| |
BGonline.org Forums
Degree of extension
Posted By: Nack Ballard In Response To: 31P-21S-65 — Rollouts — Same bot, same seed, same count, different results (Taper_Mike)
Date: Thursday, 2 February 2012, at 1:53 p.m.
The XG rollout I posted is close to the BGO value of 5.4. The 20k trial count may be insufficient, however, to wring out the variance of plays that are tied under the Ballard-Weaver error scale. Nack, didn’t you post somewhere that you and David are now rolling 61k for XG 3-ply rollouts where the results are within 0.01?
It depends on the position. In the case of first- and second-roll plays for money that are "tied" (within .01 of best) or more usually within .005, I typically like to extend to 62k (i.e., 62208).
What is the value of long extensions? Firstly, they provide a more accurate basis of comparison to data generated by other bots, plies and parameters. Secondly, when such data exists, low-variance equities provide an opportunity for crisper analysis, comparing the margins between plays of similar positions to each other. Thirdly, other small adjustments one might make for skill difference, style, surprise value, etc., might actually swing one's choice of play when the decision is already extremely close.
For me, third-roll (and subsequent-roll) positions don't generally qualify for that level of scrutiny, unless (perhaps) they're being compiled for a book or other document.
Other positions with tied plays for which I might desire long extensions are theoretical curiosities such as T4's.
Nack
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.