BGonline.org Forums

Cubeless games in cubeful rollouts

Posted By: Daniel Murphy
Date: Saturday, 3 September 2011, at 7:09 p.m.

In Response To: Cubeless games in cubeful rollouts (Timothy Chow)

So I'm not sure exactly what Daniel is getting at.

Tim, what I'm getting is no more than exploring your proposal to see which statistical display is actually more useful in a variety of positions, starting with a very simple position with no gammons (and no backgammons) and no takable redoubles.

Recall, of course, that in my proposal, you can eat your cake and have it too.

Yes, in the sense that your proposal was to be able to see both "traditional statistics" and "actual statistics," but not at the same time. Your preferred default view would be "actual statistics" because, you have written, live cube rollout "cubeless" numbers are often misleading and/or wrong and "the current practice of trying to read pseudocubeless numbers like tea leaves seems ridiculously primitive."

That said, about this position you wrote "This is not to say that I might not also want to see a cubeless rollout of the same position. If so, then I would do a cubeless rollout and want to see cubeless statistics." So I'm not sure whether, after further thought, your preference is to have one rollout that report both cubeful statistics and possibly misleading and/or wrong "pseudocubeless" statistics, or if you'd rather have a bot do two separate rollouts, one cubeful and one cubeless.

What do you think?

Let me also acknowledge that while I wrote "It's been argued that cubeless outcomes and cubeless equities are irrelevant, even worthless," in fact you withdrew the charge of "worthless" in favor of "ridiculously primitive" (and still irrelevant and misleading).

Rereading your proposal, I see that you actually proposed seeing statistics for single wins, double wins (these can be gammons at the single win cube level or wins after a further redouble) and "more than double" wins, which are wins that win more points than a double win ("Instead of "backgammons," the third number will report the percentage of higher-value games."). So I reproduce the cubeful win statistics below with different headings in place of single, double and triple:

Cube decision
Cubeful equities:
1.	Double, take	+0.3818
2.	Double, pass	+1.0000	+0.6182
3.	No double	+0.3256	-0.0562
Proper cube action:		Double, take
Rollout details
Win Single W Double W more than Double Lose Single Lose double Lose more than double Cubeful Centered 1-cube 0.6628 0.0000 0.0000 - 0.3372 0.0000 0.0000 +0.3256 Standard error 0.0000 0.0000 0.0000 - 0.0000 0.0000 0.0000 0.0000 Player White owns 2-cube 0.4722 0.0000 0.0000 - 0.5277 0.0000 0.0000 +0.3818 Standard error 0.0000 0.0000 0.0000 - 0.0000 0.0000 0.0000 0.0000
Full cubeful rollout with var.redn.
7776 games, Mersenne Twister dice gen. with seed 927765845 and quasi-random dice
Play: 4-ply cubeful prune
keep the first 0 0-ply moves and up to 12 more moves within equity 0.2
Skip pruning for 1-ply moves.
keep the first 0 2-ply moves and up to 4 more moves within equity 0.1
Skip pruning for 3-ply moves.
Cube: 4-ply cubeful prune

---

http://www.bgonline.org/forums/webbbs_config.pl?read=100568

I propose that bot rollouts allow the user to choose between two options: "Display traditional statistics" and "Display actual statistics."

http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=58574

I'm not suggesting getting rid of any features, just adding some. I am also arguing that getting used to a new mindset in which we train ourselves to think ATS directly, rather than using cubelessness as a stepping stone, is a better way to go as far as reference positions go.

http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=57713

"In my opinion, this figure is misleadingly named (because it is derived from statistics from a live-cube rollout) and it is also useless, because what we really care about is the value of the game taking the cube into account."

http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=94019

"I called the pseudocubeless numbers "ridiculously primitive" for this reason. Not useless, just ridiculously primitive."

Messages In This Thread

• Cubeless games in cubeful rollouts
  Daniel Murphy -- Friday, 2 September 2011, at 3:16 p.m.
  • Cubeless games in cubeful rollouts
    Timothy Chow -- Friday, 2 September 2011, at 4:56 p.m.
  • Cubeless games in cubeful rollouts
    Neil Robins -- Saturday, 3 September 2011, at 4:49 a.m.
    • Cubeless games in cubeful rollouts
      Timothy Chow -- Saturday, 3 September 2011, at 5:28 p.m.
      • Cubeless games in cubeful rollouts
        Daniel Murphy -- Saturday, 3 September 2011, at 7:09 p.m.
        • Cubeless games in cubeful rollouts
          Timothy Chow -- Sunday, 4 September 2011, at 6:37 p.m.
          • Cubeless games in cubeful rollouts
            Neil Robins -- Monday, 5 September 2011, at 7:53 a.m.
            • Cubeless games in cubeful rollouts
              Timothy Chow -- Tuesday, 6 September 2011, at 12:41 a.m.
  • Cubeless games in cubeful rollouts #2
    Daniel Murphy -- Sunday, 4 September 2011, at 1:17 a.m.