| |
BGonline.org Forums
Two questions about your Rollouts
Posted By: Daniel Murphy In Response To: Two questions about your Rollouts (mtuhtan)
Date: Monday, 22 August 2011, at 1:20 p.m.
2. If you decide to do them, how can 819 games be enough trials when two moves are close?
Leo's cubeless rollout of position #2 had two plays separated by 0.031 with a reported standard deviation of 0.005. More decimal places would be nice. But we can convert the reported SD's to 95% confidence intervals:
Play 1: -0.451 ±0.0098 -> -0.4608 - -0.4412
Play 2: -0.482 ±0.0098 -> -0.4918 - -0.4722
There's no overlap here. If you want, use 98% confidence intervals:
Play 1: -0.451 ±0.01165 -> -0.46265 - -0.43935
Play 2: -0.482 ±0.01165 -> -0.49365 - -0.47035
Still no overlap. Or 99% confidence intervals:
Play 1: -0.451 ±0.012875 -> -0.463875 - -0.438125
Play 2: -0.482 ±0.012875 -> -0.494875 - -0.469125
Again. So 819 trials is enough for me.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.