Hi all,
I've found an interestin position analyzed
by an expert on a web-site.
If you want, try to find your move
before reading the rest of the message.
GNU Backgammon Position
ID: 820HACC2uREDEA
Match ID : cIkEAAAAAAAA
+24-23-22-21-20-19------18-17-16-15-14-13-+
O:
| O X O O
O O | |
X |
| O O O
O O | |
X |
|
O O | |
|
|
O | |
|
|
O | |
|
|
|BAR|
|v (Cube: 1)
|
| |
|
|
| |
|
|
X | |
|
| X X
X X | | X
| Rolled 11
| O X X X
X | | X X
|
+-1--2--3--4--5--6-------7--8--9-10-11-12-+
X:
Analysis included a Snowie rollout
and I was surprised by the absence
of "my" move (13-11(2)) between
the moves rolled out. So I posted the
position to rec.games.backgammon and
did my own (gnubg) rollout.
The rollout says it's a close cal between
7-5(2) and 13-11(2), but the
strange thing is that the rollout cubeless
equities of the two moves
are very different (7-5(2) is better).
What follows is the rollout made by
Raccoon (2592 trials !!), with results
and comments. Thinking gnubg may misplay
some cube decisions (as O) he has
tried to find errors and has found
2 cases (one small error and a big one)
where the 2-ply eval is way off with
respect to the 2ply rollout.
MaX.
************ Post by Raccoon on rec.games.backgammon
****************
Good catch by Max to have found a reasonable
play -- 13/11(2) -- that
was apparently overlooked by the analyzer
at www.gammonlife.com, Mary
Hickey, a fine player who has also
analyzed problems for The Chicago
Point.
A Gnubg rollout with my usual settings
has 13/11(2) and 7/5(2) dead
even after 2592 trials:
GNU Backgammon Position
ID: 820HACC2uREDEA
Match ID : cIkEAAAAAAAA
+24-23-22-21-20-19------18-17-16-15-14-13-+
O:
| O X O O
O O | |
X |
| O O O
O O | |
X |
|
O O | |
|
|
O | |
|
|
O | |
|
|
|BAR|
|v (Cube: 1)
|
| |
|
|
| |
|
|
X | |
|
| X X
X X | | X
| Rolled 11
| O X X X
X | | X X
|
+-1--2--3--4--5--6-------7--8--9-10-11-12-+
X:
Full cubeful rollout with var.redn.,
Mersenne Twister dice gen. with
seed 1057041626 and quasi-random dice,
Play: 2-ply cubeful prune
[world class] , keep the first 0 0-ply
moves and up to 8 more moves
within equity 0.12, Skip pruning for
1-ply moves. Cube: 2-ply cubeful
prune [world class]
However, a puzzling statistic in both
Max's rollout and mine is that
7/5(2) is clearly better cubeless.
Why does 13/11(2) gain so much more
cubeful? With correct cube action,
perhaps it doesn't. Perhaps GNU
Backgammon is making cube errors after
13/11(2), or more cube errors
after that play than after 7/5(2).
I went looking for errors and found
two.
(1) A small cube error:
X 11 13/11(2) ... O 43 24/17
X 31 11/8/*7 ... O fan
GNU Backgammon Position
ID: 820HAEC2uaMAEA
Match ID : cAkAAAAAAAAA
+24-23-22-21-20-19------18-17-16-15-14-13-+
O:
| O X O O
O O | O |
|
| O O O
O O | |
|
|
O O | |
|
|
O | |
|
|
O | |
|
|
|BAR|
|v (Cube: 1)
|
| |
|
|
| |
|
|
X | | X
|
| X X
X X | | X
| On roll
| X X
X X | | X X X
|
+-1--2--3--4--5--6-------7--8--9-10-11-12-+
X:
Rollout details:
Centered 1-cube:
0.7605 0.0481 0.0012 - 0.2395
0.0391 0.0009 CL +0.5303 CF +0.7144
[0.0011 0.0012 0.0002 - 0.0011
0.0013 0.0001 CL 0.0025 CF 0.0056]
Player White owns 2-cube:
0.7691 0.0596 0.0018 - 0.2309
0.0297 0.0006 CL +1.1385 CF +1.0194
[0.0014 0.0025 0.0002 - 0.0014
0.0012 0.0002 CL 0.0074 CF 0.0086]
Full cubeful rollout with var.redn.,
648 games, Mersenne Twister dice
gen. with seed 1057041626 and quasi-random
dice, Play: 2-ply cubeful
prune [world class], keep the first
0 0-ply moves and up to 8 more
moves within equity 0.12, Skip pruning
for 1-ply moves. Cube: 2-ply
cubeful prune [world class]
(2) A huge cube error:
X 11 13/11(2) ... O 55 6/1(3)
GNU Backgammon Position
ID: n28DACC2udEAEA
Match ID : cAkAAAAAAAAA
+24-23-22-21-20-19------18-17-16-15-14-13-+
O:
| O X O O
O | |
|
| O O O
O | |
|
| O O
| |
|
| O O
| |
|
| O O
| |
|
|
|BAR|
|v (Cube: 1)
|
| |
|
|
| |
|
|
X | |
|
| X X
X X | | X
X | On roll
| O X X X
X | | X X X
|
+-1--2--3--4--5--6-------7--8--9-10-11-12-+
X:
Full cubeful rollout with var.redn.,
382 games, Mersenne Twister dice
gen. with seed 1057041626 and quasi-random
dice, Play: 2-ply cubeful
prune [world class], keep the first
0 0-ply moves and up to 8 more
moves within equity 0.12, Skip pruning
for 1-ply moves. Cube: 2-ply
cubeful prune [world class]