/pɪ'dæntɪk/ - [pe·dan·tic]
pedantic
NOTE: This version of Pedantic has been superceded with a rewritten/refactored version found here.
Pedantic is a chess engine designed to play standard chess via one of the many UCI chess clients (i.e. Arena, BanksiaGUI or Cute Chess for instance.) Any client, GUI, analysis tool designed to work with engines that support the UCI protocol should be able to work with Pedantic. As such, Pedantic is not designed to be used alone to play chess, but it does support some command-line functionality (more on that later.)
Pedantic is my first step in advancing my knowledge of computer chess to the point of competing against some of the strongest chess engines int he world. I chose C# as the development language due to my familiarity and it's usefulness over C++ to flesh out my initial ideas. When I have fully explored the latest and cutting edge techniques in computer chess, I hope to eventually port Pedantic to a more performant language such as C++ or perhaps Rust. Throughout my journey of developing Pedantic, I am often reminded of basketball legend, Michael Jordan's quote, "you reach, I teach." However, in this case as I reach to explore the chess programming body-of-work, Pedantic teaches me along every captivating step.
- Windows 11
- Microsoft Visual Studio Community 2022
- C# / .NET 7.0
- Multi-threaded, command-line, console application
- Handcrafted evaluation with nearly 2000 tapered weights
- UCI protocol for interoperability with chess clients
- Opening book library in Polyglot binary format
- Standard Chess
- Perft Testing
- Label Training Data
- Learning (Logistical Regression)
- Multi-threaded search
- Principal Variation/Negamax Search
- Quiesence Search
- Aspiration Window
- Iterative Deepening
- Null Move Pruning
- Razoring
- Futility Pruning
- Late Move Reduction
- Check Extension
- Transposition Tables (search, evaluation, pawn structure)
- Pondering
- Magic Bitboard Move Generation
- Lazy Evaluation
- Static Exchange Evaluation (SEE)
- Transform PGN Files into Labeled Training Data
- Random Sample Selection from Training Data (currently 75 million positions and growing)
- Evaluation weight tuning using Texel inspired local search
- Logistic Regression using Local Search
- Performance Enhanced Regression (16X faster than simple Texel)
- Tuned weights automatically written to weights file (pedantic.json)
Description:
The pedantic chess engine.
Usage:
Pedantic [command] [options]
Options:
--version Show version information
-?, -h, --help Show help and usage information
Commands:
uci Start the pedantic application in UCI mode.
perft Run a standard Perft test.
label Pre-process and label PGN data.
learn Optimize evaluation function using training data.
weights Display the default weights used by evaluation.
Description:
Start the pedantic application in UCI mode.
Usage:
Pedantic uci [options]
Options:
--input <input> Specify a file to read UCI commands from.
--error <error> Output errors to specified file.
--random If specified, adds small random amounts to positional evaluations. [default: false]
-?, -h, --help Show help and usage information
Description:
Run a standard Perft test.
Usage:
Pedantic perft [options]
Options:
--type <Average|Details|Normal> Specifies the perft variant to execute. [default: Normal]
--depth <depth> Specifies the maximum depth for the perft test. [default: 6]
--fen <fen> Specifies the starting position if other than the default.
-?, -h, --help Show help and usage information
Description:
Pre-process and label PGN data.
Usage:
Pedantic label [options]
Options:
--pgn <pgn> Specifies a PGN input file.
--data <data> Specifies the name of the comma-delimited, output file.
--maxpos <maxpos> Specify the maximum positions to output. [default: 8000000]
-?, -h, --help Show help and usage information
Description:
Optimize evaluation function using training data.
Usage:
Pedantic learn [options]
Options:
--data <data> The name of the labeled data file.
--sample <sample> Specify the number of samples to use from the training data. [default: -1]
--iter <iter> Specify the maximum number of iterations before a solution is declared. [default: 5000]
--save If specified the sample will be saved in file. [default: false]
--reset Reset all starting weights (except for piece material and phase) to zero. [default: false]
--maxtime <maxtime> Maximum duration the optimization will run before a solution is declared.
