[Computer-go] semeai example of winning rate

[Brian Sheppard]

> It seems to me that improvements in playout policy would apply to any time
control.

Well, I have to believe Mogo on this one. They said that fillboard helps
(only) on deep searches on the 19x19 board. So improvement in playout policy
might not apply to any time control.

If your search is exhaustive, then your intuition might be pretty solid. But
MCTS is highly selective, seldom searching more than a few moves per node.
Consequently, the balance between exploration and exploitation is more
delicate.

On a time-limited search, it may pay to concentrate on possibilities that
are likely to be correct. If you have more time, then you can admit more
options.

Brian