Re: [Bug-gnubg] Positive error rate in statistics

[Jon Kinsey]

Holger wrote:
> Hi Øystein,
>
> At 12:44 15.02.05 +0100, you wrote:
>
> > I'm able to recreate this bug with the files you atteched. However I can't
> > recreate this with matches I play against GNU Backgammon. I can't explain
> > this?
>
>
> This only happened very rarely and only with a few matches played on TMG.
> Never occurred here with other types of matches.
>
> My usual procedure is this: I batch-analyse my matches with a Perl script
> fed gnubg-nogui that imports (in this case) the .tmg file and saves the
> analysed match as .sgf. Later I look at the match with the GUI version.
>
> Btw, on searching the mailing list I've found something related:
> http://lists.gnu.org/archive/html/bug-gnubg/2003-09/msg00309.html

I looked at a similar problem this is what I said at the time:

-----------------

Christian Anthon wrote:

> when a tmg match is resigned after the dice have been rolled the
> non-move is reported by tmg as 0/0. The import filter mis-interprets
> this and the result is that the assigned match-equity is 50%, messing up
> the statistics rather badly.


I've had a quick look and this may be a more general problem of
analysing invalid moves.  The 0/0 is treats as a fan in tmg files, but a
fan isn't possible on move 21.

Does anyone know how invalid moves are analysed?

More technically: in AnalyzeMove() move 83 is copied to position 19
(there are only 82 possible moves).

Jon

signature.asc
Description: OpenPGP digital signature