|
BGonline.org Forums
Rollout Results Posted
Posted By: Paul Weaver In Response To: Magriel New York Times column position 1977-06-09 (leobueno)
Date: Wednesday, 25 December 2013, at 8:20 p.m.
Four rollout results are posted below.
If you look at the data under Position 1, you might think that making the 4pt is better than 13/9 6/1* (0.648 vs .654). However, if you look at the data under Positions 2 and 3, you might think that 13/9 6/1* is better than making the 4pt (.652 vs .663).
Position 1 gives Blue's equity after 13/9 6/1* as 0.654, while Position 3 gives Blue's equity after 13/9 6/1* as 0.652. I can live with this difference of 0.002. However, Position 1 gives Blue's equity after making the 4pt as 0.648 while Position 2 gives Blue's equity after making the 4pt as 0.663. This difference of 0.015 is too large to be attributable to small sample size.
What is going on here? The Position 1 rollout was performed using 3-ply for checker plays and XG rollout for cube decisions. Position 5 below shows that XG roller incorrectly thinks that White has a double. I believe that XG roller's doubling when it should not is what causes its own rollout to go astray in Position 1.
I believe the best play is actually 13/9 6/1* (I determine this from the rollout data under Positions 2 and 3, 0.652 vs 0.663) and I believe that Position 1 gives the wrong answer, regardless of how many times it is rolled out.
I started using Jellyfish nineteen years ago. When you wanted to do a checker play rollout with Jellyfish, you had to create a different file for each checker play and then roll out each of the files.
Currently, when you want to use to XG determine the best checker play in Position 1 and then determine if White has a double after the plays, you have to create separate files and do separate rollouts. Maybe one day XG or another bot will allow the user to do everything in one file. When XG allows us to do, this, it will be able to correct the mistake it made in Position 1 as a result of incorrectly doubling.
Position 1.
White is Player 2
score: 0
pip: 141Unlimited Game pip: 159
score: 0
Blue is Player 1XGID=-aa---E-CB--bC-b-b-c-bb-B-:0:0:1:54:0:0:0:0:10 Blue to play 54
1. Rollout1 9/4 8/4 eq: -0.648
Player:
Opponent:34.71% (G:10.53% B:0.49%)
65.29% (G:23.84% B:2.24%)Conf.: ± 0.007 (-0.655...-0.641) - [91.5%]
Duration: 1 hour 40 minutes2. Rollout1 13/9 6/1* eq: -0.654 (-0.006)
Player:
Opponent:35.81% (G:9.21% B:0.38%)
64.19% (G:23.66% B:2.06%)Conf.: ± 0.006 (-0.660...-0.649) - [8.5%]
Duration: 1 hour 52 minutes3. Rollout1 24/20 6/1* eq: -0.695 (-0.047)
Player:
Opponent:36.13% (G:9.05% B:0.42%)
63.87% (G:29.16% B:1.63%)Conf.: ± 0.008 (-0.703...-0.688) - [0.0%]
Duration: 1 hour 57 minutes1 10368 Games rolled with Variance Reduction.
Moves: 3-ply, cube decisions: XG RollereXtreme Gammon Version: 2.10
Position 2.
White is Player 2
score: 0
pip: 141Unlimited Game pip: 150
score: 0
Blue is Player 1XGID=-aa-B-E-BA--bC-b-b-c-bb-B-:0:0:-1:00:0:0:0:0:10 White on roll, cube action?
Analyzed in Rollout No double Double/Take Player Winning Chances: 65.25% (G:23.13% B:1.98%) 65.14% (G:23.82% B:2.21%) Opponent Winning Chances: 34.75% (G:10.25% B:0.45%) 34.86% (G:10.59% B:0.49%) Cubeless Equities +0.449 +0.904 Cubeful Equities No double: +0.663 ±0.005 (+0.658..+0.668) Double/Take: +0.636 (-0.027) ±0.007 (+0.629..+0.643) Double/Pass: +1.000 (+0.337) Best Cube action: No double / Take Percentage of wrong pass needed to make the double decision right: 6.8% Rollout details 10368 Games rolled with Variance Reduction.
Moves: 3-ply, cube decisions: XG RollerDouble Decision confidence: 100.0% Take Decision confidence: 100.0% Duration: 2 hours 11 minutes eXtreme Gammon Version: 2.10
Position 3.
White is Player 2
score: 0
pip: 142Unlimited Game pip: 150
score: 0
Blue is Player 1XGID=aAa---D-CC--bB-b-b-c-bb-B-:0:0:-1:00:0:0:0:0:10 White on roll, cube action?
Analyzed in Rollout No double Double/Take Player Winning Chances: 64.16% (G:23.66% B:2.00%) 64.24% (G:23.75% B:2.07%) Opponent Winning Chances: 35.84% (G:9.23% B:0.40%) 35.76% (G:9.33% B:0.42%) Cubeless Equities +0.443 +0.891 Cubeful Equities No double: +0.652 ±0.006 (+0.646..+0.658) Double/Take: +0.609 (-0.044) ±0.007 (+0.601..+0.616) Double/Pass: +1.000 (+0.348) Best Cube action: No double / Take Percentage of wrong pass needed to make the double decision right: 10.1% Rollout details 10368 Games rolled with Variance Reduction.
Moves: 3-ply, cube decisions: XG RollerDouble Decision confidence: 100.0% Take Decision confidence: 100.0% Duration: 3 hours 16 minutes eXtreme Gammon Version: 2.10
Position 4.
White is Player 2
score: 0
pip: 142Unlimited Game pip: 150
score: 0
Blue is Player 1XGID=aAa---D-CB--bC-b-b-cAbb-A-:0:0:-1:00:0:0:0:0:10 White on roll, cube action?
Analyzed in Rollout No double Double/Take Player Winning Chances: 63.76% (G:28.76% B:1.52%) 63.89% (G:29.05% B:1.48%) Opponent Winning Chances: 36.24% (G:9.01% B:0.42%) 36.11% (G:9.18% B:0.45%) Cubeless Equities +0.484 +0.973 Cubeful Equities No double: +0.618 (-0.074) ±0.013 (+0.606..+0.631) Double/Take: +0.692 ±0.023 (+0.669..+0.715) Double/Pass: +1.000 (+0.308) Best Cube action: Double / Take Rollout details 1296 Games rolled with Variance Reduction.
Moves: 3-ply, cube decisions: XG RollerDouble Decision confidence: 100.0% Take Decision confidence: 100.0% Duration: 13 minutes 00 second eXtreme Gammon Version: 2.10
Position 5.
White is Player 2
score: 0
pip: 141Unlimited Game pip: 150
score: 0
Blue is Player 1XGID=-aa-B-E-BA--bC-b-b-c-bb-B-:0:0:-1:00:0:0:0:0:10 White on roll, cube action?
Analyzed in XG Roller No double Double/Take Player Winning Chances: 64.83% (G:23.10% B:2.07%) 65.24% (G:23.17% B:2.01%) Opponent Winning Chances: 35.17% (G:9.64% B:0.41%) 34.76% (G:9.67% B:0.38%) Cubeless Equities +0.448 +0.912 Cubeful Equities No double: +0.662 (-0.001) Double/Take: +0.663 Double/Pass: +1.000 (+0.337) Best Cube action: Double / Take eXtreme Gammon Version: 2.10
|
BGonline.org Forums is maintained by Stick with WebBBS 5.12.