the big thread about cube decisions - Page 7 - Backgammon Forum - Discuss Backgammon Strategy, Clubs and Books

I can only repeat what I wrote in my post about the checker play: “Should White double after this play? Maybe yes, maybe no, but I would take if he does.”

I can’t say anything more, because I put the question to a bot after submitting that post.

Mike

Quote

04-19-2015 , 01:04 PM

#152

Robertie

old hand

Join Date: May 2004 Posts: 1,954

Black trails by 16 in the race and has the weaker home board. Rolls containing a 6, 5, or 2 don't do much for Black. Black's men on his 21-point are somewhat blocked.

Doubling looks very premature and it must be a trivial take.

Quote

04-19-2015 , 02:10 PM

#153

Alpha Fish

Pooh-Bah

Join Date: Aug 2008 Posts: 3,838

interesting position, move 10 checker to 9 and I'm doubling

Quote

04-20-2015 , 02:01 AM

#154

bleep69

grinder

Join Date: Sep 2008 Posts: 483

I reckon it's a double/take. Black has lots of ammo bearing down on the lonely white fellow in his board, as well as the threat of hitting the guy on 18. However, on white's side, 3's are duplicated for black, he is close to semi-trapping the 2 black checkers, and he's ahead in the race.

Quote

04-20-2015 , 03:10 AM

#155

Kamba

enthusiast

Join Date: Feb 2015 Posts: 93

I wouldn't double here. I don't think I have that many great dice and I'm behind in the race. I get the feeling that if I don't throw some nice dice right now, I could lose this one.

No double / Take

Quote

04-21-2015 , 04:28 AM

#156

yogiman

adept

Join Date: Jun 2012 Posts: 784

Code:

Cube analysis
Rollout cubeless equity  +0,333

Cubeful equities:
1. No double            +0,426
2. Double, pass         +1,000  ( +0,574)
3. Double, take         +0,373  ( -0,053)
Proper cube action: No double, take (8,4%)

This is the position that arises after the right move in my problems thread. I think you should certainly double because lots of players would pass this one.

Quote

04-21-2015 , 05:28 AM

#157

Kamba

enthusiast

Join Date: Feb 2015 Posts: 93

Oops... i was thinking I could sneak in a beaver as white :-)

Are you saying that this is a good position to "play the player"?

Quote

04-21-2015 , 08:45 AM

#158

yogiman

adept

Join Date: Jun 2012 Posts: 784

I think that is true for all those risky complicated positions, and that's one of the reasons I would say bye bye to pr.

Quote

04-24-2015 , 03:17 AM

#159

yogiman

adept

Join Date: Jun 2012 Posts: 784

Position ID: O88GAChsNwTgDA Match ID: UgkAAAAAAAAE

White - Pips 97

Black - Pips 155
Black on roll. Cube action?
Created with www.BGdiagram.com

Quote

04-24-2015 , 02:15 PM

#160

_Z_

adept

Join Date: Jul 2004 Posts: 722

I think this is a pass, and maybe a big one. One problem for White is even if gets both guys home, he could still lose if Black still has his decent defensive position. And it's going to be very tough to get both guys home in any case since Black will have great outfield control. Another problem for White is he is very brittle -- if he doesn't roll a 6 or 5 on the next roll, his home board is often going to crash badly.

Wouldn't shock me if this was too good, but you have to double in practice even if you think it's a little too good.

Double/Pass.

Quote

04-25-2015 , 03:12 PM

#161

Kamba

enthusiast

Join Date: Feb 2015 Posts: 93

I'm doubling here hoping for a quick drop. I don't know if its too good but I know that I am not good at playing these positions if white escapes so I want to end it now.

So, double/drop.

Quote

04-26-2015 , 01:45 AM

#162

Taper_Mike

adept

Join Date: Sep 2009 Posts: 1,030

Grunch. This is my thinking before reading the posts of others.

The big picture is that Black’s position should be improving on each turn, while White’s can only weaken. White must hope to escape and run. If he his hit, he must hope to do it again. At any time, White can roll badly, and be forced to break a point in his front position. Right now, White’s list of cracking rolls includes the small doublets: 11, 22, 33, and 44. In addition, any 4 that does not launch a runner into the outfield must be played from White’s 6pt.

