https://github.com/boweihan/tsfish
TypeScript Chess Engine with UCI Interface
https://github.com/boweihan/tsfish
Last synced: 2 months ago
JSON representation
TypeScript Chess Engine with UCI Interface
- Host: GitHub
- URL: https://github.com/boweihan/tsfish
- Owner: boweihan
- Created: 2023-05-23T23:27:49.000Z (about 2 years ago)
- Default Branch: master
- Last Pushed: 2023-09-18T14:30:19.000Z (over 1 year ago)
- Last Synced: 2025-01-31T09:45:39.572Z (4 months ago)
- Language: TypeScript
- Size: 4.27 MB
- Stars: 2
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# TSFish
TSFish is a chess engine with a UCI interface, written with TypeScript and NodeJS, intended for use as a bot on Lichess.org.
**_2023/09/14 update - TSFish now runs on [Bun](https://bun.sh/)._**
To install bun:
`curl -fsSL https://bun.sh/install | bash # for macOS, Linux, and WSL`
To run in development mode:
`bun start`
To build:
`bun run build`
to run the build:
`bun run build:exec`
to run tests:
`bun test`
## Implementation Notes
Interface
- UCI - https://backscattering.de/chess/uci/
Multi-threading (`src/worker`)
- NodeJS supports a form of parallelism with worker_threads, with which you can spawn separate processes to execute code. This isn't traditional multi-threading in the sense that threads can't access the execution context of their parents but worker_threads can communicate to eachother by passing events.
Board representation (`src/datatypes`)
- Bitboards represented as 64 bit BigInts. Manual enforcement of 64 bit maximum. JavaScript supports bitwise operations which makes it efficient to perform operations such as checking the existence of a piece. Shifting bits on a bitboard is an efficient way to generate moves for a piece.
Game State (`src/position`)
- The Position class is responsible for game state, board tracking, and move making.
Move generation (`src/generator`)
- Move generation uses pre-generated move caches but calculates around collisions and captures at runtime.
- Move generation is done in a pseudo-legal manner (doesn't consider checks and pins at generation time) with the exception of castling, in which intermediate steps are "check"ed.
- Pseudo-legal moves are pruned for legality post-generation in the `generateMoves()` function.
Search (`src/search`)
- Search uses a simplified Alpha-Beta algorithm (https://www.chessprogramming.org/Alpha-Beta) using the NegaMax framework.
Evaluation (`src/search`)
- Evaluation uses a symmetric evaluation function in which pieces have weighted scores and the algorithm is essentially `score = sum(scores_side) - sum(scores_opposing_side)`
## PERFT Benchmarks from startpos (approximate, on my dev laptop)
09/17/2023 - optimize around king moves and pinned pieces
- Perft(1) - 0.55 ms
- Perft(2) - 1.4 ms
- Perft(3) - 22 ms
- Perft(4) - 339 ms
- Perft(5) - 10 s
06/15/2023
- Perft(1) - 2.7 ms
- Perft(2) - 13.1 ms
- Perft(3) - 165 ms
- Perft(4) - 2.73 s
- Perft(5) - 63 s (definitely some room for improvement here!)
06/08/2023
- Perft(5) - 141.894s - 155256461 (with invalid moves)
- Perft(4) - 3.144s - 3347643 (with invalid moves)
## Improvements TBD
- Magic bitboards to speed up move generation of sliding pieces
- Quiescence search addition to search
- Improved evaluation function
- Support opening books
- Support endgame tablebases