Skip to content

Latest commit

 

History

History
79 lines (68 loc) · 3.17 KB

README.md

File metadata and controls

79 lines (68 loc) · 3.17 KB

chessy

Chessy is a Rust implementation of a classical chess engine (as opposed to the recently emerging engines based on neural networks). The goal is to provide a performant chess engine that works as a standalone and later as a WebASM file to be embedded in webpages, written with clean, performant and well tested code.

Why?

I believe that Rust is an excellent language for writing a performant chess engine - easily parallelizable, allows fine control over low level details, has const functions for precalculations and is blazingly fast.

As for why, when Stockfish already exists? Mostly for fun, it has been a great learning exercise so far :) I don't really expect to beat Stockfish, but maybe I can at least beat some humans.

How does it work?

Chessy is based on an algorithm called alpha-beta pruning (like most classical chess engines). Alpha-beta pruning is a slightly improved variation of the minimax algorithm, which tries to evaluate moves for both players in tandem and tries to find the best move for the player it is currently optimizing under the strongest move that the opponent has. This resembles the technique that strong human players also use when playing chess (or Shogi, Checkers, even Tic-Tac-Toe).

To make this efficient, there are of course a lot of specific tricks used (as we easily have to evaluate ten million positions on every move or more). We currently use Bitboards, Magic Bitboards, Transposition Tables, Quiescence Search and Move Preordering together with some interesting Rust unstable features (const functions and inline assembly). In the future, I also plan on adding Opening Databases and Endgame tables.

At what level does it play?

Not fully evaluated, but it should currently be at around 1600 ELO, from tests against humans.

How to use it?

Set up Rust and cargo, clone this repository and run cargo run --release in it (debug will be too slow). Make sure you build with the nightly toolchain (you can run rustup default nightly to switch).

There is no real CLI at the moment – it will just showcase the AI fighting it out with itself. Make sure that your terminal supports unicode!

Roadmap

START: ✓ Board Representation – BitBox and MailBox ✓ Simple CLI for interactive play. Shall include: ✓ Print Board state ✓ Make Move, update state Advanced: ✓ Check legality of move ✓ Include tests

MIDDLE:

  • Create Chess Engine: ✓ Move generator & validator ✓ Move heuristic ✓ Search strategy
    • Advanced Moves ✓ castling
      • en passant ✓ promotion ✓ Move preordering
    • Lazy move generation ✓ Quiescent search ✓ Let the engine checkmate
    • Transposition table
    • Time control (via iterative deepening)
  • Refactor code: Game state (ugly details) vs engine (Clean interface)
  • Include tests
  • More fine grained heuristics
  • End game positionals
  • Improve documentation

EXTRA COOL STUFF:

  • Undo moves in play with humans
  • Protocols of games with replays
  • Logging for debugging

END:

  • Abstract engine into its own form ✓ Write Algebraic Notation Parser
  • Interfaces
    • Simple CLI
    • Simple GUI
  • Universal Chess Interface (UCI)
  • Improvements to search strategy: End tables & Opening Books
  • Include tests