
BGonline.org Forums
Rollout Results Posted
Posted By: Paul Weaver
Date: Thursday, 19 December 2013, at 6:36 a.m.
In Response To: Aces From The Bar (Paul Weaver)
Thanks to John O'Hagan for his great concise analysis.
I am posting the rollout results of five positions below. Position 1 is the original position in my "Aces from the Bar" post. All four legal plays have been rolled out. I am proud to say that over the board, I quickly found and made the worst, which allows White to have a quite a party in Position 5 when she rolls 44 55 66, assuming she knew enough to double before rolling her joker. The Position 2 rollout result shows the cube action after the best play. The cube actions after the other plays are shown in Positions 3, 4 and 5.
I wanted to know if White had a double after any of Blue's four plays, so I rolled out the four cube actions. Am I the only one who would love it if Xavier would give us a way to do this in XG, with all the double/no double info on all four plays in the same file?
When you carefully compare the data under Position 1 to the data under Position 2, you will see that there is a problem, a discrepency. According to the rollout data in Position 1, Blue's equity after making the best play (B/24 6/5(2) 8/7) is –.640. However, according to the rollout data in Position 2, Blue's equity after making the same play is –.675. This difference of .035 is way too big to attribute to a small sample size. I believe the problem is that XG, when rolling it out in Position 1, thought it was a double. The –.640 given under Position 1 corresponds to the –.636 double/take equity given below Position 2. But Position 2 is not a double.
This means that in reality, the two best plays are only .002 apart and not .037 apart, as the Position 1 rollout data indicate. In this instance, the best play in the rollout of Position 1 holds up, but in other positions where XG makes a double/no double error, a long rollout might tell us that the best play is actually wrong by .02 or more. XG is amazing and it is the best, but it is not perfect.
When we do a checker play rollout in XG, it would be fantastic if we could have the option of simultaneously finding out if it is a double or not after each of the plays. If we had that option, then XG could avoid or correct the problem in the data I described above.
Position 1.
White is NevzatDogan
score: 0
pip: 142Unlimited Game
Jacoby Beaverpip: 161
score: 0
Blue is peeverXGID=aaEDAbCbcdbAA:0:0:1:11:0:0:3:0:10 Blue to play 11
1. Rollout^{1} Bar/24 8/7 6/5(2) eq: 0.640
Player:
Opponent:36.24% (G:8.47% B:0.41%)
63.76% (G:25.25% B:2.61%)Conf.: ± 0.006 (0.646...0.634)  [100.0%]
Duration: 2 hours 05 minutes2. Rollout^{2} Bar/24 11/10 6/5(2) eq: 0.677 (0.037)
Player:
Opponent:35.37% (G:8.57% B:0.41%)
64.63% (G:23.13% B:1.77%)Conf.: ± 0.006 (0.683...0.671)  [0.0%]
Duration: 49 minutes 32 seconds3. Rollout^{2} Bar/24 6/5(3) eq: 0.693 (0.053)
Player:
Opponent:34.56% (G:8.29% B:0.36%)
65.44% (G:22.45% B:1.67%)Conf.: ± 0.005 (0.699...0.688)  [0.0%]
Duration: 46 minutes 03 seconds4. Rollout^{2} Bar/23 6/5(2) eq: 0.713 (0.073)
Player:
Opponent:34.88% (G:8.36% B:0.39%)
65.12% (G:26.19% B:1.34%)Conf.: ± 0.008 (0.720...0.705)  [0.0%]
Duration: 48 minutes 24 seconds^{1} 20736 Games rolled with Variance Reduction.
Moves: 3ply, cube decisions: XG Roller
^{2} 10368 Games rolled with Variance Reduction.
Moves: 3ply, cube decisions: XG RollereXtreme Gammon Version: 2.10
Position 2.
White is NevzatDogan
score: 0
pip: 142Unlimited Game
Jacoby Beaverpip: 157
score: 0
Blue is peeverXGID=aaBCACAbCbcdbB:0:0:1:00:0:0:3:0:10 White on roll, cube action?
