Skip to content

Latest commit

 

History

History
55 lines (38 loc) · 1.59 KB

README.md

File metadata and controls

55 lines (38 loc) · 1.59 KB
Raw

amoeba

an UCI chess engine in D language

How to use it

Amoeba is a bare chess engine designed to communicate with a chess GUI (like xboard/winboard, arena, ...) via the UCI protocol. For example, xboard -fcp amoeba -fd installation directory -fUCI with:

  • amoeba the name of the executable
  • installation directory the directory where is the executable

About the program

When I wrote this chess engine, I tried to stay as close as possible to the published algorithms. I hope my code to be easy to read & understand.

  • algorithms used for move generation (board.d)

    • bitboard / mailbox board representation;
    • hyperbola quintessence;
    • stage move generation

  • algorithms used in search (search.d)

    • transposition table;
    • principal variation search;
    • quiescence search;
    • late move reduction (LMR);
    • in check extension;
    • null move;
    • frontier node pruning & razoring;
    • delta pruning in quiescence search;
    • internal iterative deepending (IID);
    • aspiration window;
    • iterative deepening;
    • static exchange evaluation (see) - iterative version;
    • singular move extension
    • probcut

  • algorithms used in evaluation

    • tuned weights with the "Nelder-Mead simplex method", aka amoeba (so the name of the program);
    • lazy evaluation using material + positional + tempo data;
    • full evaluation using various mobility scores & pawn structures in addition to the lazy evaluation; So, the chess knowledge is still very basic and could be improved significantly.

  • algorithm tried but discarded:

    • enhanced transposition cutoff (ETC);
    • ...