Backgammon Snowie

Snowie

Checker-play-according-to-score bug

From:   Peter Schneider
Address:   schneiderp@gmx.net
Date:   5 June 2001
Subject:   Cubeful rollout bug
Forum:   rec.games.backgammon
Google:   3b1ff47d$0$26420$64889074@news.comundo.de

Hi all, Has anybody else had bugs related to cubeful Snowie rollouts? I have useless results nearly every time. I think that the bug occurs only if -- the rollout is 3-ply *cubeful*, cube 3-ply (the bug does not appear in cubeless rollouts, not sure about the 3-plies) -- Some higher number of games are already performed in the rollout. The bug does sometimes not occur before a number of games are rolled out. This number seemed to vary but was always smaller than the necessary number of games. The bug seems to be position independent; I have not found a way to avoid it yet (other than un-checking the cubeful checkbox). Anybody else? Greetings, Peter aka the juggler

Dean Gay writes:

I seem to remember that one of the conditions necessary for the bug to appear is that it be a FULL, not a truncated rollout. I'm not entirely sure this is the case, but that's what I recall.

Christopher D. Yep writes:

Hi Peter, If you're using Snowie 3, I believe this problem specifically only occurs when you perform a rollout with checker play according to score (this specifically also means that you are performing a cubeful rollout, otherwise this checkbox won't even be available). Specifically sometimes the rollout will show one side winning gammons when gammons are impossible. Other times one side will be shown winning *zero* gammons, when clearly gammons are possible (or even likely). I don't often perform "checker play according to score" rollouts since the rollouts take considerably longer than normal cubeful rollouts (with checker plays based on cubeless money equity) without improving accuracy very much. However the general consensus among several users I have talked to is that the overall *equities* are likely correct even though the breakdown (backgammons, gammons, normal wins) is incorrect. For example, here are the rollout results that you reported on June 7, 2001: Rollout Money equity: 0,399 0,0% 6,0% 71,6% 28,4% 9,4% 0,0% 95% confidence interval: - money cubeless eq.: 0,399 ą0,007, - live cube no redouble: 0,883 ą0,008, - live cube redouble take: 0,851 ą0,009. The above hypothesis states that the money cubeless equity is 0.399 despite the fact that the displayed numbers (0.0%, 6.0%, 71.6% 28.4% 9.4% 0.0%) actually imply a 0.415 money cubeless equity. Similarly, the hypothesis states that the 0.883 and 0.851 numbers are in fact valid. One hypothesis is that Snowie always (accurately) keeps track of the overall equity, but that there is some type of display flaw that causes it to erroneously report some of the wins as gammons or vice-versa. To test this hypothesis, run the same position *without* checking "checker play according to score." In the position that you posted on June 7, 2001 (5-away/5-away with cube on 2), this will likely not make much difference in the final results. See if the equity comes close to equity that Snowie reports with "checker play according to score." Chris

Did you find the information in this article useful?

Do you have any comments you'd like to add?