[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: computer-go: Evaluating positions
On Tue, Jun 19, 2001 at 05:37:23PM -0700, Bob Myers wrote:
> > The only principle in Go, in this sense, boils down to this single
> > simple principle: "OWN as many squares as possible when the game is
> > over."
>
> No. It is "own at least one more square than your opponent".
>
> And I think these are at least somewhat different. At least in theory, a
> playing style designed to maximize average margin of victory could actually
> lead to a lower probability of victory.
Although I agree that in practice, with mere humans (or worse, computers)
playing, there indeed is a difference, I am not sure if this holds to
perfect players.
This is because we less-than-perfect players can not read things to the end,
but have to make do with approximations, probabilities, and errors, which in
the strictest mathematical sense do not belong in go, except in practice.
Wasn't there one program (Neurogo?) that tried to estimate the probability
of each point belonging to black or white, from the beginning of the game.
This sounds like a neat way to do scoring, and to handle this sort of risk
management.
-H
--
Heikki Levanto LSD - Levanto Software Development <heikki@xxxxxxxxxxxxxxxxx>