On Wed, 2008-12-10 at 18:56 +0000, tony tang wrote:
> Hi guys,
>  
> It would be interesting to know how everyone limits the number of
> possible moves at the 
> beginning of the game. Assuming the board is 19x19 with no handicap
> and black placed a stone, 
> that leaves (in theory) 390 possible positions which translates to 390
> children from the root (empty board). 
> With my understanding of UCT it appears that all 390 positions have to
> be simulated numerous times 
> and the node with the highest winrate gets explored further. *please
> correct me if im wrong*
>  
> Do most people limit their play in corners and sides, if so how?
> Pruning entire branches at this stage would 
> be a bad idea due to well documented reasons, but i have heard pruning
> being employed at this stage, if
> so what algorithms are used to smartly prune.

Some of my programs do not place any stone on the edge, unless it
touches some other stone.  I also count the diagonal, so if there is a
stone diagonally next to an edge point, I allow a move to that point.
But if there is nothing diagonally or orthogonally adjacent, I do not
allow a move to that point.  

This is not a perfect rule.   But a modification might be to not allow
any move on the edge if there is a stone of either color within
Manhattan distance 2 or 3 or so.    Of course if you use patterns, you
can start to get much more specific and consider exceptions to this
rule.

I think with some creativity you can produce other kinds of rules that
can at least limit the choices for the first several moves.   I'm too
weak to know how to construct these rules.  

If you have a large enough database of strong master games, you might be
able to construct some rules and use the database to verify that your
rules are good.  It could be based on building a hypothesis, then
testing to see if your hypothesis holds.  One must be very careful how
to interpret things - because if there are 350 possible moves, the hit
rate must be extremely high.   So if your test rule is to prohibit one
of these 350 moves,  then 1 hit out of 350 is probably way too many
because that is what we would expect from any "average" move.   You want
to find rules to discard moves that are close to certainty not the best
move.

- Don







>  
> Thanks, 
>  
> Please link me to the thread if this has already been discussed.
> 
> 
> ______________________________________________________________________
> Get Windows Live Messenger on your Mobile. Click Here!
> _______________________________________________
> computer-go mailing list
> [email protected]
> http://www.computer-go.org/mailman/listinfo/computer-go/

Attachment: signature.asc
Description: This is a digitally signed message part

_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/

Reply via email to