I don't have a head-to-head comparison handy. If I find it, I'll let you know.
However, these are the results of testing the current gnubg (0.15) weights
against our benchmark database, which I ran some time ago.
Contact benchmark. One point of total error corresponds to 0.0001565 points
per game, according to Øystein's regression analysis.
# Correct# Wrong % Tot. err. Avg. Err Err rate(per
move) Error Rate (points per game)
0-ply 74524 32961 30.67% 1122.71 0.03406177 0.010445271
0.175704115
1-ply 76040 31445 29.26% 993.33 0.031589411 0.009241559
0.155456145
2-ply 79480 28005 26.05% 790.55 0.028228748 0.007354943
0.123721075
Subtracting, I would expect 1-ply vs 0-ply to win 0.02 points per game.
-- Ian
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Mark
Higgins
Sent: 30 January 2012 01:28
To: [email protected]
Subject: [Bug-gnubg] 1-ply vs 0-ply equity improvement
Does anyone know roughly how much of a cubeless money equity improvement 1-ply
enjoys over 0-ply? (And 2-ply vs 1-ply.) I googled around for a while but
couldn't find anything.
_______________________________________________
Bug-gnubg mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/bug-gnubg
_______________________________________________
Bug-gnubg mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/bug-gnubg
