| |
BGonline.org Forums
BQ: Hit or not hit? *ROLLOUT* inconclusive
Posted By: Nack Ballard In Response To: BQ: Hit or not hit? *ROLLOUT* inconclusive (leobueno)
Date: Saturday, 23 April 2011, at 11:24 p.m.
Is this worthy of a deeper rollout?
By "deeper," do you mean more trials? If I'm remembering how to interpret the details correctly, your result of [p P20] /7 tells us that if you roll out these two plays on these settings a billion times, p is 95+% to end up better than P by a margin between .006 to .036. [That assumes that GnuBG doesn't underestimate variance; otherwise (as I suspect) it is wider.]
If all you care about is the answer to the question, "Is p the best play," then most people would consider 7k trials with a .020 margin to be passable.
If what you want is a reliable margin (as is generally my goal) for fairly "close" plays (say a top-plays margin within .03) of an early game position, I usually prefer to see 15k trials, though as long as there is at least 5k (even if the top plays are in a several-way dead heat, that's okay), I am comfortable that I won't be badly "misled" by the whims of variance.
Nack
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.