BGonline.org Forums

Denver -- D/T .. 6-5 .. 6-2

Posted By: Daniel Murphy
Date: Monday, 29 September 2008, at 9:12 p.m.

In Response To: Denver match score rollout (neilkaz)

Title: A Rollout. Subtitle: Man over machine. Or, Why We Do Rollouts. And Why We Don't Believe Them Blindly. Here's the position Neil posted in this thread:

The score is: Stepan 9, Neil 6 (match to 11 points)

Neil (Blue) on roll, cube decision?

Stepan	151
Neil	153
Position ID: Zk7wATDGNvABMA Match ID: cAlgAZAAMAAA

Neil doubled, Stepan took. Neil rolled 6-5 and played 13/7 13/8. Stepan rolled 6-2:

The score is: Stepan 9, Neil 6 (match to 11 points)

Stepan to play 62

Stepan	151
Neil	142
Position ID: xnbBATBmTvABMA Match ID: AQFrAZAAMAAA

Stepan played 13/5. And that play is best in a 2-ply rollout:

#		Ply	Move	Equity
•	1	R	13/5	-0.5474
			0.3901 0.1699 0.0167 - 0.6099 0.2145 0.0167 -0.5474 -0.5474 0.0054 0.0123 0.0041 - 0.0054 0.0066 0.0018 0.0154 0.0154
			Full cubeful rollout with var.redn.
			220 games, Mersenne Twister dice gen. with seed 835411290 and quasi-random dice
			Play: supremo 2-ply cubeful prune [world class]
			keep the first 0 0-ply moves and up to 16 more moves within equity 0.32
			Skip pruning for 1-ply moves.
			Cube: 2-ply cubeful prune [world class]
	2	R	13/7 8/6	-0.5961 ( -0.0487)
			0.3678 0.1390 0.0159 - 0.6322 0.2213 0.0131 -0.5961 -0.5961 0.0046 0.0121 0.0064 - 0.0046 0.0051 0.0025 0.0116 0.0116
			Full cubeful rollout with var.redn.
			220 games, Mersenne Twister dice gen. with seed 835411290 and quasi-random dice
			Play: supremo 2-ply cubeful prune [world class]
			keep the first 0 0-ply moves and up to 16 more moves within equity 0.32
			Skip pruning for 1-ply moves.
			Cube: 2-ply cubeful prune [world class]
	3	R	24/16	-0.6118 ( -0.0643)
			0.3992 0.1648 0.0372 - 0.6008 0.2794 0.0112 -0.6118 -0.6118 0.0048 0.0122 0.0101 - 0.0048 0.0050 0.0015 0.0124 0.0124
			Full cubeful rollout with var.redn.
			220 games, Mersenne Twister dice gen. with seed 835411290 and quasi-random dice
			Play: supremo 2-ply cubeful prune [world class]
			keep the first 0 0-ply moves and up to 16 more moves within equity 0.32
			Skip pruning for 1-ply moves.
			Cube: 2-ply cubeful prune [world class]

But Gnubg would not have made that play. Not in this rollout, not in a rollout of Neil's 6-5 play, and not in a rollout of Neil's cube. Gnubg 0-ply, 2-ply and 4-ply would have played 24/16. Gnubg 1-ply and 3-ply would have played 13/7 8/6.

Please don't think that the point is that Neil's cube rollout is wrong. No -- it's just a reminder that when using bot evaluations and rollouts we must remember that they aren't perfect.

Messages In This Thread

• Denver match score cube action
  neilkaz -- Sunday, 28 September 2008, at 1:47 a.m.
  • Denver match score cube action
    garyo -- Sunday, 28 September 2008, at 2:23 a.m.
    • Denver match score cube action
      likesbg -- Sunday, 28 September 2008, at 5:01 a.m.
  • Denver match score cube action
    Matt Cohn-Geier -- Sunday, 28 September 2008, at 6:48 a.m.
  • Denver match score cube action
    Neil Robins -- Sunday, 28 September 2008, at 8:46 a.m.
  • Denver match score rollout
    neilkaz -- Monday, 29 September 2008, at 4:19 p.m.
    • Denver match score rollout
      Stick -- Monday, 29 September 2008, at 5:41 p.m.
      • Denver match score rollout
        Coolrey -- Wednesday, 1 October 2008, at 8:08 p.m.
        • Denver match score rollout
          Matt Cohn-Geier -- Wednesday, 1 October 2008, at 8:13 p.m.
          • Denver match score rollout
            Stick -- Wednesday, 1 October 2008, at 8:33 p.m.
            • Denver match score rollout
              Matt Cohn-Geier -- Wednesday, 1 October 2008, at 8:40 p.m.
    • Denver -- D/T .. 6-5 .. 6-2
      Daniel Murphy -- Monday, 29 September 2008, at 9:12 p.m.
      • Denver -- D/T .. 6-5 .. 6-2
        Stick -- Monday, 29 September 2008, at 9:20 p.m.