Game 1, move 2, Readers 3-1?

[Zorba]

Readers moved 13/11 6/5; dorbel rolled a 6-5 for an obvious run 24/13, now it's Readers with a 3-1 to play.

XGID=-a---AD-C--AfD---c-e----B-:0:0:1:31:0:0:0:5:10

[ah_clem]

Spoiler

Obviously, we make the five point.  Almost as obvious is to cover the blot with 6/5  - not 8/5 - we want to unstack the six and leave the spare on the 8.

Where to play the three?  11/8 is safe but does nothing.  24/21 gets the backman in motion, and that's a reasonable choice. But I like bringing in another builder with 13/10. This play unstacks both our candlesticks and threatens to prime Dorbel's runner. 

13/10 6/5

[close]

[diane]

Snap, for pretty much those reasons exactly.

[AlpineDesert]

Spoiler

making the 5 with the 1 (6-5) is clear.  now that we've made our golden pt we need to put our back men in motion, especially in light of the fact that dorbel has already escaped one man.  24-21 is my choice for the 3.

[close]

[ah_clem]

Hmmm.  It seems our biggest threat at this point is Dorbel escaping the second checker.  He's making no threats to prime our runners, so the defensive move can wait. Bringing down a second builder improves our chances of making the bar point from 6/36 to 12/36 -that's double the chances!  If we can make the bar, we've developed a nice blockade that should put us in good stead for the rest of the game.  The builder on the 10 also gives us an additional 4 rolls to make the four point.  It's a good-point-making-machine.

Granted, 24/21 gives us an additional 13 rolls to make an anchor (15 if you include anchoring on his bar point) and 4 additional escaping rolls (although 5-3 would probably be played to make a home board point).  But we don't need an anchor at this point.  Priming threats or your opponent making an anchor is a cue to get your runners off the ace.  Escaping a single checker isn't one of the cues - look at his position:  he's got two ineffectual stacks and makes no real threats; the position is weak, so attack!  This is not the time to play defense - he threatens nothing worth defending against.

[Zorba]

Tough choice, but I went with 24/21. ah_clem's arguments for bringing one down are good, but there's more to 24/21 than just a defensive move: It guards dorbel's outfield much better and with many rolls, dorbel might want to unstack his heavy midpoint to develop his position. Running two checkers to the midpoint with nothing else is not that great for him.

Also, 6-3 is one of dorbel's running rolls, and gives a big swing if we get hit with it, instead of having a double shot at the runner. The builder on 11 already gives us some good flexibility upfront, another one on 10 might be a bit overkill at this point.

Should dorbel succeed in escaping two backmen, then he's not much of a favourite if we have the 4pt anchor and the race is still close. He will have trouble playing safe for the remainder of the game, and we can often turn it in to a race later if we roll good doubles at some point (for which we will usually have many turns to do so).

That said, 13/10 has a lot of merit too, and I wouldn't mind playing that move. Looks a bit like a toss-up to me.

[sixty_something]

good analysis, Zorba .. that is exactly what i was going to say

[gumpi]

Should be the split, getting more controll on dorbel´s side o the board(his position is stacked). The offense ist set up well anyway...

[Zorba]

XG evaluations:

Spoiler

XG 3-ply (eqv. to GnuBG 2-ply) prefers 13/10 6/5 by a small margin, but higher levels like 24/21 6/5.

XG roller + (this is a very fast mini rollout that can be used instead of evaluations):


X to play 31

    1. XG Roller+  24/21 6/5                    eq:+0,150
      Player  : 53,02% (G:14,59% B:0,44%)
      Opponent: 46,98% (G:9,85% B:0,38%)

    2. XG Roller+  13/10 6/5                    eq:+0,136 (-0,014)
      Player  : 52,95% (G:15,15% B:0,51%)
      Opponent: 47,05% (G:11,11% B:0,51%)

eXtreme Gammon Version: 1.14, MET: Rockwell-Kazaross

[close]

[ah_clem]

Spoiler

XG evaluations:


XG 3-ply (eqv. to GnuBG 2-ply) prefers 13/10 6/5 by a small margin, but higher levels like 24/21 6/5.

