| |
BGonline.org Forums
Pittsburgh recube action
Posted By: Frank Berger In Response To: Pittsburgh recube action (Matt Cohn-Geier)
Date: Thursday, 26 February 2009, at 6:52 p.m.
Well the funny thing is that we are talking about the very same neural net. The odd/even thing is a total mystery to me in the amount seen here. When I experience large swings with BGBlitz (which is rare) I assume that it doesn't understand the position very well, but that adding a ply descreases the quality of the estimation is beyond my understanding....
BTW if anyone is interested:
BGBlitz (TAchiAI 3):
1-ply Green 68.1 0.0 0.0 - 31.9 4.5 0.3 Equity (cubeless): 0.314 (cubeful) : 0.524 ND/T
2-ply Green 68.9 0.0 0.0 - 31.1 4.6 0.5 Equity (cubeless): 0.326 (cubeful) : 0.526 ND/T
3-ply Green 71.3 0.0 0.0 - 28.7 3.8 0.2 Equity (cubeless): 0.385 (cubeful) : 0.531 D/T
all cubes are borderline according to BGB; BGB counts the ply starting at 1
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.