Open source chess engines
I’ve been trying to train 2 of the open source chess engines that has self-learning capabilities, namely crafty and phalanx. And of course, I won’t miss GNU chess. I did something stupid — use a script to simulate chess tournament between various chess engines via xboard, and simply sit there, watching engines play themselves, enjoying my cup of green tea. A few comments:
- Crafty is all-around winner. Everybody knows that — no open source chess engine ever claimed it has significant chance to win Crafty.
- Phalanx, despite the fact that its last open source release is at Jan 2000, is the most tacticful out of all 3. Sometimes it evaluates piece position even more than real material. For a few moves it did surprise me. However, this tactical approach is very fatal during endgames. I’ve seen it lose under winning position! During endgames, when material difference is vital, it happily discards pawns for nothing.
- GNU chess, well, isn’t playing so badly. On average it is just losing a little bit to phalanx. However, I don’t think it is nice enough as well. Both crafty and phalanx can resign and do self-learning years ago, but gnuchess can’t even manage that right now! Let’s curse it!
Probably I’ll try packaging Sjeng Free later for Mandriva, and compare it with the other 3. And of course, scid package also needs tidying.
Disclaimer: I sucks at playing international chess, and is only a little bit better at Chinese chess.