Re: [Bug-gnubg] misevaluation on 0-ply

[Christian Anthon]

Hi Sven-Olav,

you've just stumbled in to one of the rare cases where human odd-ply
outplays gnubg 0-ply, and where 2-ply is always going to seriously
outplay 0-play. This is often the case where the game is more or less
decided in a roll or two.

Christian.

On Mon, Nov 10, 2008 at 1:37 PM, Sven-Olav Norén
<address@hidden> wrote:
> Hi,
>
> Hope you can read the position below. As you can see GNU makes a big misstep
> on 0-ply evaluation. GNU:s choice loses about 80% gammons instead of a
> normal gammon saving move that would lose about 40% gammon. GNU corrects
> itself on 1-ply and higher levels, so it might have no or small practical
> importance. However it bothers me as I often use 0-ply when doing
> rollouts. So I'm curious if anyone with inight in the neural nets might have
> some explanation or comment.
>
>
> Kind regards
> Sven-Olav Norén, Stockholm Sweden
>
> 2wIAAKuzqgIAAA:MIGpAAAAAAAA
>
> The score (after 0 games) is: gnubg 0, snoren 0 (match to 5 points)
> Match Information:
> Date: 2008-11-10
> Move number 3:  gnubg to play 32
>  GNU Backgammon  Position ID: 2wIAAKuzqgIAAA
>                  Match ID   : MIGpAAAAAAAA
>  +12-11-10--9--8--7-------6--5--4--3--2--1-+     O: gnubg
>  | O  O  O  O  O  O |   | O     O  O  O  O |     0 points
>  |                  |   | O     O        O |     Rolled 32
>  |                  |   |       O          |
>  |                  |   |                  |
>  |                  |   |                  |
> ^|                  |BAR|                  |     5 point match (Cube: 1)
>  |                  |   |                  | X
>  |                  |   |                  | X
>  |                  |   |                  | XX
>  |                  |   |          X  X  X | XX
>  |                  |   |       X  X  X  X | XX  0 points
>  +13-14-15-16-17-18------19-20-21-22-23-24-+     X: snoren
> Pip counts: O 88, X 16
>
> Rolled 32:     1. Cubeful 0-ply    4/2 4/1                      Eq.:  -1,407
>        0,000 0,000 0,000 - 1,000 0,433 0,000
>      2. Cubeful 0-ply    9/6 8/6                      Eq.:  -1,432 ( -0,025)
>        0,000 0,000 0,000 - 1,000 0,459 0,000
>      3. Cubeful 0-ply    9/6 7/5                      Eq.:  -1,443 ( -0,036)
>        0,000 0,000 0,000 - 1,000 0,471 0,000
>      4. Cubeful 0-ply    12/10 9/6                    Eq.:  -1,446 ( -0,040)
>        0,000 0,000 0,000 - 1,000 0,475 0,000
>      5. Cubeful 0-ply    12/9 8/6                     Eq.:  -1,471 ( -0,065)
>        0,000 0,000 0,000 - 1,000 0,501 0,000
>
> Output generated Mon Nov 10 13:29:07 2008
> by GNU Backgammon 0.90-mingw (Text Export version 1.91)
>
>
> ________________________________
> Långtråkigt? Mängder av sköna spel att fördriva tiden med! MSN Spelhallen
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/bug-gnubg
>
>