| |
BGonline.org Forums
a bias in rollouts
Posted By: Keene In Response To: a bias in rollouts (Tom Keith)
Date: Friday, 2 May 2008, at 1:39 p.m.
Theoretical question - could this perhaps explain the overestimation that GNU (And Snowie?) gives to holding games, and or backgames?
Also, I think whats being said is that the average equity on the JSD bell curve is whats being calculated rather than an absolute number, so if you get one play thats way deep on the JSD curve, and another thats less deep, then GNU is merely averaging those based on the assumption that the bell curve itself linear?
For todays bots, I dont see this as a problem, however, as BG gets ever closer to being absolutely solved, then this will become more of a problem, in terms of absolute equity at least. I suppose the solution is to come up with a way to calculate absolute equity rather than utilizing a confidence interval - which will invariably have small errors built into it.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.