| |
BGonline.org Forums
'variance reduction' versus 'MWC error totals'
Posted By: Maik Stiebler In Response To: 'variance reduction' versus 'MWC error totals' (Johan Segers)
Date: Friday, 9 September 2011, at 8:11 a.m.
1. Yes. But specify what, exactly, you mean by MWC. If you mean the value that a perfect player would achieve in a given situation playing against himself, everything is fine, but of course you can't compute that with current technology. You may be interested in this post (and the thread leading up to it) for a VR method with a different version of MWC that works practically and theoretically. In that version, you can occasionally have negative error rates! Still, it's definitely the version of VR that I'd recommend for future VR tournaments (in the very near future, it would probably be too cumbersome to implement).
2. The problem with MWC as given by the evaluation function of a real bot is that the MWC before the roll is generally not equal to the appropriate average of the MWC's after all possible rolls (Douglas Zare coined the term "Bot confusion" for that deviation). Thus, you can't directly compute a valid measure of luck from it. Thus, when bots calculate luck as "(MWC after the roll - MWC before the roll)", they don't use their estimated "MWC before the roll", but the explicitly computed average of the "MWC after the roll". And there your conservation law breaks down, because different flavors of MWC are used in each step.
3. Yes.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.