Create a computer program to play correspondence chess, perhaps a day per move. This is to simulate programs running on next generation hardware, and to stimulate the development of next generation programs. Play such computer programs against each other in a tournament.
I predict (Zobrist) hash collisions (transposition table) will be noticeably detrimental. I also predict the UCT algorithm (now used for Go 囲碁) will be effective.
Running such tournaments will be extremely computationally expensive (perhaps there could be better uses for compute cycles?), so utilize donated computer cycles in a distributed computing project. We need a means to execute a deterministic amount of chess computation no matter how fast or slow the host computer. Moves are not made on the real-time clock. Ask program authors to calibrate their programs to meet certain time controls on average.
No comments :
Post a Comment