|
BGonline.org Forums
Play A "loses more gammons"
Posted By: Matt Ryder In Response To: Play A "loses more gammons" (Timothy Chow)
Date: Monday, 18 January 2010, at 6:06 a.m.
Matt Ryder wrote:
In making the decision to drop, I need to know my gammon chances were I to take.
Timothy Chow wrote:
Of course. But I don't understand why you want to know your cubeless gammon chances if you were to take. Don't you want to know your actual gammon chances if you were to take?
The way GNU evaluates a cube decision ATS is to roll the position out both ways, with "no double" and "double, take." So you'll get your gammon chances, contingent on your taking, by looking at the stats for "double, take."
Could you point me in the direction of your information source? I have struggled to find a clear unambiguous description of exactly how this works.
Are you saying that the bot rolls out a branching decision tree for the cube and every possible future cube which may occur before its truncation point? That sound prohibitively costly in processing time. If it does indeed do this, how do you propose it collate all this information into a simple w/g/b output?
I had always assumed a GNU cubeful rollout simply truncates when it reaches a clear pass situation. This seems to be Douglas Zare's conclusion in his GammonVilage article about truncation depth:
"Full rollouts may have some dependence on absolute evaluations. Gnu Backgammon truncates the rollouts when a double is passed. If its evaluations are not accurate when there is a pass, then full rollouts may converge to the wrong value. An indicator of this is a significant difference between the bg/g/w distributions differ significantly for Double/Take and No Double for a money play position. Typically, evaluations are a larger component of the distribution after No Double."
(Italics mine.)
|
BGonline.org Forums is maintained by Stick with WebBBS 5.12.