxxxxxx> writes
On Tue, 15 Feb 2005 11:32:57 +0000, Nick Wedd <nick@xxxxxxxxxxxxxxxxx> wrote:
>
>i.e. is there a deterministic, bounded-time algorithm
>to determine that a board position is "finished"?  at
>least the way that i and other players i've played
>with
>play go, it's not clear that there is or might be such
>an algorithm.  (two consecutive passes before a game
>is considered over).

There is no such algorithm.

Knowing whether this is all my territory, or whether my opponent's group
inside it has a way to live, is difficult.

Actually this is not correct. Such an algorithm is theoretically very
easy for Chinese rules. Japanese rules are a bit more complex, but
come down to the same thing.

The game is finished when neither player wants to play any longer and
therefore pass. For a computer to implement this is easy, just
continue playing until you can't play a move that doesn't put yourself
in atari. You can speed this up a bit by not playing on points that
are only next to your color either, although in that case you have to
be careful not to put your own stones in danger by filling a critical
liberty of yourself. But to make a successful program you'll have to
do that anyway.

In practice, things are a bit more complicated though, as humans don't
like to play the game out that far.
Indeed, with Chinese/area rules, it is much simpler. Try this, with Japanese/territory rules:

. . . . . . . . . Complete 9x9 board
. . . . . . . . . # to play
. . . . . . . . . no komi
O O O O O O . . . (position adapted from Davies Life & Death p. 59)
# # # # # O O O O
. # . O # # # # #
# # . O . . # . #
# O O . . O O # O
. . . . . . . . .

Is this position "finished"?

If you can work out the answer in less than 15 minutes, you are a better player than I am. If your program can work out the answer at all, I will be impressed. I set Thomas Wolf's GoTools working on it - its estimate of the time it would need rose to over two hours, and then it gave up trying.

If the lower white group is dead, and black passes, black wins by one point. But if black passes and the white group lives, white wins. If black plays so as to make sure the group it dead, it is jigo.

Using Chinese/area rules it is much simpler. Black just plays to ensure the white group is dead - lots of moves will do but h1 is most convincing - and wins by one point.

Some people prefer Japanese/territory rules because they make players think about difficult questions like this. If I am running a computer Go tournament, I prefer Chinese/area rules, because then (competent) programs play so as to clear up difficult positions like this one, and I am not left with messy unfinished positions to adjudicate.

Nick
--
Nick Wedd nick@xxxxxxxxxxxxxxxxx
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/