| |
BGonline.org Forums
Agg calculation
Posted By: Nack Ballard In Response To: Agg calculation (Matt Ryder)
Date: Friday, 7 January 2011, at 4:46 a.m.
Note that in Agg assessments of AtS rollouts, I'd be inclined to heavily weight against Snowie (or remove it from the mix altogether).
Agreed. I ignore Snowie's AtS results (notice they don't appear in my data summaries) except in cases where it's the only ones that exist and seem worth mentioning, and even then I regard them as "AA" (artificial AtS) or "Fake AtS" results.
A question: Does an XG 4-ply rollout not necessarily always result in an accuracy improvement over an XG 3-ply rollout with a corresponding number of trials? If it does, then why factor the XG 3-ply number in the agg at all?.
It's a fair question. I may be slower than most to discard bots' results other than those of the newest and brightest. I've seen too many "better-bot" or higher-ply results later re-overturned, and there could be a parity (odd/even ply) issue not yet discovered.
Last but not least, there is the matter of variance: nearly all positions that have been rolled on XG 4-ply have been rolled on XG 3-ply with many more trials. Consider opening 41 [S $23] ~46, [S $6] /92, [S $8] <173, [S $22] ^25. What are we to make of that? I'd rather allow all the results to weigh in unless/until we both have clear evidence that the higher ply wipes the floor with the lower ply and the higher ply's candidate plays have been rolled at least 46k or so.
I'm delighted that Neil is posting results with even stronger settings than previous ones. Even so, consider that we now have
Opening 21
[$ S8] ~80,
[$ S19] /92,
[$ S5] <82,
[$ S3] ^57,
[$ S8] "+46(where " is my symbol for XG2 version and + is my symbol for XGRoller+ level).
So, the newest and brightest bot result seems to be telling us that the XG 3-ply result is more accurate than the XG 4-ply result. Hmm... And that Snowie was right all along.
Nack
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.