BGonline.org Forums

Rollout and source of position

Posted By: Timothy Chow
Date: Wednesday, 4 September 2019, at 12:44 a.m.

In Response To: Does this position look familiar? (Timothy Chow)

This is Position 4 from a back game quiz that Paul Weaver presented back in 2013.

On my personal computer, I have a PDF file containing XG 3-ply rollouts that I downloaded from somewhere a long time ago. (Unfortunately I'm not sure where I got it from or where you can download it today; maybe someone else knows?) At the time, I remember thinking, and perhaps commenting here on BGO, that I wasn't sure I was willing to trust an XG 3-ply rollout for such complicated positions.

Now that I have a faster computer, I have subjected these positions to XGR+/XGR+ rollouts. It's not clear to me that even these stronger rollout settings can be trusted, but in any case, what I found was that although in many cases the absolute equities changed noticeably (typically in favor of the side playing the backgame), the relative order of the plays in my XGR+ rollouts was basically the same as in the earlier 3-ply rollouts, with one exception. That exception was, of course, the position featured here.

In the PDF file, no play involving 12/10* even appeared in the list of rolled-out plays. As you can see from my rollout below, most of the top plays involve 12/10*. Paul's official answer to this problem was 9/7* 5/1 4/2*, which is certainly a contender in my rollout (given the variance), but it looks like 12/10* 5/1 4/2* is going to be the winner.

I believe that the reason that the 12/10* plays fared so poorly in the original rollouts is not the difference in strength between 3-ply and XGR+. Rather, it's because the 12/10* plays don't make it through XG's move filter. Paul, or whoever performed the rollouts, apparently took it for granted that plays that didn't make it through XG's move filter weren't worth considering. But this assumption seemed dubious to me; I had a hard time believing that 9/7* 5/1 4/2* could be good if 12/10* 5/1 4/2* were a terrible blunder. So using GNUBG, I ran every legal play through GNU 4-ply. This turned up a lot of plays that XG's move filter had rejected. In addition to the 10 moves below, I subjected several other plays to shorter and/or weaker rollouts. So I feel fairly confident that one of the top 3 plays below is really the top play (where by "top" play I mean the play that would top the rollout if it were extended; whether this is the true equilibrium play is another question).

		White is Player 2 score: 0 pip: 138 1 point match pip: 201 score: 0 Blue is Player 1
XGID=--aABBAa-Aa-A---bbcbcCBB--:0:0:1:22:0:0:0:1:10
Blue to play 22

1.	Rollout1	12/10* 5/1 4/2*	eq: -0.266
	Player: Opponent: 36.72% (G:4.11% B:0.28%) 63.28% (G:49.43% B:14.64%) Conf.: ± 0.013 (-0.279...-0.253) - [95.5%] Duration: 1 day 18 hours 04 minutes
2.	Rollout1	9/7* 5/1 4/2*	eq: -0.280 (-0.015)
	Player: Opponent: 35.98% (G:3.19% B:0.27%) 64.02% (G:48.51% B:13.66%) Conf.: ± 0.011 (-0.292...-0.269) - [4.5%] Duration: 1 day 15 hours 28 minutes
3.	Rollout1	12/10* 10/8 9/7* 4/2*	eq: -0.296 (-0.030)
	Player: Opponent: 35.20% (G:3.56% B:0.37%) 64.80% (G:47.53% B:12.56%) Conf.: ± 0.010 (-0.306...-0.286) - [0.0%] Duration: 1 day 12 hours 45 minutes
4.	Rollout2	12/10* 9/7* 5/3 4/2*	eq: -0.312 (-0.047)
	Player: Opponent: 34.39% (G:3.94% B:0.33%) 65.61% (G:46.89% B:11.51%) Conf.: ± 0.014 (-0.326...-0.298) - [0.0%] Duration: 16 hours 51 minutes
5.	Rollout2	12/10* 10/8 4/2* 3/1	eq: -0.324 (-0.058)
	Player: Opponent: 33.80% (G:3.16% B:0.03%) 66.20% (G:48.51% B:12.06%) Conf.: ± 0.015 (-0.339...-0.309) - [0.0%] Duration: 17 hours 12 minutes
6.	Rollout2	12/10* 10/8 5/3 4/2*	eq: -0.328 (-0.062)
	Player: Opponent: 33.60% (G:3.63% B:0.33%) 66.40% (G:47.04% B:10.32%) Conf.: ± 0.011 (-0.339...-0.317) - [0.0%] Duration: 15 hours 19 minutes
7.	Rollout2	12/10* 9/7* 5/1	eq: -0.328 (-0.063)
	Player: Opponent: 33.58% (G:3.49% B:0.28%) 66.42% (G:47.38% B:11.92%) Conf.: ± 0.014 (-0.342...-0.315) - [0.0%] Duration: 17 hours 37 minutes
8.	Rollout2	12/10* 9/7* 4/2* 3/1	eq: -0.335 (-0.069)
	Player: Opponent: 33.26% (G:3.85% B:0.39%) 66.74% (G:48.59% B:13.60%) Conf.: ± 0.015 (-0.350...-0.320) - [0.0%] Duration: 19 hours 45 minutes
9.	Rollout2	12/10* 9/7* 7/5 4/2*	eq: -0.342 (-0.076)
	Player: Opponent: 32.91% (G:3.20% B:0.39%) 67.09% (G:47.12% B:10.65%) Conf.: ± 0.011 (-0.353...-0.330) - [0.0%] Duration: 15 hours 44 minutes
10.	Rollout2	9/7* 7/3 4/2*	eq: -0.346 (-0.080)
	Player: Opponent: 32.70% (G:3.85% B:0.32%) 67.30% (G:46.95% B:9.53%) Conf.: ± 0.011 (-0.357...-0.335) - [0.0%] Duration: 14 hours 04 minutes
1 2592 Games rolled with Variance Reduction. Dice Seed: 271828 Moves and cube decisions: XG Roller+ Search interval: Large 2 1296 Games rolled with Variance Reduction. Dice Seed: 271828 Moves and cube decisions: XG Roller+ Search interval: Large

eXtreme Gammon Version: 2.19.207.pre-release, MET: Kazaross XG2

Messages In This Thread

• Does this position look familiar?
  Timothy Chow -- Monday, 2 September 2019, at 2:25 p.m.
  • Does this position look familiar?
    Terje Pedersen -- Monday, 2 September 2019, at 3:38 p.m.
  • Does this position look familiar?
    Alex21 -- Monday, 2 September 2019, at 6:00 p.m.
  • Does this position look familiar?
    Casper Van der Tak -- Tuesday, 3 September 2019, at 11:04 a.m.
  • Rollout and source of position
    Timothy Chow -- Wednesday, 4 September 2019, at 12:44 a.m.
    • Rollout and source of position
      Casper Van der Tak -- Wednesday, 4 September 2019, at 1:22 a.m.
      • Rollout and source of position
        Neil Robins -- Wednesday, 4 September 2019, at 7:50 p.m.
        • Rollout and source of position
          Casper Van der Tak -- Thursday, 5 September 2019, at 9:55 a.m.