Quoting Peter Drake <[EMAIL PROTECTED]>:
To those of you with Monte Carlo programs:
Consider this board configuration:
.........
..ww.....
.wBBw....
..wBBw...
...w.....
.........
......B..
.........
.........
I believe black's best move is to play in the center and escape via
the broken ladder.
1) Is this true? (This is a Go-playing question.)
It is the best move since if white captures the game is essentially over with
normal komi.
A plain monte carlo program should have problems with ladders in
general, but I
am not sure that this position is a good test of it.
The problem here is that the position is lost for black. White has 7 stones on
the board and black 5 so that is not that surprising. Normally a broken ladder
is catastrophic but here white still has the advantage.
2) Can your program find this move (and decide that it's not a good
move if the ladder-breaker is missing)?
Valkyria finds it as the best move immediately.
3) If so, how many runs does it take?
4) With fewer runs, what does the program do? If there is no ladder
breaker, does the program think the ladder works until it has read it
to the end?
I'm having a devil of a time getting my program to solve this one,
and I'm wondering if I just picked an extremely hard problem.
It is not that hard but the problem is that since black cannot win, and
in such
cases monte carlo programs tend to play any random move.
Valkyria can read ladders so it often behaves nice in these situations.
But the
weird thing is that when it played itself in this position, white played out
the ladder to the end. Since the ladder is so short, black is only able to
capture big once and thus white wins easily anyway.
-Magnus
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
