[Top][All Lists]

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

[gnugo-devel] Re: gnu-go assignment

From: bump
Subject: [gnugo-devel] Re: gnu-go assignment
Date: Wed, 13 Nov 2002 17:09:59 -0800

> Using RedHat linux 7.3. That should work alright. right?


I'll cc the following to the list. It's a first take on what
could be a reasonable project along these lines.

Let me try to formulate a more exact problem. Assume the board
can be divided into areas called local areas which are disjoint,
and such that all endgame play will occur in one of these local
areas. The local game consists of knowing the outcome in a 
local area of each of four possibilities. The following can
be computed by searching a small tree. There are more possibilities
than this of course, but this is sufficient to capture the ideas
of sente and gote.

O is the player to move, X is the opponent. Tenuki means to play
away (in a different local game).

(A) O plays first, both players alternate. Both make optimal moves
in the local area and at the end the game is scored. We also need
to know who ends up making the last move.

(B) O plays first, X tenukis, O plays again, both players alternate.

(C) X plays first, both players alternate.

(D) X plays first, O tenukis, X plays again, both players alternate.

These are experiments whose outcome can be called the local game.
Each experiment produces a score, and the information who makes
the last move. The expectation is that if X's area is being reduced
X will make the last move and conversely. Only the differences
between scores is significant.

There are then three problems:

(1) Design an algorithm for determining the local areas, assuming
the game is sufficiently far progressed. I'm not sure of the best
approach here.

(2) Given the outcomes of the four experiments, give an algorithm
for playing the global game optimally.

(3) Write some code to play the four experiments. This should be
fairly easy using GNU Go and the oracle patch.

Maybe it's a good idea to look at some positions and try to 
describe how this will work in practice.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]