| |
BGonline.org Forums
Rollout
Posted By: Hans Marius Eikseth In Response To: Rollout (ah_clem)
Date: Wednesday, 21 November 2018, at 12:49 p.m.
Hi, interesting subject that of the cube models and takepoints!
To my understanding, for any given match score and cube value, the dead cube takepoint and fully live takepoint are two extremes. The bots (Gnu at least) list these, but not the decicion-making takepoint: This you can find (often) after doing a Janowski-interpolation between the two extremes, viz. Actual_TP=lambda*FullyLive_TP+(1-lambda)*Dead_TP where lambda is a parameter between 0 and 1. Standard positions are thought to have a lambda of about two thirds, although as noted by Neil Kazaross here the lambda will probably be lower than this (less effective recube when hit). Bots have jablons they use - in Gnus case this can be gleaned from the manual (or code, as itīs open source). Advanced cube models operate with position-dependent lambdas that varies between players according to how efficient a future recube is likely to be. You can think of it as a probability of giving a strong recube without risk of losing your market (although this is not precise in bot-terms, we need to use ideas to think of such quantitative trade-offs).
Finally, you may calculate the two kinds of takepoints gammonless for any given score, then interpolate them using lambda=0,67 or somesuch value. This interpolated, gammonless takepoint you may adjust using the gammon and backgammon probabilities and prices for the position and cube value at hand.
Good luck in this endeavor!
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.