Analyzed in Rollout No double Double/Take Player Winning Chances: 64.10% (G:24.78% B:2.37%) 63.79% (G:25.20% B:2.58%) Opponent Winning Chances: 35.90% (G:8.36% B:0.40%) 36.21% (G:8.53% B:0.41%) Cubeless Equities +0.466 +0.928 Cubeful Equities No double: +0.675 ±0.004 (+0.671..+0.679) Double/Take: +0.636 (0.039) ±0.006 (+0.631..+0.642) Double/Pass: +1.000 (+0.325) Best Cube action: No double / Take Percentage of wrong pass needed to make the double decision right: 9.6% Rollout details 20736 Games rolled with Variance Reduction.
Moves: 3ply, cube decisions: XG RollerDouble Decision confidence: 100.0% Take Decision confidence: 100.0% Duration: 3 hours 21 minutes eXtreme Gammon Version: 2.10
Position 3.
White is NevzatDogan
score: 0
pip: 142Unlimited Game
Jacoby Beaverpip: 157
score: 0
Blue is peeverXGID=aaBCDAbCbcdbB:0:0:1:00:0:0:3:0:10 White on roll, cube action?
Analyzed in Rollout No double Double/Take Player Winning Chances: 64.64% (G:23.33% B:1.81%) 64.49% (G:23.54% B:1.95%) Opponent Winning Chances: 35.36% (G:8.70% B:0.42%) 35.51% (G:8.78% B:0.42%) Cubeless Equities +0.453 +0.905 Cubeful Equities No double: +0.677 ±0.006 (+0.671..+0.683) Double/Take: +0.630 (0.047) ±0.007 (+0.622..+0.637) Double/Pass: +1.000 (+0.323) Best Cube action: No double / Take Percentage of wrong pass needed to make the double decision right: 11.4% Rollout details 10368 Games rolled with Variance Reduction.
Moves: 3ply, cube decisions: XG RollerDouble Decision confidence: 100.0% Take Decision confidence: 100.0% Duration: 1 hour 27 minutes eXtreme Gammon Version: 2.10
Position 4.
White is NevzatDogan
score: 0
pip: 142Unlimited Game
Jacoby Beaverpip: 157
score: 0
Blue is peeverXGID=aaCBDAbCbcdbB:0:0:1:00:0:0:3:0:10 White on roll, cube action?
Analyzed in Rollout No double Double/Take Player Winning Chances: 65.47% (G:22.49% B:1.61%) 65.35% (G:22.58% B:1.71%) Opponent Winning Chances: 34.53% (G:8.42% B:0.37%) 34.65% (G:8.49% B:0.39%) Cubeless Equities +0.462 +0.922 Cubeful Equities No double: +0.692 ±0.015 (+0.677..+0.707) Double/Take: +0.657 (0.035) ±0.018 (+0.638..+0.675) Double/Pass: +1.000 (+0.308) Best Cube action: No double / Take Percentage of wrong pass needed to make the double decision right: 9.3% Rollout details 1296 Games rolled with Variance Reduction.
Moves: 3ply, cube decisions: XG RollerDouble Decision confidence: 99.8% Take Decision confidence: 100.0% Duration: 7 minutes 50 seconds eXtreme Gammon Version: 2.10
Position 5.
White is NevzatDogan
score: 0
pip: 142Unlimited Game
Jacoby Beaverpip: 157
score: 0
Blue is peeverXGID=aaBCDAbCbcdbAA:0:0:1:00:0:0:3:0:10 White on roll, cube action?
Analyzed in Rollout No double Double/Take Player Winning Chances: 65.26% (G:25.55% B:1.26%) 65.40% (G:26.42% B:1.40%) Opponent Winning Chances: 34.74% (G:8.08% B:0.36%) 34.60% (G:8.36% B:0.41%) Cubeless Equities +0.489 +0.997 Cubeful Equities No double: +0.699 (0.031) ±0.014 (+0.685..+0.713) Double/Take: +0.730 ±0.018 (+0.712..+0.748) Double/Pass: +1.000 (+0.270) Best Cube action: Double / Take Rollout details 1296 Games rolled with Variance Reduction.
Moves: 3ply, cube decisions: XG RollerDouble Decision confidence: 99.6% Take Decision confidence: 100.0% Duration: 27 minutes 09 seconds eXtreme Gammon Version: 2.10

BGonline.org Forums is maintained by Stick with WebBBS 5.12.