Gnubg has similar results.  2-ply analysis and 0-ply rollout favor bringing down a builder.  2-ply rollout favors the split.  Interesting.  I've been training with 2-ply analysis (analyzing all my matches) so it's to be expected that I'd tend to favor the same plays as gnubg 2-ply analysis.  It's not always right, but it's usually close enough.  (c:

This is a close one, although the builder seemed obvious to me.  Recognizing that the split does more than just slot an anchor is key here.


   

    1. Cubeful 2-ply    13/10 6/5                    MWC:  51.28%
       0.536 0.159 0.004 - 0.464 0.110 0.004

    2. Cubeful 2-ply    24/21 6/5                    MWC:  51.17% ( -0.10%)
       0.532 0.146 0.004 - 0.468 0.099 0.003
        2-ply cubeful prune [world class]
   


    1. Rollout          13/10 6/5                    Eq.:  +0.152
       0.529 0.156 0.006 - 0.471 0.119 0.006 CL  +0.104 CF  +0.152
      [0.002 0.002 0.001 - 0.002 0.001 0.000 CL   0.006 CF   0.014]
       
    2. Rollout          24/21 6/5                    Eq.:  +0.134 ( -0.018)
       0.525 0.147 0.006 - 0.475 0.103 0.006 CL  +0.098 CF  +0.134
      [0.002 0.002 0.001 - 0.002 0.001 0.001 CL   0.006 CF   0.014]
        Full cubeful rollout with var.redn.
        1296 games, Mersenne Twister dice gen. with seed 892351531 and quasi-random dice
        Play: 0-ply cubeful prune [expert]
        Cube: 0-ply cubeful prune [expert]
 



    1. Rollout          24/21 6/5                    Eq.:  +0.175
       0.529 0.150 0.005 - 0.471 0.103 0.007 CL  +0.115 CF  +0.175
      [0.002 0.002 0.001 - 0.002 0.002 0.001 CL   0.005 CF   0.011]
        Full cubeful rollout with var.redn.
     
    2. Rollout          13/10 6/5                    Eq.:  +0.159 ( -0.016)
       0.528 0.161 0.007 - 0.472 0.115 0.007 CL  +0.106 CF  +0.159
      [0.002 0.002 0.001 - 0.002 0.002 0.001 CL   0.005 CF   0.011]
        Full cubeful rollout with var.redn.
        1296 games, Mersenne Twister dice gen. with seed 892351531 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]

[close]

[ah_clem]

Spoiler

 
Note that all the bot results have the two plays winning almost the same percentage of games, the difference is in the number of gammons.  13/10 wins more gammons, but also loses more gammons, suggesting that the choice of play here is heavily influenced by the match score.  At gammon-go we'd play the builder, and at gammon-save we'd split.

Good position and good discussion.

[close]

[sixty_something]

Good position and good discussion. 

i would suggest this was an excellent discussion especially considering that the move at first blush seems patently obvious, at least it does to me .. only recently did i begin following these matches and my thinking was that 90% or more of the moves would be no brainers .. i would have said this was a no brainer with 99% certainty .. well, guess who has egg on his face

kudos to those of you who came up with this idea and have implemented it 

[Zorba]

A late XG rollout result:

Spoiler

Pretty much in line with previous results:

    1. Rollout¹    24/21 6/5                    eq:+0,149
      Player  : 52,64% (G:14,85% B:0,64%)
      Opponent: 47,36% (G:10,22% B:0,60%)
      Confidence: ± 0,005 (+0,144<E<+0,154)
      Duration: 2 hours 04 minutes

    2. Rollout¹    13/10 6/5                    eq:+0,136 (-0,013)
      Player  : 52,54% (G:15,46% B:0,73%)
      Opponent: 47,46% (G:11,55% B:0,78%)
      Confidence: ± 0,005 (+0,131<E<+0,141)
      Duration: 2 hours 08 minutes

¹ 31104 Games rolled with Variance Reduction.
  Moves: 1 ply, cube decisions: 3 ply Red

eXtreme Gammon Version: 1.14, MET: Rockwell-Kazaross

[close]