|
BGonline.org Forums
Largest −x−x score
Posted By: Nack Ballard In Response To: Largest −x−x score (Sean Garber)
Date: Saturday, 6 May 2017, at 10:24 p.m.
White is Player 2
score: 0
pip: 21337 point match pip: 41
score: 0
Blue is Player 1XGID=-EED----c-ccccA-----------:1:1:1:00:0:0:0:37:10 3-ply rollout
Analyzed in Rollout No redouble Redouble/Take Player Winning Chances: 72.33% (G:63.76% B:0.09%) 52.82% (G:46.97% B:0.04%) Opponent Winning Chances: 27.67% (G:2.67% B:0.00%) 47.18% (G:4.98% B:0.00%) Cubeless Equities +1.120 +1.094 Cubeful Equities No redouble: +1.149 ±0.056 (+1.093..+1.205) Redouble/Take: +0.769 (-0.380) ±0.098 (+0.671..+0.867) Redouble/Pass: +1.000 (-0.149) Best Cube action: No redouble / Take Rollout details 2592 Games rolled with Variance Reduction.
Moves: 3-ply, cube decisions: XG RollerDouble Decision confidence: 100.0% Take Decision confidence: 100.0% Duration: 32 minutes 10 seconds
3-ply rollouts are not enough.
The problem with this contest is that the winning position will look something like this one. We need a very high percentage of gammon wins, but the opponent has to have enough wins to take. XG can't handle this family of positions. What setting would we need that we could trust even a teeny little bit with this? 5-ply rollouts? That would take too long on my machine.
Point well taken, Sean.
I had suggested using 3-ply rollouts as a submission/starting point because I did not want to tax bgonliner's CPU time, and the use of "snakes" (15-checker rolling primes) had not occurred to me. With that family of positions, yes -- rollouts of 3-ply, 4-ply or even 5-ply are unreliable.
Your 3-ply rollout, which is copied above, alleges that the position falls inside an enormous TG/T window. However, for the same position and at the same score (x = 37), I did a 4-ply rollout (3k), and a 5-ply rollout (only 220 trials, but still delayed this posting by several hours), and they indicate huge failure -- a bust.
Here is a data summary of the rollouts:
.....3-ply (2k)....NR 1.149, T 0.769...+0.056, +0.098
..+0.033, +0.078
.....4-ply (5k)....NR 0.726, T 0.783
.....5-ply (.2k)...NR 0.651, T 0.522 ..+0.112, +0.236[I also did a 4-ply money (x = infinity) rollout of 6k (not posted), which yields NR .646, T .875, with confidence intervals of +.043 and +.151.]
Details of the 4-ply and 5-ply rollouts can be found below.
Therefore, I'll implement a rule for this endeavor: NO SNAKES! I seriously doubt a snake position can produce the highest "x," though we have no way of knowing until bots can reasonably handle them.
My solution (not yet posted/revealed), which I'm still hoping someone can beat, does NOT employ a snake, has much lower ci's, and my "x" is not remotely as high as 37. :). That position is supported as TG/T by a 4-ply rollout (and its 3-ply rollout is not way off the mark).
Nack
White is Player 2
score: 0
pip: 21337 point match pip: 41
score: 0
Blue is Player 1XGID=-EED----c-ccccA-----------:1:1:1:00:0:0:0:37:10 4-ply rollout
Analyzed in Rollout No redouble Redouble/Take Player Winning Chances: 58.10% (G:51.71% B:0.03%) 53.47% (G:47.12% B:0.02%) Opponent Winning Chances: 41.90% (G:3.97% B:0.11%) 46.53% (G:4.25% B:0.23%) Cubeless Equities +0.686 +1.136 Cubeful Equities No redouble: +0.726 (-0.057) ±0.033 (+0.694..+0.759) Redouble/Take: +0.783 ±0.078 (+0.705..+0.861) Redouble/Pass: +1.000 (+0.217) Best Cube action: Redouble / Take Rollout details 3888 Games rolled with Variance Reduction.
Dice Seed: 76568815
Moves and cube decisions: 4-plyDouble Decision confidence: 90.5% Take Decision confidence: 100.0% Duration: 6 hours 01 minute eXtreme Gammon Version: 2.10, MET: Kazaross XG2
White is Player 2
score: 0
pip: 21337 point match pip: 41
score: 0
Blue is Player 1XGID=-EED----c-ccccA-----------:1:1:1:00:0:0:0:37:10 5-ply rollout
Analyzed in Rollout No redouble Redouble/Take Player Winning Chances: 55.69% (G:50.13% B:0.05%) 52.12% (G:47.51% B:0.03%) Opponent Winning Chances: 44.31% (G:5.17% B:0.36%) 47.88% (G:4.90% B:0.25%) Cubeless Equities +0.605 +1.073 Cubeful Equities No redouble: +0.651 ±0.112 (+0.539..+0.763) Redouble/Take: +0.522 (-0.129) ±0.236 (+0.287..+0.758) Redouble/Pass: +1.000 (+0.349) Best Cube action: No redouble / Take Percentage of wrong pass needed to make the double decision right: 21.2% Rollout details 220 Games rolled with Variance Reduction.
Dice Seed: 58068548
Moves and cube decisions: 5-plyDouble Decision confidence: 83.3% Take Decision confidence: 100.0% Duration: 9 hours 26 minutes eXtreme Gammon Version: 2.10, MET: Kazaross XG2
|
BGonline.org Forums is maintained by Stick with WebBBS 5.12.