|
|
This was the first released version. I put the RCX only version on the Mindstorms site to see if anybody was interested.
It worked properly, compiled down to 24K leaving enough room for some simple robotics... and it beat me! (although I'm terrible at chess).
It was still a little bit rough around the edges: The 50 move draw rule and three move repetiton were not implemented; however it did fully support en passant, castling and pawn promotion. Undo was available but not linked into the user interface. The lookahead level could be set at the start to between 1 and 6 ply, although 5 and 6 ply were extremely slow on the RCX. The RCX could play black or white.
Thanks to the efficent negaScout algorithm, even though there is no move ordering, it took about a minute per move (mid game, less at the opening and endgame) on the RCX with two move lookahead (it deepened automatically to a maximum depth of eight on captures and checks) , and was bearable (and plays much better than the first version) on three and four ply lookahead.
The rating algorithm needed a few tweaks and there was room for improvement in the speed of the search, but I was quite proud of it.
Anyway, back to the compiler, for version 1.0....
|
|