The only bit of good new for White is that Black is not ready to pounce on this turn. Black has only one checker targeting his inner board. Black’s checker on the 10pt is somewhat out of place. Black does not have a firm hold on the outfield. White still leads in the race by almost 60 pips.

Is that enough to ward off a cube? With two checkers already hit, and White’s front position threatening to crumble, Black definitely has market losers (if he has not already lost his market). Is this too good to double? I suppose that is possible, but with so much work still do, I would cube.

Can White take? Here am less certain. My immediate reaction upon seeing the position was to take, but having now reviewed it, I have talked myself into a pass.

Mike

Quote

04-26-2015 , 07:16 AM

#163

yogiman

adept

Join Date: Jun 2012 Posts: 784

Code:

Cube analysis
Rollout cubeless equity  +0,591

Cubeful equities:
1. Double, pass         +1,000
2. Double, take         +1,026  ( +0,026)
3. No double            +0,844  ( -0,156)
Proper cube action: Redouble, pass

This is a narrow pass, but whatever you realistically do with the three free black checkers, it will remain a pass. Ofcourse, if you move 24w to 23 to make the 2-point, it will be a take.

However, changing black's position into a 34- instead of a 35-backgame will lead to a convincing take, in this case also whatever you do with the three free black checkers:

Code:

Cube analysis
Rollout cubeless equity  +0,464

Cubeful equities:
1. Double, take         +0,748
2. Double, pass         +1,000  ( +0,252)
3. No double            +0,745  ( -0,004)
Proper cube action: Redouble, take

Give life to those three dead checkers of white, and it is no double anymore:

Code:

Cube analysis
Rollout cubeless equity  +0,404

Cubeful equities:
1. No double            +0,704
2. Double, pass         +1,000  ( +0,296)
3. Double, take         +0,605  ( -0,099)
Proper cube action: No redouble, take (25,0%)

Code:

Cube analysis
Rollout cubeless equity  +0,333

Cubeful equities:
1. No double            +0,606
2. Double, pass         +1,000  ( +0,394)
3. Double, take         +0,404  ( -0,203)
Proper cube action: No redouble, take (34,0%)

Quote

04-26-2015 , 11:20 AM

#164

Taper_Mike

adept

Join Date: Sep 2009 Posts: 1,030

Once again, you have given us some nice variants. One key feature seems to be whether White can play a 4 inside without cracking. I took your second variant and moved one checker, and then two, off the 20pt. The first diagram below is a take, and the second is a pass. When White has even one spare on the 20pt, he can take.

White - Pips 97

Black - Pips 158
Black on roll. Cube action is Double/Take.

Position ID: d9wGAChsNwTADQ Match ID: UgkAAAAAAAAE

White - Pips 94

Black - Pips 158
Black on roll. Cube action is Double/Pass.

Position ID: 99gGAChsNwTADQ Match ID: UgkAAAAAAAAE

Mike

Quote

04-26-2015 , 11:44 AM

#165

_Z_

adept

Join Date: Jul 2004 Posts: 722

Due to Gnu's consistently low evaluations of containment positions, truncated rollouts of these positions will not give an accurate result.

Here's an XG rollout of position 17. It's a solid pass by .15. A full Gnu 2-ply rollout will yield similar results.

Code:

No redouble
  Player Winning Chances:   73.97% (G:25.53% B:1.05%)
  Opponent Winning Chances: 26.03% (G:10.87% B:0.15%)
Redouble/Take
  Player Winning Chances:   74.72% (G:26.35% B:1.10%)
  Opponent Winning Chances: 25.28% (G:10.70% B:0.14%)

Cubeless Equities: No Double=+0.635, Double=+1.321

Cubeful Equities:
       No redouble:     +0.883 (-0.117)
       Redouble/Take:   +1.149 (+0.149)
       Redouble/Pass:   +1.000

Best Cube action: Redouble / Pass

Rollout:
  1362 Games rolled with Variance Reduction.
  Moves: 3-ply, cube decisions: XG Roller
  Confidence No Double: ± 0.011 (+0.872..+0.894)
  Confidence Double:    ± 0.019 (+1.130..+1.168)

Last edited by _Z_; 04-26-2015 at 11:52 AM. Reason: Playing with formatting

Quote

04-26-2015 , 12:24 PM

#166

Taper_Mike

adept

Join Date: Sep 2009 Posts: 1,030

Quote:

Originally Posted by _Z_

Due to Gnu’s [and XG’s] consistently low evaluations of containment positions, truncated rollouts of these positions will not give an accurate result.