--error <error> Error threshold for terminating optimization loop. [default: 0]
-?, -h, --help Show help and usage information
Description:
Display the evaluation weights.
Usage:
Pedantic weights [options]
Options:
-?, -h, --help Show help and usage information
Creating training data involves reading in a specially prepared PGN file that contains 1 or more games. Typically, this could include millions of games. To prepare a PGN for labeling it must first be processed by command-line utility pgn-extract in such a manner as to clean up incorrect file formatting and to convert the PGN movetext from SAN format to the alebraic notation specified by UCI. And example of pgn-extract usage is shown below:
pgn-extract -Wuci -oclean.pgn -D -s -C -N -V -tstartpos.txt --fixtagstrings -w128 --minmoves 8 --nobadresults example.pgn
After processing the output created by pgn-extract with Pedantic's label command, a comma-delimited file will be created that looks like the following:
Hash,Ply,GamePly,FEN,Result
7AD41D2CC8E1829E,8,135,rnbqk2r/ppp1ppbp/3p1np1/8/3PPP2/2N5/PPP3PP/R1BQKBNR w KQkq - 1 5,0.5
A1B0F1DAA007E3A3,9,135,rnbqk2r/ppp1ppbp/3p1np1/8/3PPP2/2N2N2/PPP3PP/R1BQKB1R b KQkq - 2 5,0.5
AB181C6D755E0A48,10,135,rnbq1rk1/ppp1ppbp/3p1np1/8/3PPP2/2N2N2/PPP3PP/R1BQKB1R w KQ - 3 6,0.5
FE15BDCF4CDEC83B,11,135,rnbq1rk1/ppp1ppbp/3p1np1/8/3PPP2/2NB1N2/PPP3PP/R1BQK2R b KQ - 4 6,0.5
This training data file can then be used to train Pedantic.
While my first attempt at developing a chess engine took place over 40 years ago in the early 1980s, it was never something I was proud of except as being the motivation to learn C and to sweeten my resume. It was buggy, slow, and only with a lot of luck and a bit of prayer could it successfully complete a game without crashing. It did, however, inspire my interest in computer chess, artificial intelligence and a bet made in 1968 by International Chess Master David Levy that no chess computer would be able to defeat him in a chess match over the next ten years. To make a long story short, Mr. Levy won his bet by defeating all challengers ending with his win over CHESS 4.7 in 1978. However, like with all machines the advancement of computer chess has been relentless. Computer chess programs started beating top level players by the end of of the 1980s and in 1997, a program by IBM named Deep Blue defeated the world champion and Grandmaster, Garry Kasparov. Today, computer chess programs are no longer measured against human competitors because they are simply too strong. Instead, software developers continue to advanced the science by competing, program against program to achieve the acclaim of strongest chess engine in the world. With Pedantic, I humbly endeavor to join this elite group of developers by learning all the algorithms, heuristics and minutia so that I can create my own strong chess engine.
- Chess Programming Wiki
Inspired everything and continues to inspire!
- Minimal Chess
Inspired UCI protocol, time control, transposition table aging and engine organization. - CPW Engine
Inspired implementatin of search, razoring and quiesence. - Crafty
Inspired late move reduction and null move pruning technique.
Caplinger, Noah. Chess Algorithms. USA: Lulu Press, 2021. Jordan, Bill. Advanced Chess Programming. Orlando: Independently Published, 2022. Jordan, Bill. How to Write a Bitboard Chess Engine. Orlando: Independently Published: 2020. Jordan, Bill. Teach a Chess Program Strategy. Orlando: Independently Published: 2022. Rustad-Elliotr, Rhys. Fast Chess Move Generation With Magic Bitboards. 2017. Levy, David. Computer Gamesmanship. London: Century Publishing, 1983. Levy, David. The Joy of Computer Chess. Englewood Cliffs: Prentice-Hall, 1984. Levy, David and Monty NewBorn. How Computers Play Chess. New York and Tokyo: Ishi Press International, 2009. Slate, David and Larry Atkin. "Chess 4.5: The Northwestern University Chess Program." Computer Chess Compendium. London: B. T. Batsford, 1988.
