BGonline.org Forums

XGR+ RO, more candidates

Posted By: Michael Depreli
Date: Wednesday, 17 December 2014, at 4:29 p.m.

In Response To: XGR+ RO, more candidates (Timothy Chow)

I've studied the position a bit more now and here's what I've found:

After for example a non homeboard blotty play like 8/4 6/5 XG seems to play all the replies quite well on all settings. However after 6/5 6/2 it has a nightmare. So much so that it misplays EVERY roll on 1-ply most of which are whoppers with cheese by leaving a blot on the 16 point. It doesn't even manage to play the disengaging rolls 55 and 66 correctly.

To make matters worse for rolls 33, 22 and 11 the best plays don't even make it to 1-Ply! hence no settings will find the best play. Some allowance will have to be made that the correct moves to these three rolls don't make it to the rollout results.

Due to 1 ply misplaying this so badly this needs to be rolled out using gigantic search interval at 3-ply otherwise most of the best plays won't make it through.

Having already posted a 4-ply rollout I've now actually noticed that for the replies to 6/5 6/2 4-ply seems to play slightly worse than 3-ply I'm guessing due to odd / even ply bot confusion.

GNUbg seems to understand this position better and because it's possible to set the search intervals yourself it can be set up to avoid the pitfalls of XG for the rollout. A rollout had 8/4 6/5 best but only 0.003 over 6/5 6/2.

Michael