| |
BGonline.org Forums
15552-game rollout of 41Z-65R-43
Posted By: Daniel Murphy In Response To: 15552-game rollout of 41Z-65R-43 (Timothy Chow)
Date: Wednesday, 16 June 2010, at 1:57 a.m.
Nack says that for opening rollouts, his experience demonstrates that truncated rollouts are more reliable for a fixed amount of computation time.
Actually, he said he trusts a 1k truncated rollout of an opening play more than a 5k full rollout, and that, apparently, is due to having done multiple 5k-ish rollouts of the same position with different seeds and gotten disparate results.
Well, the truncated rollout is reliant on evaluations at the end of the truncation depth. That's bad. But at least the result will (probably) have a low SE and (probably) be repeatable. If we're comparing such a rollout to, say, a longer full rollout with a difference of 0.010 and SDs of 0.020, which doesn't tell you much, I'd find the shorter, truncated rollout more trustworthy too.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.