Skip to content

An exploration of methods to improve the classic Artificial Intelligence algorithm minimax as used by an adversial chess playing model

License

Notifications You must be signed in to change notification settings

CordetG/AIChessPlayer

 
 

Repository files navigation

AIChessPlayer

An exploration of methods to improve the classic Artificial Intelligence algorithm Minimax as used by an adversial chess-playing agent

Install Requirements:

# Used libraries include chess, gmpy2, and numpy.
# The node class within also requires python 3.7+.
pip install -r requirements.txt

Chess

The name of the game is pawn chess. Each player starts with a row of pawns with all of the normal legal moves. If one pawn gets to the other end of the board the game is over. If there is a stalemate they tie.

This project uses the python Chess Library.

The building block of the chess library is a 64 bit integer state representation. To track all pawns for a particular player for example each 1 or 0 in the binary of the integer corresponds to a position on the board. As such, each player has an integer to track their pawns, rooks, knights, bishops, king, and queen.

Using binary representations decreases processing time.

# Binary representations where each column is full of ones
# and each 1 represents a pawn. The binary itself is
# read from the top row right to left. This has been
# adjusted to match the game for viewing purposes.
# See STATEVARS.md for more state representations.

COLUMN H: int representation: 9259542123273814144
COLUMN H: chess.board('7p/7p/7p/7p/7p/7p/7p/7p')
# [[0 0 0 0 0 0 0 1]
#  [0 0 0 0 0 0 0 1]
#  [0 0 0 0 0 0 0 1]
#  [0 0 0 0 0 0 0 1]
#  [0 0 0 0 0 0 0 1]
#  [0 0 0 0 0 0 0 1]
#  [0 0 0 0 0 0 0 1]
#  [0 0 0 0 0 0 0 1]]

Environment

The created environment contains a customized board and game class to adjust for the nuances of pawn chess. The victory condition has been replaced in addition to the move mechanics - a custom class has been used to simplify the move making process.


Movement

Movement is controlled using normal mechanics example: A2 to A4. Using our own class called "make a move" a player can move a piece from position 1 to position 2.


Victory

Victory is defined by a pawn making it to the other side. This is checked using a flag in the chess library called BB_BACKRANKS. Doing a bitwise 'and' with this and the position of all pawns for a player will return 0 if no pawn has made it to the back and a different positive integer otherwise.


Algorithms

  1. Minimax
    • Alpha Beta Pruning

Heuristics

  • Maximize Number Of Pieces
  • Keep Pawns Diagonally Supported
  • Side By Side Pawns
  • Distance From Starting Location
  • Stacked Pawns
  • Piece Could BCaptured

Tools

  • Board to Array: Convert any 64 bit int into a 1 or 2d array for binary free heuristics processing.
  • Pawn Board to Int - Take in a board and return its int representation.
  • Count 1s - Count the number of ones in a binary to give the total number of pawns on the board.

Contributors

  • Amila Ferron
  • Cordet Gula
  • Nicole Kurtz
  • Ryan Filgas

About

An exploration of methods to improve the classic Artificial Intelligence algorithm minimax as used by an adversial chess playing model

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%