Re: [gnugo-devel] Inaccurate estimate_score

[Darren Cook]

> I don't think you can hope for consistently reliable ranges, but this
> may very well be a bug anyway. The algorithm for estimate_score was
> changed in 3.7.2. Can you try it again with 3.7.2 or 3.7.3 (or current
> CVS) and report the results?

Here are the results using 3.7.3; I've repeated the 3.6 results below.

= W+60.5   [correct]

= W+61.1 (upper bound: 61.1, lower: 61.1)

= 59.6  [better: should be 60.5]

= 62.1  [better: should be 61.5]

experimental_score's range is now correct, so may be useful to me; I'll
try it on some more positions.

estimate_score is still strange: surely upper and lower bounds should
only be equal when the only moves left are filling dame? Or does
upper/lower bound only take into account unsettled groups? If so
clarifying the docs would be useful.

Darren





[1]:
showboard
final_score
estimate_score
experimental_score black
experimental_score white


[2]:
=
   A B C D E F G H J
 9 . X X . . . . . . 9
 8 X . X O O O . . . 8
 7 X X X O . O + . . 7
 6 . . X O . . O . . 6
 5 O O O . + . . O O 5
 4 . . X O O . . . . 4
 3 X X X O . O O O . 3
 2 . . . O . O . . O 2     WHITE (O) has captured 12 stones
 1 O O O O . . . . . 1     BLACK (X) has captured 1 stones
   A B C D E F G H J

= W+60.5

= W+59.5 (upper bound: 59.5, lower: 59.5)

= 58.0

= 60.5