Evidently, the same is true of the variants I posted. A truncated rollout showed the first of those to be a take, but when I performed a full rollout, the result was pass.

White - Pips 97

Black - Pips 158
Black on roll. Cube action is Double/Pass.

Code:

XGID=-aaBBCB---A-------bbcCBcc-:2:1:1:00:0:0:0:0:10

X:Player 1   O:Player 2
Score is X:0 O:0. Unlimited Game
Pip count  X: 158  O: 97 X-O: 0-0
Cube: 4, X own cube
X on roll, cube action

Analyzed in Rollout
No redouble
  Player Winning Chances:   72.31% (G:26.08% B:1.21%)
  Opponent Winning Chances: 27.69% (G:12.61% B:0.27%)
Redouble/Take
  Player Winning Chances:   73.10% (G:26.52% B:1.28%)
  Opponent Winning Chances: 26.90% (G:12.54% B:0.24%)

Cubeless Equities: No Double=+0.590, Double=+1.225

Cubeful Equities:
       No redouble:     +0.863 (-0.137)
       Redouble/Take:   +1.059 (+0.059)
       Redouble/Pass:   +1.000

Best Cube action: Redouble / Pass

Rollout:
  1296 Games rolled with Variance Reduction.
  Dice Seed: 50198947
  Moves: 3-ply, cube decisions: XG Roller
  Confidence No Double: ± 0.011 (+0.852..+0.875)
  Confidence Double:    ± 0.022 (+1.037..+1.081)

  Double Decision confidence: 100.0%
  Take Decision confidence: 100.0%

Rollout by Taper_Mike
2015-Apr-26
eXtreme Gammon Version: 2.10.199.2658
Cube Decisions 017a.xgp

Mike

Quote

04-26-2015 , 12:54 PM

#167

yogiman

adept

Join Date: Jun 2012 Posts: 784

We can't afford to be too punctilious about cube decisions. We serve our memory better with sacrificing a little equity imo.

Quote

04-29-2015 , 03:33 AM

#168

yogiman

adept

Join Date: Jun 2012 Posts: 784

Quote:

Originally Posted by Taper_Mike

Bill Robertie and the many other users of Snowie back in the day when computers were a whole lot slower know the value of a truncated rollout. When the number of trials is small, a trunc is usually better than a full rollout.

The trunc used by Yogiman is very good. It uses GnuBg's "world class" 2-ply evals to a depth of 10 plies. The only thing I would change is the number of trials. I would like to see a minimum of 324, and twice that many for "difficult" positions. Note that 324 = 1296/4, while 648 = 1296/2. A good compromise might be to use the same number of trials as XG does with its XGR++ trunc. That number is 360.

Just my two cents.

Mike

I go in Gnu to settings>rollouts, and get two windows. The number of trials is in the window of general settings, whereas the tickling box for "cube decisions same as checker play" is in the window of play settings. How do I get a different number of trials for both without manually changing the number every time?

Quote

04-29-2015 , 05:10 AM

#169

Taper_Mike

adept

Join Date: Sep 2009 Posts: 1,030

Quote:

Originally Posted by yogiman

One way is to set up a rollout, and then save it using "Save As" at the bottom of the same window. You can use any of five names: a, b, c, d, or e. After saving one, then you can change the setup, and save the alternate using a different name. When you save a rollout, the number of trials is one of the parameters that is stored.

When you want to make a rollout using one of your saved rollouts, do not click "Rollout" in the Move Panel. Instead, click one of the letters: a, b, c, d, or e.

