BGonline.org Forums

GNU 2-ply RO vc XG 3/3 RO

Posted By: Robert-Jan Veldhuizen (Zorba)
Date: Saturday, 4 July 2009, at 5:21 p.m.

In Response To: GNU 2-ply RO vc XG 3/3 RO (MaX)

red means reduced evaluation, a setting also found on Snowie (like 25%, 33% and 50%) and older GnuBG versions.

It means the third ply (GnuBG 2-ply) of the lookahead doesn't use all 36 combinations of dice rolls, but some fraction of it where some ordering algoritm is used to get a representative sample, equity-wise.

I remember testing these settings on GnuBG and the result was that it worked better for cube decisions than for checker play.

(GnuBG ply notation starting at 0 below)

50% 2-ply worked well for checker play, extremely close to 100% (normal) 2-ply at almost twice the speed. 33% started showing some differences and was less interesting to use therefore, ditto for 25%.

For cube decisions, 50% was almost indistinguishable from 100% and 33% was very good too. Even 25% still produced very good results, 4 times faster, not as good as 100% 2-ply, but still a lot better than f.i. 0-ply cube.

I don't know what reduction percentage xG uses, I'd guess 50%. If so, it's probably a welcome extra setting that produces results that are clearly more reliable than 0-ply, and significantly faster than full 100% 2-ply.

(xG ply notation:) If you want to use xG's "red 3-ply", I think it's most useful to use for cube decisions, in combination with a lower setting for checker play. You could do high-speed rollouts using 1-ply (or 2-ply) checker play and 3-ply red. cube decisions, pretty good, and probably better than f.i. 3-ply/3-ply truncated, and much faster than full 3-ply.

Messages In This Thread

• I propose to make a XG rollout for this position
  aoleg1 -- Wednesday, 1 July 2009, at 1:11 p.m.
  • here you are....
    Andreas -- Wednesday, 1 July 2009, at 5:15 p.m.
    • 3/3 XG RO
      Matt Cohn-Geier -- Wednesday, 1 July 2009, at 8:00 p.m.
      • 3 vs. 3 red...
        Chuck Bower -- Wednesday, 1 July 2009, at 8:23 p.m.
        • GNU 2-ply RO
          Matt Cohn-Geier -- Thursday, 2 July 2009, at 4:37 a.m.
          • GNU 2-ply RO vc XG 3/3 RO
            MaX -- Thursday, 2 July 2009, at 7:27 a.m.
            • GNU 2-ply RO vc XG 3/3 RO
              Stick -- Thursday, 2 July 2009, at 7:36 a.m.
              • GNU 2-ply RO vc XG 3/3 RO
                MaX -- Thursday, 2 July 2009, at 8:36 a.m.
              • GNU 2-ply RO vc XG 3/3 RO
                Mislav Radica -- Thursday, 2 July 2009, at 12:49 p.m.
            • GNU 2-ply RO vc XG 3/3 RO
              Robert-Jan Veldhuizen (Zorba) -- Saturday, 4 July 2009, at 5:21 p.m.
              • xG red = 33%
                Robert-Jan Veldhuizen (Zorba) -- Sunday, 5 July 2009, at 12:21 a.m.
          • GNU 2-ply RO
            Mislav Radica -- Thursday, 2 July 2009, at 12:35 p.m.
      • 3/3 XG RO
        Mislav Radica -- Thursday, 2 July 2009, at 12:41 p.m.