Like David pointed out, cycles can be very complex. In theory. But
with random playouts I believe only triple-ko is possible. And rare
at that. The reason is simple: it will only keep going on if there's
no choice. Otherwise a deviation will be played at some point,
reducing the situation. For there to be no choice I believe it can
only be triple ko. It means all empty points are either a single eye
or part of the ko. Of the three moves in the ko, one is forbidden by
the direct ko detection, one is suicide so only one choice remains.
I don't do super-ko detection at all. In MC playouts I have a move-
limit which is three times the boardsize. In practice this limit is
almost never reached. I believe not even once in a thousand full
games. So in practice it has a pretty negligable effect.
Of course the situation can get different if you impose more rules on
the playout. If for example you disallow certain types of self-atari
or make the move selection more deterministic in some way more types
of cycles may become possible.
Mark
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
