Re: [Bug-gnubg] Use of Two-Sided DB in matches

[Michael Petch]

Because Gnugb plays based on the score and cube value (When in Match play). When it evaluates it doesn’t blindly use the 2 sided bearoff database because the static values the 2 sided database don’t account for score by itself. Jankowski’s forumlas are applied to the values that  come from the DB and the eval continues. The bigger question is, is Jankowski’s formula at fault in how we use it. 

I don’t see this as a bug still. But it doesn’t mean the methodology has been verified.

On 04/09/09 10:56 PM, "Neil Robins" <address@hidden> wrote:

It has occurred to me that in a position like this previously discussed one:

http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=47213

GNU Backgammon  Position ID: gwEAQCoAAAAAAA
                 Match ID   : cAmgACAAGAAA
 +13-14-15-16-17-18------19-20-21-22-23-24-+     O: gnubg
 |                  |   | O              O | OOO 2 points
 |                  |   | O              O | OO  
 |                  |   |                  | OO  
 |                  |   |                  | OO  
 |                  |   |                  | OO
v|                  |BAR|                  |     5 point match (Cube: 1)
 |                  |   |                  | XX
 |                  |   |                  | XX  
 |                  |   |                  | XX  
 |                  |   |                  | XX  On roll
 |                  |   | X  X  X     X    | XXX 3 points
 +12-11-10--9--8--7-------6--5--4--3--2--1-+     X: Neil

The MWC chance following Double/Take is directly obtainable from the Two-Sided Database. Doubling produces an automatic redouble so the situation is in effect cubeless.
The DB gives Cubeless equity: +0.4707.

Therefore where x = MWC/100
x - (1 - x) =  0.4707
x = 0.73535
MWC = 73.535% after Double/Take
This is in very close agreement with the XG rollout.

Is there any reason GNU analysis cannot use this exact method?

I feel I still get complete nonsense here with the latest windows build.

7-ply cubeless MWC  67.39% (Money:  +0.387)
  0.694 0.000 0.000 - 0.306 0.000 0.000
Cubeful equities:
1. No double            70.06%
2. Double, pass         75.08%  (  5.01%)
3. Double, take         69.36%  ( -0.70%)
Proper cube action: No double, take (12.3%)

It also sure worries me to get rollout results that are quite certainly wrong.

Rollout details:
Centered 1-cube:
  0.728 0.000 0.000 - 0.272 0.000 0.000 CL  68.25% CF  70.66%
 [0.000 0.000 0.000 - 0.000 0.000 0.000 CL   0.00% CF   0.00%]
Player gnubg owns 2-cube:
  0.732 0.000 0.000 - 0.268 0.000 0.000 CL  81.71% CF  73.16%
 [0.000 0.000 0.000 - 0.000 0.000 0.000 CL   0.00% CF   0.00%]
Full cubeful rollout with var.redn.
129600 games, Mersenne Twister dice gen. with seed 865036646 and quasi-random dice
Play:  2-ply cubeful prune [world class]
keep the first 0 0-ply moves and up to 8 more moves within equity 0.12
Skip pruning for 1-ply moves.
Cube: 2-ply cubeful prune [world class]

---

_______________________________________________
Bug-gnubg mailing list
address@hidden
http://lists.gnu.org/mailman/listinfo/bug-gnubg