[gnugo-devel] data mining in 9x9 GNU Go 3.7.4 games ?

[alain Baeckeroot]

Hi

On CGOS GNU Go 3.7.4 has played more than 11 000 games on a 9x9 board
against lots of differents bots including strong ones. This has provided
superko violation examples and stepped on some bugs.

A mail in the archive speak about statistical analysis for tuning something
(the opening ?), but i don't remember what, when, why ;)

One thing i am doing is sorting the games by opponent, and separate the rare 
strong ones (>=1500 ELO, namely Anchorman, Lazarus, Neurogo, Viking5 and
GNU Go based SluGo, twinbot, ggexp, twinbot, and CVS). These can probably used
for tuning, based on stats, so it should be quite fast and easy to do.

I am also extracting all big loss against weak bots, as generally they show
a bug or a blunder. This need to watch games, and will take longer time...

Any idea,tools, comment or volunteer would be nice :)
Alain