Re:Deep learning for chess
by toolslive on Tuesday December 02, 2014 @04:10AM
(#8686316)
the Houdini (and I think Rybka too) evaluation function is tweaked by letting the engine play zillions of micro games against itself in a tournament. One such micro game lasts a few seconds, and each of the players has a different setting of the parameters for the evaluation function (material, position, ...). You could apply the same meta strategy here.
[ Parent
| on HN ]
Stockfish has a distributed network to allow anyone to donate computer time to test new patches. Currently running nearly 400 games/minute.