Re: [Bug-gnubg] Cubeless equity in cubeful rollouts

[Christian Anthon]

Without consulting the code, I'm guessing the following: For money
games the cubeless equity is a fairly well defined thing, as the
relative cost of a point, gammon or backgammon is independent of the
cube value. For matches it is different. Suppose the there is a
double/take in a game in your example game. Now, should the cubeless
equity of such a game be calculated with the cube at two or one.
Neither is probably a very good idea, but I'm guessing that the
evaluation is done with the cube at two, which explains the difference
in numbers that you are observing.

The bottom line: don't use the CL equities for matches.

Christian.

On Sun, Dec 14, 2008 at 7:59 PM,  <address@hidden> wrote:
> I have a question concerning the cubeless equity derived from cubeful
> rollouts
>
>
>
> Date: Dezember 14, 2008
> Move number 3: X to play 32
>
>     GNU Backgammon  Position ID: mGdrBgDLzInBAA
>                     Match ID   : cImpACAAAAAA
>     +24-23-22-21-20-19------18-17-16-15-14-13-+  O: somebody
>     |          O     O |   | X  O  O  O  O  X |  2 points
>     |          O     O |   | X  O  O     O  X |
>     |                O |   |                  |
>     |                O |   |                  |
>     |                  |   |                  |
>     |                  |BAR|                  |v 5 point match (Cube: 1)
>     |                  |   |                  |
>     |                  |   |                  |
>     |                  |   |    X             |
>     | X        X     X |   |    X           O |  Rolled 32
>     | X  X     X     X |   |    X     X     O |  0 points
>     +-1--2--3--4--5--6-------7--8--9-10-11-12-+  X: You
> Pip counts: O 124, X 120
>
> * You moves 18/15*/13
>
>
> This is a cubeless rollout result:
>
>     1. Rollout          8/6 8/5                      Eq.:  -0,008
>        0,498 0,015 0,000 - 0,502 0,014 0,001 CL  -0,008
>       [0,001 0,001 0,000 - 0,001 0,001 0,000 CL   0,003]
>         Full cubeless rollout with var.redn.
>         1296 games, Mersenne Twister dice gen. with seed 812588963 and
> quasi-random dice
>         Play: 0-ply cubeful prune
>         Cube: 0-ply cubeful prune
>     2. Rollout          18/15*/13                    Eq.:  -0,134 ( -0,127)
>        0,464 0,174 0,005 - 0,536 0,174 0,009 CL  -0,134
>       [0,002 0,002 0,001 - 0,002 0,002 0,001 CL   0,006]
>         Full cubeless rollout with var.redn.
>         1296 games, Mersenne Twister dice gen. with seed 812588963 and
> quasi-random dice
>         Play: 0-ply cubeful prune
>         Cube: 0-ply cubeful prune
>
>
> And this is a cubeful rollout result:
>
>     1. Rollout          18/15*/13                    Eq.:  +0,031
>        0,465 0,176 0,006 - 0,535 0,176 0,015 CL  -0,025 CF  +0,031
>       [0,002 0,002 0,001 - 0,002 0,002 0,002 CL   0,006 CF   0,012]
>         Full cubeful rollout with var.redn.
>         1296 games, Mersenne Twister dice gen. with seed 812585055 and
> quasi-random dice
>         Play: 0-ply cubeful prune [expert]
>         Cube: 0-ply cubeful prune [expert]
>     2. Rollout          8/6 8/5                      Eq.:  -0,009 ( -0,040)
>        0,500 0,017 0,000 - 0,500 0,014 0,001 CL  -0,001 CF  -0,009
>       [0,001 0,001 0,000 - 0,001 0,001 0,000 CL   0,003 CF   0,006]
>         Full cubeful rollout with var.redn.
>         1296 games, Mersenne Twister dice gen. with seed 812585055 and
> quasi-random dice
>         Play: 0-ply cubeful prune [expert]
>         Cube: 0-ply cubeful prune [expert]
>
>
>
> I know why we should not expect the w/g/bg breakdowns to be the same (with
> the cubeful rollout being truncated in case of a double/pass), but anyway in
> this case they are close. What I don't understand at all is why the cubeless
> equity of 18/15*/13 differs so much in the cubeful and in the cubeless
> rollout.
>
>
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/bug-gnubg
>
>