| |
BGonline.org Forums
A possible correction for what you posted at 2+2
Posted By: Michael Depreli In Response To: A possible correction for what you posted at 2+2 (Neil Robins)
Date: Sunday, 6 March 2011, at 12:16 p.m.
Yesterday I was looking at a game where three missed doubles by my opponent weren't picked up by XGR+ because it didn't look beyond 1-ply.
Note about cube
The search interval affects only checker play. However when using eXtreme Gammon 3-ply, cube decisions are first analyzed in 1-ply (cubeful). If the decision of not doubling is 0.200 more than doubling, the analysis stops. If less than 0.200 then a 3-ply is preformed.
In the process of this study we check also the effect of the 0.200 threshold by forcing all cube decisions to go to 3-ply.
On the sample set, there are 21434 non obvious cube decisions (non obvious meaning less than 0.200 equity between No Double and Double)
Using the 0.200 threshold results in 3 cube error in match play (0.019, 0.019 and 0.017) and 3 in Money games (0.052, 0.044 and 0.033)
This represents an error every 3692 decisions, or every 1901 games. The estimated cost in Elo term is 0.06.
It could be an idea of improvement to apply the search interval factor to the 0.200 threshold. In this case the large interval (0.300 thresholds) would result in a single error, while the huge (0.400 threshold) would catch them all. But again, the gain is very minimal.
Note: Version 1.14 did introduce the fact that the search interval modify the cube threshold. So with a Huge interval the threshold to use 3-ply is 0.400.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.