Peripheral

A chess engine in development. I'm learning Rust and engine dev as I go, so any feedback is appreciated.

Lichess

The bot runs on Lichess.org! (Currently running on a 2GHz Odroid N2+)

Rating estimations

Rating estimate of version 1.1.3:

~2055 CCRL blitz (+95 against 1.0.0)

~2069 Lichess (all time controls). Probably not representative outside of Lichess, since the bot plays on relatively weak hardware, potentially against cherry-picking bots, etc. etc.

SPRT results and ELO calculations can be found here.

Search features:

• iterative deepening
• alpha-beta negamax
• extended null-move reductions
• transposition table
  • current replacement scheme:
    1. previous PV always remains
    2. previous iteration gets replaced
    3. depth-preferred replacement
• move ordering:
  1. TT move
  2. MVV-LVA

Eval features:

• PeSTO eval

How to build

Since version 0.1.7, the uci repository has been combined with this repository, so building from source should now be trivial:

In the root directory, run cargo build --release

UCI commands

• default UCI commands (not 100% complete, but sufficient for compatibility with e.g. Cute Chess)
• help Show list of known commands
• d Print current board
• eval Static eval of position
• run Run main function of the bot (multipurpose debug command, runs run_bot() located in lib.rs)
• make [move] Make move (e.g. e2e4)
• undo Undo last move made
• probe Probe current position in the transposition table
• gen Get the TT generation of the last search
• hist Print the history of stored keys for threefold detection
• quit Quit

Benchmarks

Some Criterion benchmarks used for development can be run with cargo bench --bench bench

One statistic which I'm proud of:

perft(6)takes on average430 ms, which is~276 MNPS!(bulk counting, no hashing)

(No-bulk: 2372 ms, 50 MNPS)

Perft test suite

Run a perft test suite with cargo test --release -- --nocapture This performs perft on 32 tricky positions and compares it with expected perft results.