mayersjd/Chesster

Have AIs battle in games of chess!

Chesster

This provides a standard for a Chess AI hackathon and a graphical display of matches.

Participation

Installation

To participate you must clone this repository through git, and create a branch (off of the development branch)

git checkout development
git checkout -b YourName

Before you start coding you should ensure that Chesster runs on your system via installing it and it's dependencies.

pip install .

After that you will have the chesster command available on your system. Try it out by running:

chesster RandomAI RandomAI

This will start a single game match between two instances of the RandomAI.

To try your hand at playing against the RandomAI try running:

chesster RandomAI Human

Note that you must enter your move in UCI notation.

To see all of the options that chesster has availabe to you run:

chesster -h

Adding an AI

AIs in chesster are all a child class of BaseAi which is defined in src/chesster/ai/base.py. Creating your own would be easily accomplished by copying the implementation of RandomAi which is defined in src/chesster/ai/random.py. Note that you will need to change the name of the class witn the copied file, and a few of the documentation strings. Finally, add your new class to the AI dictionary within src/chesster/ai/__init__.py.

To test out that your new copy of the RandomAI works as intended you can do so in one of two ways:

• Reinstall Chesster with pip install ., and run the chesster command.
• Run the src/chesster/cli/__init__.py file directly
  • This will be easier to use within a fancy IDE like PyCharm. After it works as expected you'll be free to build your AI as you see fit! The requirements for your AI are simple and are in the make_move method. Firstly your AI must be able to make descions based on only two pieces of information
• board: A Board object from the chess python library
  • This object contains the information about the current board state and previous moves.
  • Hint: board.turn will tell your AI it's color
  • Hint: board.legal_moves will tell you all the possible legal moves. If your AI attempts to make an illegal move it will lose!
  • Hint: You can attempt to modify the board object but that won't actually do anything as your AI is only given a copy of the board object.
• timer: A custom Timer object within Chesster
  • Hint: There are several types of timers, but you're likely only interested in the timer.seconds_left property
  • Hint: You can attempt to stop the timer with timer.stop(), but that won't actually do anything as your AI is only given a copy of the timer object. Finally, your AI must return a single object: chess.Move. (Remember that if your AI submits an illegal move it loses).

How to Teach your AI

How you implement your AI's learning will likely be very different from everyone else, but your best start is the information in the JSON output that chesster creates of each match. By default Chesster will not save it, so be sure to specifiy it! To do so on the command line run:

chesster RandomAI RandomAI --record_file record.json

The resulting file is dense with information as it knows the board state of every move that took place in the game! The best way to understand it is to look at the MatchRecord object, specifically the to_dict method. In Python JSON object are basically just dictionaries so each object within the records module has a to_dict and from_dict method. Note that MatchRecord keeps track of the number of wins required in a match, the number of wins each color got and a list of GameRecord objects. Each of the GameRecord objects knows what AI each color was, the initial board state, the eventual GameResult and a list of each Move. From this information you should be able to teach your AI how to play chess, as well as debug what happens during a game. Something of use may be that you can specifiy the initial state of the board (in the FEN notation). This could be useful for debugging, or teaching your AI.