Renegade is a free and open source chess engine. ♞
It is written in C++ from scratch, and values readability and simplicity. As the engine communicates through the UCI protocol, it can be connected to almost all chess user interfaces.
The engine is fairly strong, and regularly competes on Lichess over at https://lichess.org/@/RenegadeEngine, as well as in tournaments organized by the engine testing community.
- Uses bitboards to represent the location of the pieces on the board, which allows for fast and efficient operations
- A redundant mailbox is used to query what piece is on a given square, and a board hash vector is kept around for repetition checking
- Move generation relies on a number of precalculated tables, including plain magic bitboards for sliding pieces
- The engine uses a fairly standard fail-soft alpha-beta pruning framework with iterative deepening and principal variation search
- A number of move ordering and pruning methods are implemented to make search more efficient (see
Search.cpp)
- Renegade makes use of modern NNUE (efficiently updatable neural network) technology for accurate position evaluation
- Its neural network was trained entirely on self-play data, amounting to over 2.7 billion positions
- The network architecture is a (768->1024)x2->1 perspective net with SCReLU activation
Renegade - like most chess engines - is a command line application implementing the UCI protocol. It should be used alongside a graphical user interface, such as Cute Chess.
As one might expect, Renegade can also be interacted with using the terminal. Most UCI commands are supported including limiting search to a specific depth, time limit or node count.
Scores are scaled according to the estimated outcome of the game, an evaluation of 100 centipawns represents a 50% likelihood of winning.
If the engine doesn't receive the uci command, it defaults to pretty print outputs (if the compiler supports it):
Renegade chess engine 1.1.0 [Jun 26 2024 20:50:06]
go depth 7
1/ 1 20 0ms 156knps h= 0.0% -> +0.27 29-54-17 [d2d4]
2/ 4 110 0ms 378knps h= 0.0% -> +0.34 31-54-16 [d2d4 d7d5]
3/ 3 265 0ms 572knps h= 0.0% -> +0.22 27-55-18 [e2e4 e7e5 g1f3]
4/ 6 685 0ms 732knps h= 0.0% -> +0.44 33-53-14 [e2e4 e7e5 g1f3 b8c6]
5/ 7 1473 1ms 952knps h= 0.0% -> +0.25 28-54-18 [d2d4 e7e6 g1f3 g8f6 a2a3]
6/ 8 2113 1ms 1089knps h= 0.0% -> +0.22 27-55-18 [d2d4 e7e6 g1f3 g8f6 a2a3 (+1)]
7/11 5495 3ms 1607knps h= 0.0% -> +0.20 27-55-18 [g1f3 d7d5 g2g3 g7g6 c2c4 (+2)]
bestmove g1f3
Some useful custom commands are also implemented, such as eval, draw and fen.
The recommended compiler is Clang 16. It should be noted that the engine makes use of modern hardware instructions for the best possible performance.
Windows and Linux executables can be found for each release.
Getting this far would not have been possible without the members of the Engine Programming Discord, and the decades of prior research done into chess programming.
In particular, Renegade makes use of many ideas from Akimbo, Ethereal, Stockfish, Stormphrax, Viridithas and Wahoo. These are fantastic engines, and you are highly encouraged to check them out!
The neural networks were trained with bullet, and win-draw-loss models have been calculated using Stockfish's WDL tool.
Additionally, I would also like to thank everyone who gave me feedback, and those who spent the time trying out and testing the engine!