| |
BGonline.org Forums
A study: How confident can we be of the confidence interval!
Posted By: eXtreme Gammon In Response To: 4-ply queue, Nov 20 update (eXtreme Gammon)
Date: Sunday, 21 November 2010, at 2:44 a.m.
Ah.. i found the data.
I actually made 1068 3-ply Rollout of 1296 games using different seed.
conf theo actual 99.90% 0.10% 0.09% 99.50% 0.50% 0.37% 99.00% 1.00% 0.94% 98.00% 2.00% 1.69% 97.00% 3.00% 2.81% 96.00% 4.00% 4.03% 95.00% 5.00% 4.68% 90.00% 10.00% 9.64% 80.00% 20.00% 17.98% 70.00% 30.00% 28.93% 60.00% 40.00% 39.33% 50.00% 50.00% 48.60% 25.00% 75.00% 74.34% 10.00% 90.00% 89.89% Conf is the confidence interval Theo is the percentage of RO which CI should not include the actual value of the RO (as determined by the overall 1068x1296=316,128 games [CI is +/- 0.001]) is the actual number of RO which did not include the actual value of the RO Graphically here is how it looks:
![]()
Obviously for large CI (99% and more) I probably need more data, but overall results fit the theory. Also this study was mode on one single position:
is Player 2
score: 1
pip: 1537 point match pip: 161
score: 0
is Player 1
XGID=--aBa-D-BA--dC---bbbc-BA--:0:0:1:63:0:1:0:7:10 to play 63
1. Rollout1 13/4* eq: -0.700
Player:
Opponent:34.07% (G:10.03% B:0.61%)
65.93% (G:21.21% B:1.44%)Conf: ± 0.002 (-0.702...-0.698)
Duration: 15 hours 22 minutes1 206853 Games rolled with Variance Reduction.
Dice Seed: 801
Moves and cube decisions: 3 plyeXtreme Gammon Version: 2.00.150.alpha, MET: Rockwell-Kazaross
Comments are welcomed.
Xavier Dufaure de Citres
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.