obvious is not always right - Page 7 - Backgammon Forum - Discuss Backgammon Strategy, Clubs and Books

Three of these are easy enough. Start by putting a cork in the bottle with 12/7. Next, play 23/18, and then realize that continuing all the way to the 8pt is the wrong idea. That would pay off to White’s joker 61, and even if White did not hit, would give Black a bad 44 and an awkward 33 on his next turn. You don’t want to slot the back of the prime until you have a cover within range. So play the third of these 5s by moving 23/18 again.

The hardest part is the last 5, and it is not very hard at all. There are only two choices: 23/18 and 18/13. Usually, the best way to contain a trapped checker is to position builders within direct range of the back end of the prime. That serves two purposes. If White cannot move up to the front edge of the prime, then you can slot the back. But in case White moves up or jumps out, it gives you outfield control.

In this position, however, playing 18/13 would mean leaving two blots behind. White might hit once or twice, or even point on your head. If that were to happen, you might find yourself on the roof against a four-point board. I think you would still be a favorite, but that could change fast if White were to roll an ace.

To guard against that, I would quietly play the last 5 out to the 18pt. It has the hidden advantage that some of White’s 4s would leave shots.

23/18(3), 12/7

Mike

Quote

01-29-2015 , 01:29 AM

#153

bleep69

grinder

Join Date: Sep 2008 Posts: 483

It feels good to get one right every now and again. My main reasoning for abandoning 23/8 was the long delay in bringing up anything to cover with.

Quote

01-29-2015 , 06:16 AM

#154

yogiman

adept

Join Date: Jun 2012 Posts: 784

Taper_Mike has informed me by pm that -z- is right. It was not noticed by the Gnu tutor because in 2-ply it is a slight double and a take. We would all have dropped in the shoes of white, and that would be the right decision. But howmany of us would have went on for the gammon? To end all confusion I give the results with white in possession of the cube.

White - Pips 88

Black - Pips 172
Black to Play 5-5
Created with www.BGdiagram.com

Code:

1. 23/18(3) 12/7                Eq.:  +0,653
       0,800 0,187 0,009 - 0,200 0,067 0,006 CL  +0,722 CF  +0,653
      
2. 23/8 12/7                    Eq.:  +0,510 ( -0,144)
       0,757 0,189 0,009 - 0,243 0,102 0,011 CL  +0,597 CF  +0,510

 Truncated cubeful rollout (depth 10) with var.redn. 
        147 games, Mersenne Twister dice gen. with seed 770449777 and quasi-random dice 
        Stop when std.errs. are small enough: ratio 0,1 (min. 144 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

01-29-2015 , 07:32 AM

#155

Taper_Mike

adept

Join Date: Sep 2009 Posts: 1,030

Black on roll. Cube action?
XGID=-a-BBBBA----A----a-cd-fCB-:2:1:1:00:0:0:0:0:10
White - Pips 88

Black - Pips 172

Below is the rollout of the cube decision I made for Yogiman. As was surmised by Z, the proper cube action is Redouble/Pass. In addition to the XG rollout below, I also made a 1k GnuBg rollout that yielded the same result.

It is interesting to note that an XGR++ eval rates this as Redouble/Take, as does the truncated GnuBg rollout I created that mimics XGR++.

A question for Z: Have you checked GnuBg since new neural nets were introduced in version 1.00? Do you know whether the new neural nets have the problems with containment cubes that you alluded to?

Mike

2k XG2 Rollout – Unlimited Game, No Jacoby, No Beavers

Code:

XGID=-a-BBBBA----A----a-cd-fCB-: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: 172  O: 88 X-O: 0-0
Cube: 4, X own cube
X on roll, cube action

Analyzed in Rollout
No redouble
  Player Winning Chances:   73.86% (G:21.67% B:1.08%)
  Opponent Winning Chances: 26.14% (G:11.64% B:1.60%)
Redouble/Take
  Player Winning Chances:   75.04% (G:22.29% B:1.14%)
  Opponent Winning Chances: 24.96% (G:11.18% B:1.49%)

Cubeless Equities: No Double=+0.5722, Double=+1.2170

Cubeful Equities:
       No redouble:     +0.8598 (-0.1402)
       Redouble/Take:   +1.0730 (+0.0730)
       Redouble/Pass:   +1.0000

Best Cube action: Redouble / Pass

Rollout:
  2592 Games rolled with Variance Reduction.
  Dice Seed: 58166652
  Moves: 3-ply, cube decisions: XG Roller
  Confidence No Double: ±0.0103 (+0.8494..+0.8701)
  Confidence Double:    ±0.0175 (+1.0555..+1.0905)

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

eXtreme Gammon Version: 2.10

Quote

01-29-2015 , 10:15 AM

#156

yogiman

adept

Join Date: Jun 2012 Posts: 784

Position ID: 2F6DATBmO8EBMA Match ID: cAkFAAAAAAAE

White - Pips 139

Black - Pips 135
Black to Play 2-1
Created with www.BGdiagram.com

Quote

01-29-2015 , 11:41 AM

#157

uberkuber

veteran

Join Date: Jul 2009 Posts: 3,032

24/22 8/7 with some 6s duplication.

