Re: [Bug-gnubg] Suggestion ( Equity DiffHistogram)

[Joern Thyssen]

On Sat, May 31, 2003 at 09:03:55PM -0700, Ned Cross wrote
> 
> I'm having a hard time understanding where the equity values and "best move"
> come from, since values in the temp map do not agree with the values in my
> analysis window.
> 
> In the attached position, the temp map calls 24/13 the best move for a roll
> of 65 (2-ply), but 2-ply analysis window shows 23/18 24/18 as better by .003
> cubeful, and by .001 cubeless.  Also the 2-ply equity listed in the temp map
> is +0.693, yet 2-ply cubeful analysis gives 24/13 an equity of +0.344, and
> the cubeless equity is +0.468.
> 
> I thought for a moment GNUBG might not be normalizing the "temp map" equity
> to the cube value, but +0.693 isn't quite the same as (+0.344 * 2).
> 
> Am I mis-interpreting the numbers, or is there an issue with the way GNUBG
> populates the temp map?

You've actually discovered two bugs:

(1) the equity shown in the temp map was calculated with the wrong
    player owning the cube. 

    In your example the opponent holds the cube, but the equities were
    calculated as if you owned the cube. Hence the difference between
    +0.693 and +0.344. Apperently the cube is worth 0.349 :-)

(2) the default move filter used by the temp map had some illegal values
    for the 2-ply movefilter, which effectively meant that the moves
    were only evaluated at 0-ply. gnubg is now using the "normal" move
    filter for the temp map. Personally I think it would be overkill to
    let the user change the move filter.

Jørn