>
>
> Not sure anyone has been following rec.games.backgammon, but I am
(And so
> are others) wondering how this position can be " Too good to
redouble, take"
>
Uuhh, I was under the impression that no such thing
exists ... if it's too good to redouble, then it's a pass. By definition.
> I can understand that its possible to be Too
good to redouble, take if the
> opponent can be gammoned, I'm just not sure how this can be possible
with
> the following position. This rollout was done as Grandmaster checker
(3ply),
> supremo cube(2ply) for both players. The result was:
> > The score (after 2 games) is: mpetch 1, Monty
4 (match to 15 points)
> Match Information:
>
> Date: April 13, 2008
> Move number 67: Monty on roll, cube decision? > GNU Backgammon Position ID: BwAAmG1bCkEAAA
> Match ID
: UQngARAAIAAA
> +13-14-15-16-17-18------19-20-21-22-23-24-+ O:
mpetch
> | X |
O | X O | OO 1 point
> | |
O | O | OO
> | |
| O | OO
> | |
| |
OO
> | |
| |
OO
> v| |BAR|
|
15 point match
> | |
| |
> | |
| |
> | |
| |
> | |
| X X X X X |
On roll
> | X X X | |
X X X X X | 4 points
> +12-11-10--9--8--7-------6--5--4--3--2--1-+ X:
Monty (Cube: 2)
> Pip counts: O 53, X 99
>
> * Monty doubles
>
> Alert: wrong double ( -0.025)! [very bad]
> Cube analysis
> Rollout cubeless equity +0.587 (Money: +0.587)
>
> Cubeful equities:
> 1. No double +1.015
> 2. Double, pass +1.000 ( -0.015)
> 3. Double, take +0.990 ( -0.025)
> Proper cube action: Too good to redouble, take
> Rollout details:
> Player Monty owns 2-cube:
> 0.795 0.000 0.000 - 0.205 0.003 0.000 CL +0.587 CF +1.015
> [0.001 0.000 0.000 - 0.001 0.000 0.000 CL 0.003 CF
0.005]
> Player mpetch owns 4-cube:
> 0.800 0.000 0.000 - 0.200 0.004 0.001 CL +1.178 CF +0.990
> [0.001 0.000 0.000 - 0.001 0.000 0.000 CL 0.006 CF
0.007]
> Full cubeful rollout with var.redn.
> 1296 games, Mersenne Twister dice gen. with seed 824585691 and quasi-random
> dice
> Play: grandmaster 3-ply cubeful prune [grandmaster]
> keep the first 0 0-ply moves and up to 8 more moves within equity
0.16
> Skip pruning for 1-ply moves.
> keep the first 0 2-ply moves and up to 2 more moves within equity
0.04
> Cube: 2-ply cubeful prune [world class]
Maybe it's an oddity given by the statistical nature
of a rollout: the std err on the equity of the take and/or the pass
is so high that the two equities can cross the 1.000 border randomly (and
here we have stumbled on a case where they are inconsistent).
However I've noticed that evaluations swings *a lot*
(I mean really a lot) between even and odd plies:
GNU Backgammon Position ID: BwAAmG1bCkEAAA
Match ID : UQngARAAIAAA +13-14-15-16-17-18------19-20-21-22-23-24-+
O: gnubg | X
| O | X O
| OO 1 point |
| O |
O | OO |
| |
O | OO |
| |
| OO |
| |
| OO v|
|BAR|
| 15 point match |
| |
| |
| |
| |
| |
| |
| | X X X X X
| On roll | X X
X | | X X X X X |
4 points +12-11-10--9--8--7-------6--5--4--3--2--1-+
X: MaX (Cube: 2)
Cube analysis 0-ply cubeless MWC 65.350% (Money: +0.3038) 0.6530 0.0000 0.0000 - 0.3470 0.0020 0.0002 Cubeful equities: 1. No double 66.955% 2. Double, pass 71.200%
( 4.245%) 3. Double, take 64.479%
( -2.476%) Proper cube action: No redouble, take (36.8%)
That's HUGE !! I don't know if we are allowed to keep
on blaming this on the odd/even effect (especially since Snowie
seems to much less affected by this).
I'm wondering if we have a bug in the eval code for
odd plies ...