RustyKnight is a Rust-based chess engine designed for speed and clarity, implementing solid evaluation heuristics and advanced search techniques. This README provides a comprehensive list of all evaluation terms and search methods used in the engine.
The evaluation function determines the quality of a position using the following terms and heuristics:
- Calculates the value of all pieces for both sides.
- Uses a material hash table to cache computed material values.
- Special score for bishop pairs (+30 for two bishops).
- Evaluates piece activity and placement using middle-game and end-game tables.
- The final piece square value is a blend based on the phase of the game.
- Checkmate: Assigns a large positive/negative score depending on the side to move.
- Stalemate: Assigns a contempt value (draw score).
The engine assesses rook positioning and activity with the following heuristics:
- Rooks on Open Files: A rook on a file with no pawns is valued highly (+25 for each side).
- Rooks on Semi-Open Files: A rook on a file with only friendly pawns is valuable (+15 for each side).
- Rooks on the Seventh Rank: A rook on the opponent's second to last rank is a powerful piece, often controlling key squares and attacking pawns (+20 for white on the 7th rank, +20 for black on the 2nd rank).
Analyzes pawn structure to detect weaknesses and strengths:
- Doubled Pawns: Penalized by -20 per extra pawn on the same file.
- Isolated Pawns: Penalized by -15 if there are no adjacent pawns on neighboring files.
- Connected Pawns: Bonus of +3 for pawns supporting each other diagonally.
- Passed Pawns: Bonuses increase with rank advancement: Ranks 2–6: +5 to +80 (based on proximity to promotion)
- Pawn Chains: Additional +5 bonus for pawns protected from behind by another pawn.
- Backward Pawns: Appropiate penalty is given for backward pawns.
RustyKnight implements several search algorithms and pruning techniques:
- Main recursive search algorithm using the negamax variant of minimax.
- Integrates alpha-beta pruning for efficiency.
- Supports Principal Variation (PV) search.
- Extends search at leaf nodes by examining only tactical moves (captures, promotions) to avoid the horizon effect.
- Searches incrementally deeper, using results from shallower searches to improve move ordering.
- Hash Move: Move recommended by the transposition table.
- Killer Moves: Quiet moves causing cutoffs in previous searches at the same depth.
- History Heuristic: Tracks which moves historically perform well.
- Used to sort captures using the most valuable victim and least valuable attacker.
- Memoizes positions already evaluated to prevent redundant computation and speed up search.
- Futility Pruning: Skips moves unlikely to improve alpha.
- Late Move Pruning (LMP): Prunes quiet moves deeper into the move list.
- Late Move Reductions (LMR): Reduces search depth for less promising moves.
- Null Move Pruning: Searches positions where the side to move passes (makes a "null" move) to detect threats.
- Internal Iterative Deepening (IID): Used to find a good move when hash move is unavailable.
- UCI Protocol Support: Communicates with chess GUIs via the Universal Chess Interface.
- Zobrist Hashing: Efficient position hashing for transposition tables and caching.
All main logic is implemented in src/main.rs.
For details, refer to function definitions such as evaluate, negamax, order_moves, quiesce, evaluate_pawn_structure, and iterative_deepening.
Contributions are welcome! Please open issues or pull requests to suggest improvements, new evaluation terms, or advanced search features.
This project is open source and available under the GPL 3.0 License.
![@google-labs-jules[bot]](https://avatars.githubusercontent.com/in/842251?s=64&v=4){kind=small}