[Bug-gnubg] Re: Gnu bugs?

[Jim Segrave]

On Mon 03 Nov 2003 (00:08 +0000), ghjk wrote:
> Dear Jim,
> I have attached for you a new .sgf. This time is NOT
> from editing but from a real match.
> In the fourth game we have the same wrong. It wants to
> double in my first roll. There is the sign: ?!
> After Analyse / Analyse match the cube wrong is ok,
> but the ?! remains.
> Please investigate it and reply.

There's a bug. When you have tutor mode enabled, gnubg analyses the
move you have made, including checking if you could have and should
have doubled. The check for 'could have' does not include a check for
the first move of the game, so in games where you should double at the
first opportunity, you get the tutor analysis saying you made an error
in not doubling. The code which will pop up the warning window doesn't
run, because gnubg is only interested in warning you about your
chequer play at that poing, but the analysis results do have your cube
play marked as an error. Recent versions of gnubg save the results of
the tutor analysis whenever possible, so the erroneous criticism of
the cube play does show up. I'm looking for an elegant way to fix this.

> Some others questions:
> 1) When sometimes I do Analyse / Analyse match, ?GNU
> Backgammon? crashes.
>      Try the .sgf I have sent you. Sometimes it crashes, sometimes not.

It does crash for me, but this is gnubg under Linux, not Windows. I
have tried 0, 1 and 2 ply analysis, all without problems

> 2) Why after Analyse match gnubg has only 2050 ?Error based
 abs. FIBS rating?? Older versions had 2200.

Research by others, particularly Kees van Doel, showed that 2050 is a
better approximation to the correct number. It does not mean that
gnubg has become weaker, it hasn't

> 3) In a match I have played yesterday the ?Error based abs. FIBS
> rating? for me it was a? negative number. Is it possible?
> Unfortunally I have not the .sgf file.  Thank you in advance for
> your reply.

That certainly seems unreasonable. However, given a high enough
chequer error rate yes, it can come out negative. For the sake of
argument, assume your cube error rate is 0.

your fibs rating is calculated as:

2050 - (Ce * (8798 + 25526/n))

where n is the match length and Ce is your chequer error rate 0 to
perhaps .9 (although that would require some work to achieve). So, in
a 1 point match, that's

2050 - Ce * 34324. An error rate greater than 59.7 millipoints/move
will give a negative rating.

an error rate over 118 mpts/move in a 3 point match will also gie a
negative rating. 

I think we need to review this end of the calculation - most of the
work was concentrated on the top end of the score.

-- 
Jim Segrave           address@hidden