RE: [computer-go] Interesting problem with Monte Carlo Go

["Peter McKenzie" <peter_mckenzie@xxxxxxxxxxx>]

I'm no expert but I think you have identified the basic problem with the 
Monte Carlo approach, namely that it is fooled by positions where there are 
only a small number of moves that lead to the optimal outcome.

It seems that the common approach to (partially) solving this problem is to 
adopt some sort of hybrid approach which combines search and Monte Carlo.

Regards,
Peter McKenzie

> From: Peter Drake <drake@xxxxxxxxxxxxxxxxx>
> Reply-To: computer-go <computer-go@xxxxxxxxxxxxxxxxx>
> To: Computer Go <computer-go@xxxxxxxxxxxxxxxxx>
> Subject: [computer-go] Interesting problem with Monte Carlo Go
> Date: Thu, 2 Jun 2005 16:51:30 -0700
>
> Consider this situation:
>
> . w . . . . .
> w . w B B B .
> . . B w w B .
> . B w . w B .
> . . B w B w .
> . . . B . . w
> . . . . . w .
>
> If black moves in the center, five white stones are captured.  On the  
> other hand, there are two places where white can capture a single  black 
> stone and escape.  The first player to make one of these moves  will win.  
> (Raw) Monte Carlo will judge that the white stones are  alive, because they 
> live in the majority of random completions of the  game.
>
> There may be a flaw in this particular problem, but I want to address  the 
> situation where there are many ways to live but only one way to  die (or 
> vice versa).
>
> Thoughts?
>
> Peter Drake
> Assistant Professor of Computer Science
> Lewis & Clark College
> http://www.lclark.edu/~drake/
>
>
> _______________________________________________
> computer-go mailing list
> computer-go@xxxxxxxxxxxxxxxxx
> http://www.computer-go.org/mailman/listinfo/computer-go/
_________________________________________________________________
Check out the latest video  @  http://xtra.co.nz/streaming

_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/