Re: [bug-gnubg] bug in gammon-save eval

[Jim Segrave]

On Fri 25 Apr 2003 (13:45 -0700), Ned Cross wrote:
> GNU is choosing a move that loses more gammons because it has a fantasy that 
> there are actual game-winning chances from this position.  Very strange.
> 
> -Ned
> 
> 
> black (O, 10 pts) vs. white (X, 1 pts) (Match to 15)
> 
> Game number 6
> 
> Match Information:
> 
> black's rating: n/a
> white's rating: n/a
> Date: n/a
> Event: n/a
> Round: n/a
> Place: n/a

> Annotator: n/a
> Comments: n/a
> Move number 47: X to play 34
> 
>  GNU Backgammon  Position ID: HQAAYLYbsgAAAA
>                  Match ID   : UYnxAaAACAAA
>  +13-14-15-16-17-18------19-20-21-22-23-24-+     O: black
>  | X                |   |             O  O | OOO 10 points
>  |                  |   |             O    | OO  
>  |                  |   |             O    | OO  
>  |                  |   |                  | OO  
>  |                  |   |                  | OO 
> v|                  |BAR|                  |     15 point match
>  |                  |   |                  |    
>  |                  |   |                  |     
>  |                  |   |    X             |     
>  | X                |   | X  X  X  X     X |     Rolled 34
>  | X     X          |   | X  X  X  X     X |     1 point
>  +12-11-10--9--8--7-------6--5--4--3--2--1-+     X: white (Cube: 2)
> Pip counts: O 7, X 90
> 
> * white moves 13/6
> 
> Cube analysis
> 2-ply cubeless equity  -1.630 (Money:  -1.936)
>   0.000 0.000 0.000 - 1.000 0.936 0.000
> Cubeful equities:
>   2-ply cubeful 100% speed [world class]
> 1. No double            -1.630
> 2. Double, pass         +1.000  ( +2.630)
> 3. Double, take         -1.894  ( -0.264)
> Proper cube action: No redouble, take (9.1%)
> 
> Rolled 34 (-0.023):
>      1. Cubeful 2-ply    13/10 12/8                   Eq.:  -1.629
>        0.008 0.002 0.000 - 0.992 0.974 0.000
> *    2. Cubeful 2-ply    13/6                         Eq.:  -1.647 ( -0.018)
>        0.000 0.000 0.000 - 1.000 0.962 0.000
>      3. Cubeful 2-ply    12/9 10/6                    Eq.:  -1.663 ( -0.034)
>        0.000 0.000 0.000 - 1.000 0.986 0.000
>      4. Cubeful 2-ply    13/9 12/9                    Eq.:  -1.663 ( -0.034)
>        0.000 0.000 0.000 - 1.000 0.986 0.000
>      5. Cubeful 2-ply    12/5                         Eq.:  -1.665 ( -0.036)
>        0.000 0.000 0.000 - 1.000 0.988 0.000
>      6. Cubeful 2-ply    13/9 10/7                    Eq.:  -1.667 ( -0.038)
>        0.000 0.000 0.000 - 1.000 0.991 0.000

If you have the player settings for gnubg different than the settings you
are using for analysis, then sometimes the analysis will give a
different choice of best move than the one gnubg has made. You can see
some very interesting results if you set the player to expert and
analysis to world class or supremo. Gnubg will sometimes rate some
moves it has made as a blunder

But I find the analysis odd:

There's no contact and X has pieces off. So gammon chances are, by
definition, zero. And chances of winning are also 0. So where do the
.008 and .002 numbers for win/win gammon come on the top move in the
analysis?

The second thing is that O is going to be off in 2 rolls worst case,
no matter what. The only slight chance gnubg has to save the gammon is
to get one man in with this roll, then roll a high double on the next
one. 13/10 12/8 is a guaranteed gammon losing move, as even a high
double on the next roll will only get all the men in, none will be
borne off. 

-- 
Jim Segrave           address@hidden