Re: [Bug-gnubg] Problems with Gnubg 04-Mar-2009 release
From:
Massimiliano Maini
Subject:
Re: [Bug-gnubg] Problems with Gnubg 04-Mar-2009 release
Date:
Mon, 16 Mar 2009 16:55:42 +0100
address@hidden wrote on 14/03/2009 12:36:59:
> Dear Christian, Joaquín and all, > > I have some further diagnosis and observations
since yesterday that
> I would like to share with you upon Joaquín's request: > > 8) I realize now that my older version (Sep-08)
is analyzing the
> match "on-the-go" [player0, analysis and evaluation settings
in both
> versions all set as checker play="supremo", cube play= "worldclass",
> tutor decision="same as evaluation"]. What I mean with this
is that
> when I for example make a very lucky, very unlucky, doubtful, bad,
> very bad move etc, I immediately see its effects (bold, italics,
> colored etc.) in the "Game Record" panel after I confirm
my move by
> clicking on the dice. > > On the other hand, the newer version (Mar-09)
does not do this "on-
> the-go" analysis right after the move. All comments (bold, italics,
> bad, very bad, lucky, unlucky etc.) only appear in the "Game
Record"
> panel, if and only if you perform a full analysis of the match
> afterwards. In addition, the missing "delta equity" for
each move
> with respect to the average dice roll in the parenthesis also
> appears surprisingly after the analysis!
I've uploaded a new version with the most recent code
(20090316): I'm not sure all the issues are gone (Christian thinks
there's still something left), but at least I can see the colors/comments
in the game record as soon as a move is made. > This is basically the reason why it is taking
so short for the older
> version to analyze the match (since it is just analyzing gnubg's
> moves - half the work!), whereas the newer version starts everything
> from scratch. I double-checked the settings for everything and can
> assure you that they are the same. Does this mean anything to you?
> Do you encounter the same problem? Is it in anyway related to an
> already addressed error?
To be checked, but I'm also nuder the impression that
the analysis does not take into account the "analysis" done by
the tutor. As you said, if the tutor and analysis levels are the same, analysis is
already done and, in the past it wasn't necessary to do it again (making
the analysis at the end of the match very fast). > C) As I wrote in my original mail, those cases
were definitely wins
> or losses for Gnubg and its equity was already probably -1.000 or
> +1.000. So Joaquín's argument makes sense. Since all possible moves
> are having the same equity value at that instant, Gnubg cannot
> differentiate between them and probably picks "any random"
one
> amongst them. That's the only reasonable explanation for it.
I think it has already been reported and, if I'm not
wrong, an easy solution would be to sort the moves by cubeful equity
and then by cubeless equity: when all of them have cubeful equity
+1.000 (or -1.000), this should prevent odd-looking moves. Not a big deal
however, the match is already won/lost ...