Quote:
Originally Posted by Robertie
I didn't actually consider this play at first, but I should have, because it does have a point. It's the only play that doesn't hit and blocks 1-1, thereby giving White a cracking number. However, the cost is high (a ruined board). I did some 3-ply rollouts matching this play against 8/4(2) and it did significantly worse (although better than any hitting play), so I'll stick with my original recommendation.
I did also some rollout. Your explanation of 11 from white in the next gives a little light. I did a 10k RO with 3ply checker/cube, variance reduction. 8/4(2) came in as 6th, near whopper range. And to decide between first and second, we have to extend to even 40 or 80k, if we look at the confidence interval. I played several deep backgames from both sides against XG and he played in my opinion like a human. He tries to solve the game at protobackgame otherwise, but also has no fear to go in full back game modus. And then, i was really stunned about his understanding of timing.
Position 83a is as deepest as i could imagine to stand in a backgame. So i think the philosophical question (or epistemological) is, what can we really know for certain in complex backgammon positions. I think, if we use equity estimations from bots as a reference, then this is the archimedic point. But if we are not sure, that there is a flaw in the bot estimation and the move doesn't fit in our conceptual framework of the game, then we have to admit, that there is no certainty. My opinion is: there is so much order from noise, which can't be analyzed, that only with a really massive simulation we can see a little light. And if we think, that the bots are way off in deep back games, then this is south pole exploration. And for sure a "goal line stand". That is in my amygdala, since i've read advanced backgammon.
XGID=----A-CaBA-aaB---bbbbbbDB-:1:1:1:22:0:0:3:0:10
1. Rollout¹ 8/4 6/4(2) eq:-0,812
Player : 29,97% (G:0,68% B:0,03%)
Opponent: 70,03% (G:44,15% B:12,11%)
Confidence: 0,010 (-0,822<E<-0,802)
Duration: 5 hours 45 minutes
2. Rollout¹ 9/7* 7/5 6/2 eq:-0,815 (-0,003)
Player : 31,73% (G:0,98% B:0,05%)
Opponent: 68,27% (G:45,27% B:16,11%)
Confidence: 0,014 (-0,829<E<-0,801)
Duration: 9 hours 33 minutes
3. Rollout¹ 6/4(2) 6/2 eq:-0,836 (-0,024)
Player : 28,23% (G:0,73% B:0,03%)
Opponent: 71,77% (G:43,76% B:10,38%)
Confidence: 0,009 (-0,845<E<-0,827)
Duration: 4 hours 31 minutes
4. Rollout¹ 8/2 6/4 eq:-0,853 (-0,041)
Player : 27,91% (G:0,70% B:0,03%)
Opponent: 72,09% (G:44,44% B:10,50%)
Confidence: 0,010 (-0,863<E<-0,843)
Duration: 4 hours 28 minutes
5. Rollout¹ 8/6 8/2 eq:-0,877 (-0,065)
Player : 27,05% (G:0,55% B:0,03%)
Opponent: 72,95% (G:44,83% B:10,26%)
Confidence: 0,009 (-0,886<E<-0,868)
Duration: 4 hours 00 minute
6. Rollout¹ 8/4(2) eq:-0,888 (-0,076)
Player : 26,62% (G:0,51% B:0,02%)
Opponent: 73,38% (G:45,16% B:10,00%)
Confidence: 0,008 (-0,896<E<-0,880)
Duration: 3 hours 05 minutes
¹ 10368 Games rolled with Variance Reduction.
Moves and cube decisions: 3 ply
eXtreme Gammon Version: 1.21
Last edited by higonefive; 11-21-2010 at 09:05 AM.