|
BGonline.org Forums
Inconsistent XGR++ Evaluations
Posted By: Nack Ballard
Date: Friday, 16 June 2023, at 7:45 p.m.
David Rockwell and I have noticed three scenarios (that might or might not be related) where XG does not consistently display the same evaluation. I'll address them individually:
Scenario 1: A significant checker-play equity difference in at least one specific position.Set up the opening Nackgammon position one roll shy of reaching it; for example as below. Punch up White's roll of 21; it yields a ++ equity (to the fourth decimal place) of –.0813.
[You could also, for example, move one of White's bar checkers to the 24pt and her 8pt spare back to her 14pt, make the roll 62, and look at the equity for bar/23 14/8.]
The problem is that occasionally XG claims a different equity. So far, David and I have also seen –.0858, –.0849, and (I think) –.0847. Once the error appears in some XG instance and you tinker with it, it eventually "rights" itself and shows –.0813.
So far, we are unable to reproduce such a misevaluation at will.
White is Player 2
score: 0
pip: 197Unlimited Game
Jacoby Beaverpip: 194
score: 0
Blue is Player 1XGID=baa---D-C---dD---c-d---BB-:0:0:-1:21:0:0:3:0:10 White to play 21
1. XG Roller++ Bar/24 Bar/23 eq: -0.0813
Player:
Opponent:48.07% (G:12.40% B:0.56%)
51.93% (G:14.34% B:0.76%)eXtreme Gammon Version: 2.10
Scenario 2: Other checker-play equity differences.David has been transcribing some early game positions to the fourth decimal place and some inconsistencies have arisen. I'll let him post his comments separately.
Scenario 3: Cube-action equities.I suppose it could happen in almost any type of position, with some more likely than others. I've noticed it when working with fifth-roll cubes lately. As an example, the XGR++ evaluation for 61P-43U-55P-62S-C (see diagram below) is "Double +0.469, No Double +.0468 (–.001)," which I record in cubrac shorthand as [D 469–1].
It probably doesn't apply to this exact position, but the following hypothetical description demonstrates the problem:
I double-check the same position later and instead of [D 469–1], the evaluation punches up as [D 470–1] or [D 468–1] or [D 469–2] or [D 469–0]. It's never more than a .001 difference (or if it is, then it's some other traceable problem such as wrong score or Jacoby being unchecked) and it happens with relative infrequency.
[Note that I'm working with three decimal places. The actual difference might invisibly be only .0001 or .0002 and it is rounding the opposite direction in such instances.]
Has anyone else noticed inconsistent XGR++ evaluations?
Nack
White is Player 2
score: 0
pip: 152Unlimited Game
Jacoby Beaverpip: 140
score: 0
Blue is Player 1XGID=---Ba-EBB--adBa--c-e----B-:0:0:1:00:0:0:3:0:10 Blue on roll, cube action?
Analyzed in XG Roller++ No double Double/Take Player Winning Chances: 59.50% (G:25.56% B:0.90%) 59.66% (G:26.15% B:0.93%) Opponent Winning Chances: 40.50% (G:8.02% B:0.38%) 40.34% (G:7.83% B:0.38%) Cubeless Equities +0.371 +0.764 Cubeful Equities No double: +0.468 (-0.001) Double/Take: +0.469 Double/Pass: +1.000 (+0.531) Best Cube action: Double / Take eXtreme Gammon Version: 2.10
|
BGonline.org Forums is maintained by Stick with WebBBS 5.12.