| |
BGonline.org Forums
major disagreement between GnuBG 0-4 ply and rollout...or, am I doing it wrong?!
Posted By: Timothy Chow In Response To: major disagreement between GnuBG 0-4 ply and rollout...or, am I doing it wrong?! (bb766)
Date: Friday, 17 May 2013, at 11:58 p.m.
With some experience you'll develop a sense for when a rollout is likely to agree/disagree with the initial evaluation. Most of the time they'll agree. In general, cube decisions are more likely to lead to disagreement than checkerplay decisions are, because an accurate estimate of absolute equity (and not just relative equity) is required for a cube decision. Also, if it's a position where one side has to perform some task that the bot is known to be poor at, then you can expect discrepancies. In your example, White's containment play is difficult for the bot.
As a general piece of advice, I'd recommend never "trusting" the bot in the sense of treating it as an oracle. Instead, treat it as a tool for helping you understand the game. When the bot says you're wrong, come up with a theory as to what's going on. Test it by cross-examining the bot with variations of the position. If you still can't make sense of what it's telling you, then ask others for help, or just give up and move on to the next problem. You'll do better this way than by just blindly accepting that the bot's favored play is "correct."
As for the rollout statistics, the pseudocubeless win/gammon/backgammon figures are explained here. Confidence values are explained here. If these aren't sufficiently clear then feel free to ask more questions.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.