|
BGonline.org Forums
cubeless RO
Posted By: Rick Janowski In Response To: cubeless RO (Bob Koca)
Date: Tuesday, 16 July 2013, at 7:42 p.m.
Yes we concur in the calculation and assumptions made.
Its an interesting point you made about distance to travel from the take point to the doubling window - it is a lot closer like you say. Where there are are no gammons the distance is roughly 22.5% at the point of taking to say 65% (in the middle of the doubling window) - a distance of 42.5%. For positions where W = 1.00 and L = 1.87 (as it is for the taker) the distance is about 31%. However, there is another factor which will tend to reduce efficiency - the width of the doubling window. In gammonless games the width is 30% (50% to 80%), but in this position its only 20% (65% to 85% roughly). So the two effects may balance out. I agree however that certain positions may have higher efficiency than races, but the values wont generally be a lot higher than in long races.
Where XG has a good handle on the raw equity of a position in its evaluations, which it normally does, its cube handling in its rollouts normally leaves little to be desired. Problems when they occur are when its evaluations of raw equity are way off the mark. Having the neural net learn cube effiencies won't help that much in such situations in my view.
|
BGonline.org Forums is maintained by Stick with WebBBS 5.12.