Re: [Bug-gnubg] 0-ply cubeful equities

[Joern Thyssen]

On Tue, Jun 10, 2003 at 05:12:25AM -0400, address@hidden wrote
> They look much better.  I noticed 0-ply doubles medium races too soon.
> Does the race net simply play the checker properly or should I
> generate a bearoff database.  

No, the race net is almost as good as bearoff databases, so that's
probably a waste of time.

> Can 0-ply access some other input to cube these properly?

A 0-ply cube decision has no way to calculate market losers or the
volatility of a position, so put very simple: it doubles when it's
Y% into the market window. 

Y is related to the cube efficiency X used in Janowski's formula. For
races we calculate X as

#define RACEFACTOR 0.00125
#define RACECOEFF 0.55
#define RACE_MAX_EFF 0.7
#define RACE_MIN_EFF 0.6

      rEff = anPips[1]*RACEFACTOR + RACECOEFF;
      if (rEff > RACE_MAX_EFF)
        return RACE_MAX_EFF;
      else
       {
        if (rEff < RACE_MIN_EFF)
          return RACE_MIN_EFF;
        else
          return rEff;
       }
     }

so X is between 0.7 and 0.6 depending on the length of the race. When I
have some spare time I can calculate what Y that corresponds to.

Jørn