Quote

01-29-2015 , 02:02 PM

#158

_Z_

adept

Join Date: Jul 2004 Posts: 720

Quote:

Originally Posted by Taper_Mike

A question for Z: Have you checked GnuBg since new neural nets were introduced in version 1.00? Do you know whether the new neural nets have the problems with containment cubes that you alluded to?

I have not investigated this closely, though I think the problems might still be there. For Problem 30, a gnu rollout largely agrees with XG -- the D/T equity is 1.05 or so. But the 2-ply eval for D/T is 0.87. So it's off by close to 0.2. I think I'm using a recent version, it says it's:

Version GNU Backgammon 1.04.000-mingw 20141021

and

* Contact neural network evaluator:
- version 1.00, 250 inputs, 128 hidden units.

Last edited by _Z_; 01-29-2015 at 02:07 PM.

Quote

01-29-2015 , 04:52 PM

#159

_Z_

adept

Join Date: Jul 2004 Posts: 720

Problem 31 --

13/11 8/7 is kind of popping out at me, largely due to the duplication of 6s.

It seems a little funny because White has the better priming structure, so usually you'd want to split to combat the prime instead of trying to build a counter-prime where you are outgunned, but it looks like Black would be stepping into too much of an attack if he plays something like 24/22 24/23 (probably the best split). There are 11 White checkers there that are looking to blow Black off the board.

Not particularly confident about this one though.

Quote

01-30-2015 , 01:54 AM

#160

bleep69

grinder

Join Date: Sep 2008 Posts: 483

I didn't want to split the back men, given the ammo that white has lurking, which led me to 13/11 8/7.

Only after reading the responses did I notice the handy duplication of 6's, sadly.

Quote

01-30-2015 , 06:21 AM

#161

Taper_Mike

adept

Join Date: Sep 2009 Posts: 1,030

OTB, I doubt I would have found slotting the bar point. I was leaning towards two-up (24/23 24/22) before I read the other posts. Once you decide it is too dangerous to break anchor, then it makes sense to go for the counter-prime.

Mike

Quote

01-31-2015 , 07:48 AM

#162

yogiman

adept

Join Date: Jun 2012 Posts: 784

Code:

    1. 13/11 8/7                    Eq.:  -0,184
       0,453 0,167 0,012 - 0,547 0,210 0,019 CL  -0,144 CF  -0,184
     
    2. 24/23 24/22                  Eq.:  -0,287 ( -0,104)
       0,438 0,173 0,016 - 0,562 0,270 0,016 CL  -0,221 CF  -0,287
     
    3. 13/10                        Eq.:  -0,297 ( -0,114)
       0,424 0,154 0,012 - 0,576 0,206 0,015 CL  -0,207 CF  -0,297
      
        Truncated cubeful rollout (depth 10) with var.redn.
        146 games, Mersenne Twister dice gen. with seed 770326265 and quasi-random dice
        Stop when std.errs. are small enough: ratio 0,1 (min. 144 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]

13/10 doesn't do very much. If white throws a 6, than in his next turn he will have a very strong double.
If we move one of white's four checkers on his 6 point to his 13 point, 24/23 24/22 will be slightly better. This is somewhat surprising as white's chance to make a point on one of those heads is almost 50%, but black's board is still better. So timing is an important consideration here: in the latter white's pipcount will be 14 points more than black's. However, the original 4 points difference is little, and black's hind checkers have somewhat more room than white's. If white throws a 6 he will hit, but no panick, in more than halve of the cases black's chance to hit back from the bar will be 60%.

Quote

01-31-2015 , 12:03 PM

#163

yogiman

adept

Join Date: Jun 2012 Posts: 784

Position ID: rJ3QACOwt+0AAA Match ID: QQkbAAAAAAAE

White - Pips 150

Black - Pips 112
Black to Play 6-6
Created with www.BGdiagram.com

Quote

01-31-2015 , 03:36 PM

#164

Taper_Mike

adept

Join Date: Sep 2009 Posts: 1,030

I may get dinged here, because the play that seems obvious is the play I want to make! Hitting is what seems clear. That will force White forward, and perhaps give a tempo if he dances. Then I would make the 3pt, rather than the 2pt. The gaps seem less dangerous that way.

