RE: computer-go: [Question] Nullmove

["Xu, Mousheng (SEA)" <moushengxu@xxxxxxxxxxxxxxxx>]

Jeff,

	OK. Then in this case, as I guessed before, you might need to
think about "NULLMOVE" or its equivalent, because you calculate locally
and need to know if there is a more valuable place globally. But if you
always consider globally (e.g. HandTalk, I guess), then there is no need
to consider "NULLMOVE" except maybe at end game.

	"The enemy's best position is my best position" is a Go idiom.
But in many cases, it is not true. If your example, assume the best move
for 'O' is d3, but the best move for 'X' might or might not be d3 --
e.g., it could be f3, depending on the environment.

-- Mousheng Xu


For my program, I don't do a NULL move, but basically the equivelant:  I

analyze the board for both white and black (assume the computer is
white).  
Then I analyze the best white moves for whether or not they help, and
(more 
importantly) whether they need to be done now (can an attack on a group
or 
defense wait?).  If everything in white's best moves indicates that
waiting 
is a viable option, I look at black's (the player's) best moves.
Perhaps 
this next statement is that of a very inexperienced (13k) player, but I
have 
the program assume that black's best move is white's best defense:

12345
.....a
..X..b
.XOX.c
.....d

Example: Black's best move is d3 (according to my program), so that
would by 
white's best defense.

Well, that's my take on NULL move.

Jeff


Xu, Mousheng wrote:
>I have been trying to understand the lengthy discussions about nullmove
>for a few days, but still have no clue why you guys are interested in
>it. So here comes my question.
>
>In a game tree, because of the limitation of computer power, usually
you
>decide to stop while the width or the depth of the tree reaches a
>certain number (say, 3 nodes wide, 4 nodes deep). Since the depth of
the
>terminal nodes are the same in 99% cases, the terminal evaluation can
be
>assumed to be equally fair to each node. If I understand correctly, you
>guys are talking about to take nullmove into consideration while
>evaluating the terminal node so that the evaluation is closer to
>reality. But why do you care about the absolute values of your
>evaluation? You only care about the relative order of the evaluations,
>right?
>
>I am sure I am missing something.
>
>Thanks.
>
>-- Mousheng Xu

______________________________________________________
Get Your Private, Free Email at http://www.hotmail.com