Re: [Bug-gnubg] Strange rollout result ?

[Joaquín Koifman]

Hi,

It had happened to me that even the rollout of a move, with the cube in the middle and a position far from a gammon, gives an equity of more than 1.000 because Gnubg is taking the very next cube which is indeed a pass.
In this position seems to happen the same.

About the Double / Pass, look that the difference in equity between the D/P and the ND is 0.000. Gnubg is probably not taking into account an equity greater than 1.000 in the ND case if a checker had been borne off.
After all, Gnubg 'recognizes' in some way that mistake in the evaluation and gives the right decision.

Joaquin

2009/5/18 Massimiliano Maini <address@hidden>

Hi all,

this looks odd to me:

 GNU Backgammon  Position ID: cP8BAJBtGwCwBQ
                 Match ID   : cAlgAXAACAAA
 +13-14-15-16-17-18------19-20-21-22-23-24-+     O: gnubg
 |                  | O | O  O  X  X  X    | O   7 points
 |                  |   | O  O  X  X       | O  
 |                  |   | O  O             |    
 |                  |   | O                |    
 |                  |   | 9                |    
v|                  |BAR|                  |     11 point match (Cube: 1)
 |                  |   |                  |    
 |                  |   |                  |    
 |                  |   |                  |    
 |                  |   | X  X  X  X  X    |     On roll
 |                  |   | X  X  X  X  X    |     1 point
 +12-11-10--9--8--7-------6--5--4--3--2--1-+     X: xxx


Rollout details:
Centered 1-cube:
  0.8109 0.0000 0.0000 - 0.1891 0.0571 0.0050 CL  +0.5695 CF  +1.1167
 [0.0019 0.0000 0.0000 - 0.0019 0.0014 0.0010 CL   0.0043 CF   0.0100]
Player gnubg owns 2-cube:
  0.8194 0.0000 0.0000 - 0.1806 0.0566 0.0071 CL  +1.1630 CF  +1.1385
 [0.0018 0.0000 0.0000 - 0.0018 0.0015 0.0014 CL   0.0086 CF   0.0100]
Full cubeful rollout with var.redn.
1296 games, Mersenne Twister dice gen. with seed 858145082 and quasi-random dice
Play:  2-ply cubeful prune [world class]
keep the first 0 0-ply moves and up to 12 more moves within equity 0.16
Skip pruning for 1-ply moves.
Cube: 2-ply cubeful prune [world class]

In MWC:

Rollout details:
Centered 1-cube:
  0.8109 0.0000 0.0000 - 0.1891 0.0571 0.0050 CL  19.082% CF  21.356%
 [0.0019 0.0000 0.0000 - 0.0019 0.0014 0.0010 CL   0.018% CF   0.041%]
Player gnubg owns 2-cube:
  0.8194 0.0000 0.0000 - 0.1806 0.0566 0.0071 CL  21.548% CF  21.446%
 [0.0018 0.0000 0.0000 - 0.0018 0.0015 0.0014 CL   0.036% CF   0.042%]
Full cubeful rollout with var.redn.
1296 games, Mersenne Twister dice gen. with seed 858145082 and quasi-random dice
Play:  2-ply cubeful prune [world class]
keep the first 0 0-ply moves and up to 12 more moves within equity 0.16
Skip pruning for 1-ply moves.
Cube: 2-ply cubeful prune [world class]


First, with ND at +1.1167 and D/T at 1.1385, gnubg indicates D/P as
the right decision, while it should be ND/P (too good).

Second, there are no gammons for X, how can it be too good ?
Is gnubg taking cobes that are in fact passes later on ?

By the way, the copy button in the hint window/analysis panel does not
copy all the information (it misses the recap with the equities and
differences dor the 3 actions ND, D/T, D/P and the final decision).

MaX.

P.S.
I have a sgf file with the match and the rollout saved.

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