9/3(2), 7/1*(2)

Mike

Quote

02-01-2015 , 07:00 AM

#165

yogiman

adept

Join Date: Jun 2012 Posts: 784

Code:

    1. 9/3(2) 7/1*(2)               Eq.:  +0,744
       0,790 0,250 0,003 - 0,210 0,019 0,000 CL  +0,813 CF  +0,744
  
    2. 8/2(2) 7/1*(2)               Eq.:  +0,598 ( -0,147)
       0,741 0,231 0,003 - 0,259 0,032 0,001 CL  +0,683 CF  +0,598
      
        Truncated cubeful rollout (depth 10) with var.redn.
        146 games, Mersenne Twister dice gen. with seed 770498836 and quasi-random dice
        Stop when std.errs. are small enough: ratio 0,1 (min. 144 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]

It was not obvious to me. 7/1*(2) 8/2(2) leaves one twopoint gap in the outerboard and facilitates the dump of checkers or an eventual pick and pass, versus 7/1*(2) 9/3(2) which leaves two onepoint gaps. If white throws a 4, the twopoint gap will be more unfavorable, but that difference doesn't look dramatic. The real nuisance is when there is one outerboard point left. In that case white's 22-point will offer some more shots than his 23 point. And if left with a 9-point it will be significantly more dangerous. So it seems that the equity difference is caused by white's potential artillery pointed on blacks outerboard.

By the way, a number of my saved codes have become useless, because the dicenumbers have changed. Is this a familiar problem?

Quote

02-01-2015 , 07:40 PM

#166

Taper_Mike

adept

Join Date: Sep 2009 Posts: 1,030

Quote:

Originally Posted by yogiman

By the way, a number of my saved codes have become useless, because the dice numbers have changed. Is this a familiar problem?

I have never had the dice roll in an ID change by itself.

It is important is to use the Position ID and Match ID together. If you paste a Position ID without a Match ID, GnuBg will have to guess which player is on roll.

If you have a GnuBgID for a certain position, and you want to change the roll, the easiest way to do that is to paste it into GnuBg, and then click Edit on the toolbar. This is the same thing as choosing Edit Position from the Edit menu. From there, click the dice on the board, and select a different roll. Lastly, click Edit to leave editing mode, and press Ctrl+C to copy the new ID.

Hope this answers your question!

Mike

Quote

02-02-2015 , 05:53 AM

#167

yogiman

adept

Join Date: Jun 2012 Posts: 784

Position ID: TbuBGQCw3QY1AA Match ID: cIkOAAAAAAAE

White - Pips 107

Black - Pips 138
Black to Play 5-3
Created with www.BGdiagram.com

Quote

02-02-2015 , 12:15 PM

#168

_Z_

adept

Join Date: Jul 2004 Posts: 720

Maybe I'm missing something, but I'm not seeing anything better than 14/6. You could play 14/9 7/4, but there may be an immediate exchange of hits, and you might not want the blot on your 4 point. 14/6 has good building power without the risk of slotting.

Quote

02-02-2015 , 12:24 PM

#169

_Z_

adept

Join Date: Jul 2004 Posts: 720

Just a little warning, and I'm not sure if it applies here -- if a bot screws up the cube action/checker play immediately after one or more of the plays, a rollout will be suspect.

Last edited by _Z_; 02-02-2015 at 12:41 PM.

Quote

02-02-2015 , 12:58 PM

#170

Robertie

old hand

Join Date: May 2004 Posts: 1,954

As with many of these positions, I'm not sure what the 'obvious' play is supposed to be. Here I'd just play 9/4 7/4, making a nice 5-point prime with the back slotted. White has two blots in his board and he'll need to break a point to hit, so if I get hit I'm guaranteed plenty of return shots.

Quote

02-02-2015 , 01:26 PM

#171

Fllecha

journeyman

Join Date: Mar 2010 Posts: 321

I think that the board is wrong placed. The 4 men i think that are stragglers and white is playing a 4 man back game. In that case i would not double hit and simply hit one checker and advance a straggler

Quote

02-02-2015 , 01:32 PM

#172

_Z_

adept

Join Date: Jul 2004 Posts: 720

Quote:

Originally Posted by Robertie

The obvious play is the one that doesn't leave a quadruple shot...

If 9/4 7/4 is right, that would be a great play!

Quote

02-02-2015 , 02:42 PM

#173

Fllecha

journeyman

Join Date: Mar 2010 Posts: 321

Quote:

Originally Posted by _Z_

If 9/4 7/4 is right, that would be a great play!

Humm great? If the position is correct placed Id say that is close to standard: oppo has two blot and gaps and we have a strong board... I expect yogy to post.

Quote

02-02-2015 , 04:31 PM

#174

Robertie

old hand

Join Date: May 2004 Posts: 1,954

Quote:

Originally Posted by Fllecha

The points in the diagram are numbered, so Black is bearing off to the lower right and White is bearing off to the upper right. It's a mutual holding game, not a backgame.

The backgame position is also interesting, however. In that case I'd play 21/16*/13*, hitting two.

Quote

02-02-2015 , 04:48 PM

#175

_Z_

adept

Join Date: Jul 2004 Posts: 720

I thought it was a backgame problem too at first glance

But I'm pretty sure yogi intended the holding game position.

I would hit two in the backgame position with the idea that I want to keep White busy while I escape and try to build the 9 point. Escape work is tougher however, so hit one and advance is quite reasonable.

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