|Subject:||[Bug-gnubg] Gnu copy Bug|
|Date:||Sun, 9 Nov 2003 11:07:15 -0800|
I did a rollout with 2 plays. After the rollout finished a clicked on copy and gnu crashed. Tested this a few other times with different rollouts and got the same results.
Question: What should I set "Stop when results is accurate" on the rollouts menu if I want to stop when I am 99% sure the results won't change (assume 360 rollouts if that matters)? Same question for 95%.
I found this position where the 3 ply evaluation is very different than a rollout (about 10%).
Side on roll has 2-23 1-20 1-14 1-11 2-6 2-5 2-4 2-3 2-2
Side not on roll has 1-24 5-1
What is interesting here is that the best position the side on roll can get to is a close out with spares on the 6,5,4 and opp. checker on the bar. In this 'best position' GNU 3 play evaluation is almost the same as in the original position. So GNU does not 'see' much chance that it won't achieve the 'best position'., but that happens often.
Why does this happen? Can anything be done to correct this problem?
|[Prev in Thread]||Current Thread||[Next in Thread]|