Re: Réf... . : Re: [Bug-gnubg] Missed doubles problem ?

[Holger]

At 13:58 29.08.2003 +0000, Joern Thyssen wrote:
> On Fri, Aug 29, 2003 at 02:11:57PM +0200, Holger wrote
> >
> > But I can't find any double decision around DP. That's worrying me more.
>
> The logic for "aronud DP" is very crude:
>
> (analysis.c, updateStatcontext):
> if( arDouble[ OUTPUT_NODOUBLE ] >= 0.95 ) {
>   /* around too good point */
> else {
>   /* around double point */
>
> Did one of the cube decisions have a no double equity above 0.95?

Yes, roll 8 of MaX:

Cube analysis
2-ply cubeless equity  +0,908 (Money:  +0,840)
  0,852 0,149 0,002 - 0,148 0,013 0,000
Cubeful equities:
1. Double, pass         +1,000
2. Double, take         +1,180  ( +0,180)
3. No double            +0,959  ( -0,041)
Proper cube action: Double, pass

But this one is listed as around TG:

                                      gnubg                MaX 

Missed doubles around DP              0                    1 (-0,078, -0,911%)
Missed doubles around TG              0                    1 (-0,041, -0,476%)
Error rate (total)                    0                    -0,119 (-1,387%)
Error rate (per cube decision)        0                    -0,011 (-0,126%)

So it must be roll 9 of MaX that gnubg thinks is around DP:

Cube analysis
2-ply cubeless equity  +0,855 (Money:  +0,795)
  0,839 0,129 0,002 - 0,161 0,013 0,000
Cubeful equities:
1. Double, pass         +1,000
2. Double, take         +1,070  ( +0,070)
3. No double            +0,922  ( -0,078)
Proper cube action: Double, pass

I can't really explain this.

> It has previously been suggested to improve the logic by calculating the
> double point (DP) and the too good point (TG), and call it "around TG
> point" if the gwc is larger than DP + (TG - DP)/2 (assuming TG>DP).

Yes, that's what I'd thought gnubg is doing right now. (Can there be a case 
with TG<DP? Equal yes.)

Regards,

        Holger