BGonline.org Forums

Small Cube Error Reported by XGR++. Why?

Posted By: Taper_Mike
Date: Saturday, 8 February 2014, at 11:20 p.m.

In Response To: Small Cube Error Reported by XGR++. Why? (Henrik Bukkjaer)

If you check the description found in the XG Help System, you will see that XGR++ uses a 5-ply trunc for checker plays and a 7-ply trunc for cube decisions. In GnuBg, I saved the settings for a pair truncated rollouts that use the same metrics.

There are only two differences between the settings used by my GnuBg rollouts and those of XGR++.

1. XGR++ uses the same seed every time.
2. XGR++ stops cube rollouts at fewer than 360 trials when reaching 0.010 Confidence (minimum 180 games).

Below are two GnuBg truncated rollouts of Henrik’s original position. Unlike the GnuBg evals posted by Neil Robbins, these truncs do not show the double as optional. Although perhaps not surprising, it is interesting that the two rollouts do not agree. One says no-double/take, and the other, double/take. Were XGR++ to use a new seed for each invocation, the same result would probably be possible in eXtreme Gammon.

Mike

---

No-double/Take

The score (after 0 games) is: gnubg 16, Taper_Mike 8 (match to 17 points, post-Crawford play)

Move number 2: Taper_Mike on roll, cube decision?

gnubg	152
Taper_Mike	156
Position ID: mOfgASHgc/ADIA Match ID: cAkgAgABQAAE

• Taper_Mike doubles

Alert: wrong double ( -0.000039)!

Cube decision
Rollout cubeless equity		+0.041890	(Money: 0)
Cubeful equities:
1.	No double	+0.155615
2.	Double, pass	+1.000000	+0.844385
3.	Double, take	+0.155576	-0.000039
Proper cube action:		No double, take (0.0%)
Rollout details
Win W g W bg   Lose L g L bg Cubeless Cubeful Centered 1-cube 0.476106 0.099209 0.002999 - 0.523894 0.151595 0.006708 +0.041890 +0.155615 Standard error 0.000895 0.000914 0.000068 - 0.000895 0.001713 0.000231 0.005132 0.001825 Player gnubg owns 2-cube 0.475991 0.099321 0.003019 - 0.524009 0.152423 0.006694 +0.155576 +0.155576 Standard error 0.000891 0.000823 0.000067 - 0.000891 0.001719 0.000233 0.001795 0.001795
Truncated cubeful rollout (depth 7) with var.redn.
360 games, Mersenne Twister dice gen. with seed 735515093 and quasi-random dice
Play: world class 2-ply cubeful prune [world class]
keep the first 0 0-ply moves and up to 8 more moves within equity 0.16
Skip pruning for 1-ply moves.
Cube: 2-ply cubeful prune [world class]
Different evaluations after 2 plies:
Play: 1-ply cubeful prune
keep the first 0 0-ply moves and up to 8 more moves within equity 0.16
Cube: 2-ply cubeful prune [world class]

 

---

Double/Take

The score (after 0 games) is: gnubg 16, Taper_Mike 8 (match to 17 points, post-Crawford play)

Move number 2: Taper_Mike on roll, cube decision?

gnubg	152
Taper_Mike	156
Position ID: mOfgASHgc/ADIA Match ID: cAkgAgABQAAE

• Taper_Mike doubles

Cube decision
Rollout cubeless equity		+0.051199	(Money: 9)
Cubeful equities:
1.	Double, take	+0.157984
2.	Double, pass	+1.000000	+0.842016
3.	No double	+0.157452	-0.000532
Proper cube action:		Double, take
Rollout details
Win W g W bg   Lose L g L bg Cubeless Cubeful Centered 1-cube 0.475280 0.101440 0.003216 - 0.524720 0.152383 0.006898 +0.051199 +0.157452 Standard error 0.000934 0.000950 0.000069 - 0.000934 0.001576 0.000234 0.005734 0.001861 Player gnubg owns 2-cube 0.475372 0.101671 0.003225 - 0.524628 0.152367 0.006903 +0.157984 +0.157984 Standard error 0.000931 0.000864 0.000067 - 0.000931 0.001556 0.000229 0.001862 0.001862
Truncated cubeful rollout (depth 7) with var.redn.
360 games, Mersenne Twister dice gen. with seed 735756070 and quasi-random dice
Play: world class 2-ply cubeful prune [world class]
keep the first 0 0-ply moves and up to 8 more moves within equity 0.16
Skip pruning for 1-ply moves.
Cube: 2-ply cubeful prune [world class]
Different evaluations after 2 plies:
Play: 1-ply cubeful prune
keep the first 0 0-ply moves and up to 8 more moves within equity 0.16
Cube: 2-ply cubeful prune [world class]

 

Messages In This Thread

• Small Cube Error Reported by XGR++. Why?
  Henrik Bukkjaer -- Friday, 7 February 2014, at 12:23 p.m.
  • Small Cube Error Reported by XGR++. Why?
    Taper_Mike -- Friday, 7 February 2014, at 1:02 p.m.
  • Small Cube Error Reported by XGR++. Why?
    Neil Robins -- Friday, 7 February 2014, at 4:05 p.m.
    • Small Cube Error Reported by XGR++. Why?
      Michael Nielsen -- Friday, 7 February 2014, at 5:46 p.m.
      • Small Cube Error Reported by XGR++. Why?
        Henrik Bukkjaer -- Saturday, 8 February 2014, at 12:13 a.m.
    • Small Cube Error Reported by XGR++. Why?
      Henrik Bukkjaer -- Saturday, 8 February 2014, at 12:02 a.m.
      • Small Cube Error Reported by XGR++. Why?
        Taper_Mike -- Saturday, 8 February 2014, at 11:20 p.m.
  • Small Cube Error Reported by XGR++. Why?
    Bob Koca -- Friday, 7 February 2014, at 7:46 p.m.
    • Small Cube Error Reported by XGR++. Why?
      Maik Stiebler -- Saturday, 8 February 2014, at 9:26 a.m.
      • Small Cube Error Reported by XGR++. Why?
        Henrik Bukkjaer -- Monday, 10 February 2014, at 2:20 p.m.
      • Small Cube Error Reported by XGR++. Why?
        Maik Stiebler -- Tuesday, 4 March 2014, at 6:59 p.m.
  • Small Cube Error Reported by XGR++. Why?
    Neil Robins -- Wednesday, 12 February 2014, at 6:18 p.m.
    • Small Cube Error Reported by XGR++. Why?
      Henrik Bukkjaer -- Wednesday, 12 February 2014, at 6:24 p.m.
    • Small Cube Error Reported by XGR++. Why?
      Matt Cohn-Geier -- Wednesday, 12 February 2014, at 7:08 p.m.
      • Small Cube Error Reported by XGR++. Why?
        Bob Koca -- Friday, 14 February 2014, at 12:55 a.m.
        • Small Cube Error Reported by XGR++. Why?
          Matt Cohn-Geier -- Friday, 14 February 2014, at 1:16 a.m.
          • Small Cube Error Reported by XGR++. Why?
            Bob Koca -- Saturday, 15 February 2014, at 7:59 p.m.