Re: [Bug-gnubg] "Joseph-ID" in benchmark db

[Joseph Heled]

Converting position probabilities into cubeful error measures is open
to personal interpretation.

My script (perr.py) provides measures for a variety of match scores [
(7,7), (3,3), (5,5), (9,9), (25,25), (2,3), (2,4), (2,5), (2,6),
(2,7), (3,4), (3,5), (3,6), (3,7), ... ]

Note: The cube decision uses my own code, not the one currently used by gnubg.

For each position, I look at the cube action using the rollout values
and the net values. If cube action is the same, the error is zero (0).
If the two actions disagree, two error statistics are computed:
   The cubeful one is the cubeful equity difference between the two
decisions, using the match equities computed from the rollout.

   The cubeless is the money equity difference between the two
evaluations, with an optional "correction". The correction looks how
"far" the net probs are from making the same decision as the rollout
ones, and weighs the error accordingly.

-Joseph

On 25 February 2012 10:49, Mark Higgins <address@hidden> wrote:
> Can you clarify what "rollout values of the outcome probability" mean pls?
> eg how do I turn those into a cubeful equity, or a cube decision?
>
> Is there any benchmark for cube decisions directly, like whether to
> double/redouble at the start of a turn, and whether the opponent should
> take/pass?
>
>
>
> On Feb 10, 2012, at 6:26 AM, Øystein Schønning-Johansen wrote:
>
>> 'r' is the seed used for the rollout, I think
>
>
> Sound likely, since there is an 'r'-line for every rollout result. But there
> is no code lines in perr.py to conferm it. I guess we can trust your memory
> on that.
>
> 'o' is the cube rollout, and the numbers are the rollout values of the
> outcome probability,
>
> -Øystein
>
>