On 09/26/10 22:53, Yann Dirson wrote:
Well, let's get serious and hunt. Looks like -t would help to
understand, so I try to get some useful info using it.
./regress.sh . nngs4.tst
480 unexpected FAIL: Correct 'R16', got 'Q17'
./interface/gnugo -t -l
regression/games/nngs/evand-gnugo-3.3.17-200303201727.sgf -L 6
Strange, the first -t run for this one on amd64 decides for K4 (which
in all other runs ranks 2nd) - unfortunately my history is too short
to show traces for it :}
Typical 1st lines of output of amd64 are:
Game status = 0.18 (0.0 = start, 1.0 = game over)
Fuseki database move at R16 with relative weight 16, pattern F-H4-309+3
Fuseki database move at K4 with relative weight 2, pattern F-H4-310+0
Fuseki Player suggests R16 with value 75
On gta02 there are no fuseki lines at all.
I won't have much time to look in more details this week - if you have
a precise idea about what to look at that will surely help.
Oh, there's no mystery here. The warnings about the fuseki databases
were real; they were misbuilt and effectively disabled. I'm sure that
if you rebuild those with fuseki*.c generated on a 32-bit platform,
those testcases will pass just fine.
/Gunnar
_______________________________________________
gnugo-devel mailing list
gnugo-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/gnugo-devel
