Re: [Bug-gnubg] Luck rate bug?

[Joern Thyssen]

On Sat, May 03, 2003 at 03:37:03PM +0200, Scott Steiner wrote
> I played and won a 7 pointer against gnu and analysed the match as
> usual.  The match statistics displayed the following (summary):
> 
>                                          gnu                  user   
> checker 
> Error rate (total)                -0,047 (-1,647%)     -0,309 (-4,289%)
> Error rate (pr. move)             -0,001 (-0,032%)     -0,009 (-0,119%)
>   
> Luck rate (total)                 +1,625 (-1,520%)     +0,987 (+54,832%)
> Luck rate (pr. move)              +0,028 (-0,026%)     +0,022 (+1,218%)
> 
> cube
> Error rate (total)                -0,033 (-0,428%)     -0,026 (-0,158%)
> Error rate (per cube decision)    -0,002 (-0,020%)     -0,004 (-0,026%)
> 
> overall
> Overall error rate (total)        -0,080 (-2,075%)     -0,335 (-4,447%)
> Overall error rate (per decision) -0,001 (-0,028%)     -0,008 (-0,106%)
> 
> MWC against current opponent        52,37%               47,63%
> 
> 
> As you can see, gnu outplayed me in both checker play and cube
> decisions.  How come I won the match with a lower luck rate then? 

Huh? You do have a higher luck rate 54.832% versus -1.520%, so you net
luck was 56.3%. According to the overall error rate you needed about
52.4% luck, so that looks fine (the small difference is due to
imperfections in the luck calculation and analysis).

> This is the first time ever with gnu that I observe that the weaker
> player wins and also has a lower luck rate, so there must be something
> wrong here. 

Nope, everything is fine!

> And also notice above that gnu has a positive luck rate in terms of
> equity points but a negative one in terms of MWC, another thing I
> found peculiar.

I think this should be an FAQ :-)

For example, at -7,-7 a 0.2 blunder (or 0.2 joker) gives you 1.2% MWC,
but the same blunder (or joker) gives you 10% MWC at DMP.

So if you roll a 0.2 joker at -7,-7 and a -0.2 anti-joker at DMP your
normalised luck rate will be 0.0, but the unnormalised one will be -8.8%
MWC.

So you can't say that the unnormalised and normalised luck rate should
have the same sign.

Jørn