| |
BGonline.org Forums
Why does XG ding no double after 21$ at 2a 2a?
Posted By: Chris Yep In Response To: Why does XG ding no double after 21$ at 2a 2a? (Bob koca)
Date: Sunday, 17 April 2016, at 4:07 a.m.
If you look at the Dice Distribution (Ctrl+Alt+D) at 3-ply, 4-ply, or 5-ply, you can see that XG doesn't think there are any market losers.
Nevertheless, even if there are no market losers, XG can still conclude that ND is a small error. In its cube analysis XG doesn't specifically look for market losers at all. It simply estimates (evaluates) the equity of ND, estimates (evaluates) the equity of doubling, and then compares these two numbers. In general, evaluations are subject to all sorts of possible errors due to misplays or misevaluations in the intermediate plies or at the horizon. Since the positions don't always transpose after the two lines (ND and D/T), the evaluations may not be exactly the same once we factor in these possible errors in the analysis. Also, as David notes in his response, there is a chance that "after ND, XG erroneously waits again next turn in some sequences."
To the above notes, I can also add the following comments:
1-ply, 2-ply: Even if XG were programmed to specifically look for market losers, these evaluation levels don't look ahead far enough for XG to see them.
3-ply to 7-ply: Even if there aren't any market losers, XG can still conclude that ND is an error based on misplays, misevaluations, etc. (as noted above).
XGR, XGR+, XGR++: Even if it's an optional-double, XG will often conclude that D is better than ND (or vice-versa). The reason is that XGR, XGR+, and XGR++ are actually rollouts. According to Xavier in a BGO post a few years ago, the random seed depends on a hash of the initial position. In particular this means that the ND and D/T lines will use different random seeds (they have different XGIDs). So, even if it's a clear optional-double, XGR(+)(+) will often say that D/T is better than ND.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.