BGonline.org Forums

Largest bot misevaluation

Posted By: Timothy Chow
Date: Monday, 21 December 2009, at 3:28 p.m.

In Response To: Largest bot misevaluation (Nack Ballard)

I haven't been playing very long by the standards of people here so I'm sure I won't be setting any records here, but by some measure, I think Position 101 below is the biggest checker-play misevaluation out of the positions in Robertie's 501 problems. If you ask GNU for a 2-ply evaluation with a normal move filter, then GNU doesn't bother going beyond 0-ply because it thinks 13/8 10/4 is very clearly (0.216) better than the alternative 13/8 13/7. However, a rollout shows that GNU 0-ply is misevaluating 13/8 13/7 by about 0.210. Even if you force it to do a 2-ply evaluation, GNU is off by about 0.113 and thinks that 13/8 10/4 is the best play by a clear margin.

Maybe this doesn't count because in some sense it's a 0-ply evaluation error, but I sort of think it should count because it's what you would get if you asked for a 2-ply evaluation under most circumstances.

By the way, Robertie errs here too; he does not even consider 13/8 13/7 in his analysis.

gnubg	127
tchow	154
Position ID: tm3AATDYTuIBMA Match ID: cIkaAAAAAAAA

• tchow moves 13/8 13/7

Alert: very bad move ( -0.216)

#		Ply	Move	Equity
	1	0	13/8 10/4	-0.340
			0.410 0.127 0.006 - 0.590 0.206 0.014
•	2	0	13/8 13/7	-0.556 ( -0.216)
			0.393 0.119 0.004 - 0.607 0.287 0.033

If we force 2-ply evaluation we get:

gnubg	127
tchow	154
Position ID: tm3AATDYTuIBMA Match ID: cIkaAAAAAAAA

• tchow moves 13/8 13/7

Alert: doubtful move ( -0.051)

#		Ply	Move	Equity
	1	2	13/8 10/4	-0.408
			0.404 0.132 0.005 - 0.596 0.216 0.017
•	2	2	13/8 13/7	-0.459 ( -0.051)
			0.414 0.116 0.003 - 0.586 0.284 0.036

Messages In This Thread

• $ Cube Action?
  Michael Depreli -- Sunday, 20 December 2009, at 11:13 a.m.
  • $ Cube Action?
    Leonardo Jerkovic -- Sunday, 20 December 2009, at 2:02 p.m.
    • $ Cube Action?
      Nack Ballard -- Sunday, 20 December 2009, at 5:40 p.m.
      • $ Cube Action?
        Nack Ballard -- Monday, 21 December 2009, at 1:15 a.m.
        • $ Cube Action?
          Leonardo Jerkovic -- Monday, 21 December 2009, at 1:48 a.m.
        • $ Cube Action?
          Petter Bengtsson -- Monday, 21 December 2009, at 1:48 a.m.
          • $ Cube Action?
            Nack Ballard -- Monday, 21 December 2009, at 2:53 a.m.
            • $ Cube Action?
              Petter Bengtsson -- Monday, 21 December 2009, at 3:04 a.m.
              • Largest bot misevaluation
                Nack Ballard -- Monday, 21 December 2009, at 4:32 a.m.
                • 54 play...and its fun to have a fast computer now
                  neilkaz -- Monday, 21 December 2009, at 4:40 a.m.
                  • 54 play...and its fun to have a fast computer now
                    Nack Ballard -- Monday, 21 December 2009, at 4:45 a.m.
                • Would you count something like this?
                  Matt Cohn-Geier -- Monday, 21 December 2009, at 4:53 a.m.
                  • Two categories: (a) checker play, (b) cube
                    Nack Ballard -- Monday, 21 December 2009, at 5:21 a.m.
                • Largest bot misevaluation Bot Data
                  Michael Depreli -- Monday, 21 December 2009, at 7:55 a.m.
                • Largest bot misevaluation
                  Timothy Chow -- Monday, 21 December 2009, at 3:28 p.m.
                  • Largest bot misevaluation
                    Bob Koca -- Wednesday, 23 December 2009, at 4:30 a.m.
    • $ Cube Action?
      neilkaz -- Sunday, 20 December 2009, at 6:39 p.m.
      • $ Cube Action?
        Leonardo Jerkovic -- Sunday, 20 December 2009, at 6:54 p.m.