Incidentally, the checkbox for "Cube decisions use same settings as Checker play" refers to the strength of analysis, not the number of trials. Suppose, for instance, you are making a 324-trial rollout. On each turn in your rollout, there may be both a cube decision and a checker play. Because cube decisions are generally faster to compute than checker plays, you might want to use a stronger setting for them. You could, for instance, use 2-ply for checker plays and 3-ply for cube decisions. No matter what you choose, however, the number of games played, i.e., the number of trials, will be 324.

Mike

Quote

04-30-2015 , 06:15 AM

#170

yogiman

adept

Join Date: Jun 2012 Posts: 784

cube settings from now on:

Truncated cubeful rollout (depth 10) with var.redn.
1296 games, Mersenne Twister dice gen. with seed 739617091 and quasi-random dice
Stop when std.errs. are small enough: ratio 0,1 (min. 1296 games)
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]

Quote

04-30-2015 , 06:19 AM

#171

yogiman

adept

Join Date: Jun 2012 Posts: 784

Position ID: uL0jAUCbmTkAFA Match ID: cAkAAAAAAAAE

White - Pips 114

Black - Pips 117
Black on roll. Cube action?
Created with www.BGdiagram.com

Quote

04-30-2015 , 08:33 AM

#172

bleep69

grinder

Join Date: Sep 2008 Posts: 483

I'll go with double/take.

Black has a few rolls that crush - 53, 52, 54, 61, 63, 62 and 64, without looking at any doublets.

At the same time, white is a tiny bit ahead, and if black fails to roll the roll the right numbers, white may be able to slip through, romp to victory and gloat for years to come.

Last edited by bleep69; 04-30-2015 at 08:46 AM. Reason: missed some rolls

Quote

05-01-2015 , 03:38 AM

#173

Taper_Mike

adept

Join Date: Sep 2009 Posts: 1,030

One of the big takeaways from Robertie’s 501 is how perilous it is to dance on a four-point board. Many passes follow.

Here, Black can hit an second checker with twelve rolls: 64, 63, 62, 61, 54, and 52. He can make his 5pt with four more: 55, 53, and 33. The first of these, 55, is a super-joker that also makes the 3pt. 66 and 65 play quite nicely, escaping both back checkers.

In case none of these good things happen, White has 45% chance to dance, giving Black the same opportunites on the next turn.

If Black can hit a second checker, he will often get the third. Because the Jacoby Rule is not in force, many gammons will follow.

Finally, if the game concludes with no more hitting, Black is still about 50/50 in the race!

I am often to conservative in my takes—I pass too often—but I see enough here that I don’t like that I will risk that mistake.

Double/pass.

Mike

P.S. @Yogiman: Your upgrade to 1296 trials will mean that the confidence intervals in your rollouts are significantly reduced. Not only will you be picking out the top play more often, but the margins between it and the other choices will also be more accurate.

Quote

05-01-2015 , 09:14 AM

#174

Robertie

old hand

Join Date: May 2004 Posts: 1,954

I'll go with double/take.

The double is clear with White on the bar and two more blots floating around. Black hits immediately with 52, 54, 61, 62, 63, and 64. 55, 33, and 66 are also pretty good, and the cube is in the middle, so gammons need to be activated.

The take also looks clear to me. White has a 4-prime with good builders and no checkers out of play. Black doesn't have many numbers to make the fifth point in his board. The sequence where Black doesn't throw a 5 or 6 and White enters happens almost 20% of the time, and White is in good shape there owning the cube.

When in doubt, take.

Quote

05-01-2015 , 02:51 PM

#175

Karol Szczerek

enthusiast

Join Date: Mar 2015 Posts: 92

Opp must have a clear take, since he has good&flexible blockade and a lot of counterplay from the bar with our board not being decisive yet.

The double is rather clear. I'd estimate at least 60% wins, with ca. half of them being gammons. This is enough to double aggressively if the volatility of next exchange of rolls is high.

This is the kind of position, which is rather easy at match scores. If I lead, this is a clear no-double, and if I trail or have some sort of gammon-promoting score (i.e. 1-0/5), I double automatically and it's a big drop most of the time.

Quote

Page 7 of 13

First

2 3 4 5 6 7 8 9 10 11 12

Last

Post Reply Subscribe

...

Page 7 of 13

First

2 3 4 5 6 7 8 9 10 11 12

Last