| |
BGonline.org Forums
GNU vs XG walking in a prime
Posted By: Daniel Murphy In Response To: GNU vs XG walking in a prime (Timothy Chow)
Date: Friday, 18 September 2009, at 12:38 p.m.
This way we would know with much greater certainty than we do now how close to optimal the bots are for this problem.
What we do already know is that if the prime is no further advanced than the 5-11 points, both Snowie and Gnubg are awful at prime-walking. Awful: unforgiveably worse than a human expert thinks he could do, allowing for the possibility that the human is not right! Jellyfish, according to rollouts that David Montgomery did, was actually better than the new bots in such long way to go primes, versus one checker, with no chance of getting another.
It's pretty clear that if the walk is long enough then the prime is sure to crack (with a 4-4 for example) and allow the checker through.
Well, just two consecutive 4-4's and/or 5-5's will break most primes. But the checker still has to get out. If the prime is a long way from home, the temporary defect doesn't matter at all. With correct play, relevant positions aren't reached until the home board is partially filled. Manually -- and it's a useful exercise -- you could roll a prime from home board to home board all day long without ever letting a checker escape. If we hypothesize (reasonably, I think) that no initial full-prime position should be much worse than, say, 3% worse than a closed board, you can see from Stick's, David's and my Snowie, Gnubg and Jellyfish rollouts posted at gammonu and bgonline, how far the bots are from perfection.
As long as we're discussing old problems with fresh perspectives, who do you think is the money game favorite in the proposition described here:
http://www.bkgm.com/rgb/rgb.cgi?view+1127
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.