| |
BGonline.org Forums
Surprising Fourth Roll 63: 65R-64P-61P-63 *ROLLOUT*
Posted By: Daniel Murphy In Response To: Surprising Fourth Roll 63: 65R-64P-61P-63 *ROLLOUT* (leobueno)
Date: Wednesday, 27 July 2011, at 9:02 p.m.
What can you say about this rollout?
Double (approximately) the Standard Error of each play to get its 95% confidence interval. For 24/15, rollout equity is -0.370 ± 0.008 so the interval is -0.378 to -0.362. For 24/21 13/7, rollout equity is -0.373 ± 0.008 so the interval is -0.381 to -0.365. So the overlap of the intervals is large, running from -0.378 to -0.365.
And so you can say that in a Gnubg 2-ply rollout the plays are close but it is uncertain which play is better. As of now 24/15 is better by 0.003. But if you extended the rollout, it wouldn't be surprising, for example, if the equity of 24/15 turned out to be -0.365 and the equity of the other play turned out to be -0.381, which would make 24/15 better by 0.016. Neither would it be surprising if the equity of 24/15 turned out to be -0.378 and of the other play -0.365, which would make the other play better by 0.13. As "not surprising" results I've chosen the four end points of your rollouts. But it's only less probable, not guaranteed, that the equities of your final results will lie within the current intervals.
Should it be extended?
(a) If you want. Are you interested enough to know whether, according to this bot, your error with either play is very likely to be less than, say, 0.020?
(b) What Stick said about the score.
(c) For really long rollouts, XG is megatimes faster (and, I hear, better).
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.