Re: [Bug-gnubg] Need help on evaluating a position

[Nis Jorgensen]

Nardy Pillards wrote:

>>I have managed to edit the position until obtaining the result listed
>>at the end of this posting. I'd like now to tell GnuBG that opponent
>>is on roll and that he is offering a cube, and would like it to
>>analyse the position and tell me if it's a take or a pass, and I can't
>>find how to do that.
> 
>  
> I assume you use the Win32 GUI build.
> 
> Settings / Players: both human
> Settings / Options /Dice: manual
> 
> In front of the names (bottom of screen) there is a small checker.
> Click on the one before your opponent's name, so it will be his/her 
> turn to roll the dice or use the cube.
> Analyse / Hint (or Ctrl + H) will show you the Hint about the cube.
> Analyse / Evaluate the evaluation of the position.

Why do you set the players to human and dice to manual? I would just:

Click on the checker next to oppenents name (or Game/Set Turn/<name of
opponent>)
Analyse/Hint

> By the way, if you have any advice about this position, I would be happy to 
> hear it.
> For the record, he doubled, I took, redoubled to 8, and rolled a 3 2, so lost.
> 
> I'd like to know if it was right to take or not.
> 
> Thanks very much for your help, and happy neww year to all :-)
> 
> Carla
> 
> The score (after 0 games) is: opponent 4, me 0 (match to 7 points)
> Match Information:
> 
> Date: décembre 31, 2003
> Move number 5: X to play 24
> 
>  GNU Backgammon  Position ID: OwAAwI0AAAAAAA
>                  Match ID   : QQnxAEAAAAAA
>  +13-14-15-16-17-18------19-20-21-22-23-24-+     O: Opponent (Cube: 2)
>  |                  |   |             O  O | OO  4 points
>  |                  |   |             O  O | OO  
>  |                  |   |             O    | OO  
>  |                  |   |                  | OO  
>  |                  |   |                  | OO 
> v|                  |BAR|                  |     7 point match
>  |                  |   |                  | X  
>  |                  |   |                  | XX  
>  |                  |   |                X | XX  
>  |                  |   |             X  X | XX  Rolled 24
>  |                  |   |    X        X  X | XX  0 points
>  +12-11-10--9--8--7-------6--5--4--3--2--1-+     X: me
> Pip counts: O 8, X 12

This is a very illustrative position - since it a big problem gnubg has
with automatic recubes. Here are the results of 0-3 ply:

Cube analysis
0-ply cubeless MWC  88,54% (Money:  +0,793)
  0,896 0,000 0,000 - 0,104 0,000 0,000
Cubeful equities:
1. Double, pass         91,25%
2. Double, take         94,82%  (  3,57%)
3. No double            89,94%  ( -1,31%)
Proper cube action: Redouble, pass

Cube analysis
1-ply cubeless MWC  88,54% (Money:  +0,793)
  0,896 0,000 0,000 - 0,104 0,000 0,000
Cubeful equities:
1. No double            89,70%
2. Double, pass         91,25%  (  1,55%)
3. Double, take         89,65%  ( -0,05%)
Proper cube action: No redouble, take (3,4%)

Cube analysis
2-ply cubeless MWC  88,54% (Money:  +0,793)
  0,896 0,000 0,000 - 0,104 0,000 0,000
Cubeful equities:
1. No double            90,22%
2. Double, pass         91,25%  (  1,03%)
3. Double, take         89,65%  ( -0,57%)
Proper cube action: No redouble, take (35,7%)

Cube analysis
3-ply cubeless MWC  88,54% (Money:  +0,793)
  0,896 0,000 0,000 - 0,104 0,000 0,000
Cubeful equities:
1. Double, take         89,65%
2. Double, pass         91,25%  (  1,60%)
3. No double            88,68%  ( -0,97%)
Proper cube action: Redouble, take

Further plies gives same result.

And here is my explanation of what is happening:

On 0-ply, gnubg doesn't see your recube to 8 - so it calculates a "cube
ownership value" which is much too low.

On 1-ply, gnubg see the recube - so it get's the value of double-take
accurate. It slightly misestimates the cube ownership value of keeping the
cube, though

On 2-ply, gnubg suddenly sees the possibility of cubing on next roll - but
not the recube to 8! Thus it advices you that doubling now is a huge error.

On 3-ply, gnubg sees all the possible cube action, and get's the same
result as an exact rollout.

Note the two cases where gnubg get's it wrong: It sees the cube to 4, but
not that the recube to 8 is automatic.

My guess is that this pattern can be refound in other similar positions.

-- 
Nis Jorgensen
IT Department
Greenpeace International
+31 6 1500 7413