[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

MABC Study #8 ROLLOUT - plus VARIANT

Posted By: Nack Ballard
Date: Saturday, 26 August 2023, at 3:42 a.m.

In Response To: MABC Study #8 ROLLOUT (Jason Lee)





White is Player 2

score: 4
pip: 152
11 point match
pip: 154
score: 1

Blue is Player 1
XGID=-a-aBBB-C---cD---bce----B-:0:0:1:52:1:4:0:11:10
Blue to play 52

1.Rollout113/11 8/3* eq: +0.639
Player:
Opponent:		61.43% (G:23.26% B:1.12%)
38.57% (G:11.60% B:0.96%)		Conf.: ± 0.006 (+0.633...+0.644) - [72.6%]
Duration: 2 hours 04 minutes
2.Rollout18/3* 3/1* eq: +0.636 (-0.003)
Player:
Opponent:		60.59% (G:24.70% B:0.71%)
39.41% (G:11.05% B:0.73%)		Conf.: ± 0.005 (+0.631...+0.641) - [22.2%]
Duration: 2 hours 04 minutes
3.Rollout124/22 8/3* eq: +0.633 (-0.005)
Player:
Opponent:		61.24% (G:22.42% B:1.00%)
38.76% (G:10.70% B:0.66%)		Conf.: ± 0.005 (+0.628...+0.639) - [5.3%]
Duration: 2 hours 07 minutes
1 10368 Games rolled with Variance Reduction.
Dice Seed: 19289287
Moves: 3-ply, cube decisions: XG Roller

eXtreme Gammon Version: 2.10, MET: Rockwell-Kazaross

For opening rollouts, huge interval takes about 1.75 times as long to roll as normal interval. In terms of ply, XGR+ takes about 80 times as long to roll as 3-ply (or about 11 times as long as XGR, or about 4.5 times as long as 4-ply) yet in parallel testing do not produce significant differences in the margin. I believe that rolling with high ply is a misspent time investment except for backgames and pure containment positions.

For most positions, I use either 3-ply or (usually) the popular 3/R (3-ply checker, XGR cube) setting. In this way, I can iterate 5k (i.e., 5184) trials or more and still save a lot of time.

To my taste, the variance in 1k (i.e., 1296) trials is too high for early game positions. Even when 100% conf. is achieved for the best play, it is harder to draw firm conclusions about candidate comparisons from the high-variance margins.

Ths rollout displayed above of the original MABC Study #8 position is 10k (i.e., 10,368 trials), using the standard 3/R setting.

The rollout displayed below (rolled to 5k) is a variant with a 1-pip difference: a White spare is moved from her 7pt to her 8pt. The ranking of plays is reversed, though with no drastic changes.





White is Player 2

score: 4
pip: 153
11 point match
pip: 154
score: 1

Blue is Player 1
XGID=-a-aBBB-C---cD---cbe----B-:0:0:1:52:1:4:0:11:10
Blue to play 52

1.Rollout124/22 8/3* eq: +0.645
Player:
Opponent:		61.43% (G:22.60% B:0.99%)
38.57% (G:10.56% B:0.65%)		Conf.: ± 0.007 (+0.638...+0.652) - [98.2%]
Duration: 52 minutes 05 seconds
2.Rollout18/3* 3/1* eq: +0.634 (-0.011)
Player:
Opponent:		60.62% (G:24.73% B:0.72%)
39.38% (G:10.95% B:0.86%)		Conf.: ± 0.008 (+0.627...+0.642) - [1.7%]
Duration: 55 minutes 45 seconds
3.Rollout113/11 8/3* eq: +0.630 (-0.016)
Player:
Opponent:		61.10% (G:23.48% B:1.14%)
38.90% (G:11.62% B:0.97%)		Conf.: ± 0.008 (+0.622...+0.637) - [0.1%]
Duration: 53 minutes 12 seconds
1 5184 Games rolled with Variance Reduction.
Dice Seed: 4300210
Moves: 3-ply, cube decisions: XG Roller

eXtreme Gammon Version: 2.10, MET: Rockwell-Kazaross

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

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