Re: [Bug-gnubg] Dropped by move filter

[Robert-Jan Veldhuizen]

On Tue, Jun 24, 2008 at 11:32 AM, Achim Mueller <address@hidden> wrote:

The point is not a technical bug but the fact that gnubg totally misevaluates a position at 0ply that actually should be handled easily.
I put some more positions at http://www.acepoint.de/index.php?id=37. This section will be extended from time to time.
Another idea is to open a section at www.gnubg.org that can be filled by readers and users also.

Actually this one is not so easy I think. A quick 0/2-ply rollout suggests that this is a heavily cube-influenced checker play:

    1. Rollout          21/15 8/4                    Eq.:  +0,431
        61,3%   7,5%   0,4% -  38,7%   3,4%   0,1% CL  +0,283 CF  +0,431
      [  0,1%   0,1%   0,1% -   0,1%   0,1%   0,0% CL   0,003 CF   0,007]
        Full cubeful rollout with var.redn.
        1296 games, Mersenne Twister dice gen. with seed 826408977 and quasi-random dice
        Play: 0-ply cubeful prune [expert]
        Cube: 2-ply cubeful prune [world class]
    2. Rollout          21/11                        Eq.:  +0,393 ( -0,037)
        60,6%   5,7%   0,4% -  39,4%   4,8%   0,1% CL  +0,230 CF  +0,393
      [  0,1%   0,1%   0,1% -   0,1%   0,1%   0,0% CL   0,003 CF   0,008]
        Full cubeful rollout with var.redn.
        1296 games, Mersenne Twister dice gen. with seed 826408977 and quasi-random dice
        Play: 0-ply cubeful prune [expert]
        Cube: 2-ply cubeful prune [world class]
    3. Rollout          21/17*/11                    Eq.:  +0,365 ( -0,066)
        64,1%  11,7%   0,4% -  35,9%  12,2%   0,2% CL  +0,293 CF  +0,365
      [  0,1%   0,1%   0,0% -   0,1%   0,1%   0,0% CL   0,003 CF   0,007]
        Full cubeful rollout with var.redn.
        1296 games, Mersenne Twister dice gen. with seed 826408977 and quasi-random dice
        Play: 0-ply cubeful prune [expert]
        Cube: 2-ply cubeful prune [world class]

As you can see, the #3 play, gnubg 0-ply's favourite, still has the highest cubeless equity. It's not surprising that without a lookahead, gnubg fails to see the future cube implications here.

I guess what's going on, is that a hit leads to overkill after a dance and subsequent cube. So, cubeful, you don't reap full benefits of this play when it works. The alternative play 21/15 8/4 is less volatile and supposedly leads to more efficient cube usage (after a miss).

Greetings,
-- 
Robert-Jan Veldhuizen