by Phil Simborg
14 December 2007
A lot of people say that backgammon is now fully in the computer age. They say that thanks to the perfections of Snowie, Jellyfish, and GNUBG, there no longer is any mystery or debate about what is right: we simply have to ask the bots and they will tell us.
I think these people are right - but they are only right about 97 percent of the time. I believe there are still many plays and situations that the bots simply do not fully understand.
|The rest of this article (3.44 K) is premium content. Please subscribe below.|
Snowie may not get it but GNU does. When in doubt, use 2 bots.
I show the same eval as you do on Snowie but GNU shows:
1. Cubeful 2-ply 20/16 20/14 Eq.: +1.188 0.895 0.427 0.025 - 0.105 0.019 0.000 2-ply cubeful prune [world class] 2. Cubeful 2-ply 24/14 Eq.: +1.187 ( -0.001) 0.909 0.394 0.017 - 0.091 0.011 0.000 2-ply cubeful prune [world class]
1. Cubeful 4-ply 24/14 Eq.: +1.183 0.909 0.392 0.018 - 0.091 0.013 0.000 4-ply cubeful prune 2. Cubeful 4-ply 20/16 20/14 Eq.: +1.173 ( -0.010) 0.891 0.424 0.025 - 0.109 0.020 0.001 4-ply cubeful prune
10 times is a very small samplle. What happens those times white moves the 2 checkers out and black rolls 1-1, 3-1 or 3-3? Did any of the 10 games include one of these rolls? Further, even if it did, these rolls would show up ~1.1 times in an average 10 games each way. That extra 10% is a very large margin in this situation.
Perhaps those 4 rolls are cause for the decidedly "wimpy" move 24-14 that a snowie 3-ply 11-depth 1080 game sample favors by 0.053.
Just because most pros favor it and it "feels" better (even to a sampling of pros - or all pros for that matter) doesn't necessarily make it the correct play.
Not only that, but perhaps at the time we're able to send the black checker back, we actually give black more timing?
Not only that, but the gnu 4-ply supports the 24-14.
oh.... btw... I would have made the 20-14,20-16 play myself.
You must be signed in to post comments.