|
BGonline.org Forums
Rollout Result Posted
Posted By: sita In Response To: Rollout Result Posted (Paul Weaver)
Date: Thursday, 9 January 2014, at 3:57 p.m.
Is a 1297 rollout better than a 1296 rollout? I think so. If increasing the number of trials in a rollout by one never increases its accuracy, then to be consistent, you have to believe that a rollout of 2592 is no better than a rollout of 1296.
Well, I can believe that trial 1328 (1296+36) adds more accuracy to the result, than trial 1297 does. So, for the same price (every trial costs the same computing power in average), I receive more value. And that's why I try to optimize the value I get for the computing power, by stopping at a multiple of 36. If I didn't believe this, I could as well switch off variance reduction.
Profound concept: If a baby never grows from one second to the next, then it never grows over the course of its entire life.
And if pulling out a single hair won't change the fact that somebody has full hair, everybody has full hair.
|
BGonline.org Forums is maintained by Stick with WebBBS 5.12.