kish 1.1.3

A high-performance Turkish Draughts (Dama) engine

kish

A high-performance Turkish Draughts (Dama) engine with Python bindings.

Built with Rust and PyO3 for maximum speed. Designed for both UI applications (simple API with algebraic notation) and machine learning (fast bitboard access).

Installation

pip install kish

Or build from source:

cd kish-py
pip install maturin
maturin develop --release

Quick Start

import kish

# Create a new game with standard starting position
board = kish.Board()
print(f"Turn: {board.turn}")  # Turn: White

# Get all legal moves
actions = board.actions()
print(f"Legal moves: {len(actions)}")  # Legal moves: 23

# Make a move (returns new board - immutable)
new_board = board.apply(actions[0])
print(f"Move: {actions[0].notation()}")  # e.g., "a3-a4"

# Check game status
status = new_board.status()
if status.is_in_progress():
    print("Game continues")
elif status.is_draw():
    print("Draw!")
else:
    print(f"Winner: {status.winner()}")

Game with History (Draw Detection)

Use Game for full game management with proper draw detection:

import kish

game = kish.Game()

# Play moves
while not game.status().is_over():
    actions = game.actions()
    if not actions:
        break

    # Make move (mutates game state)
    game.make_move(actions[0])

    # Draw detection
    if game.is_threefold_repetition():
        print("Draw by repetition!")
        break

    if game.halfmove_clock >= 50:
        print("Draw by 50-move rule!")
        break

# Undo moves
game.undo_move()
print(f"Moves played: {game.move_count}")

Custom Positions

import kish

# Create a custom position
board = kish.Board.from_squares(
    turn=kish.Team.White,
    white_squares=[kish.Square.D4, kish.Square.E3],
    black_squares=[kish.Square.D5, kish.Square.F6],
    king_squares=[kish.Square.D4],  # D4 is a king
)

# Query pieces
print(f"White pieces: {[str(s) for s in board.white_pieces()]}")
print(f"Kings: {[str(s) for s in board.kings()]}")

Machine Learning

Bitboard Access

Fast access to raw bitboard representation for neural network input:

import kish
import numpy as np

board = kish.Board()

# Get individual bitboards (u64 integers)
white = board.white_bitboard()
black = board.black_bitboard()
kings = board.kings_bitboard()

# Get all at once (most efficient)
white, black, kings, turn = board.bitboards()

# As numpy array
arr = np.array(board.to_array(), dtype=np.uint64)
# arr = [white_pieces, black_pieces, kings, turn]

# Reconstruct from bitboards
board = kish.Board.from_bitboards(turn=0, white=white, black=black, kings=kings)

Bit Plane Conversion for CNNs

import numpy as np

def to_bit_planes(board):
    """Convert board to 4x8x8 tensor for CNN input."""
    w, b, k, turn = board.bitboards()
    planes = np.zeros((4, 8, 8), dtype=np.float32)

    for i in range(64):
        row, col = i // 8, i % 8
        planes[0, row, col] = (w >> i) & 1  # white pieces
        planes[1, row, col] = (b >> i) & 1  # black pieces
        planes[2, row, col] = (k >> i) & 1  # kings
    planes[3, :, :] = turn  # turn plane

    return planes

# Faster version using numpy
def to_bit_planes_fast(board):
    """Optimized bit plane conversion."""
    w, b, k, turn = board.bitboards()
    planes = np.zeros((4, 64), dtype=np.float32)
    planes[0] = np.unpackbits(np.array([w], '>u8').view(np.uint8))[::-1]
    planes[1] = np.unpackbits(np.array([b], '>u8').view(np.uint8))[::-1]
    planes[2] = np.unpackbits(np.array([k], '>u8').view(np.uint8))[::-1]
    planes[3] = turn
    return planes.reshape(4, 8, 8)

Action Features for Policy Networks

# Get action features for ML
for action in board.actions():
    # Source and destination
    src = action.source()
    dst = action.destination()

    # Move properties
    is_capture = action.is_capture()
    is_promotion = action.is_promotion()
    capture_count = action.capture_count()

    # Captured pieces (for reward shaping)
    captured = action.captured_pieces()      # List[Square]
    captured_bb = action.captured_bitboard() # u64

    # Raw delta for applying moves (XOR with board state)
    white_delta, black_delta, kings_delta = action.delta()
    delta_arr = np.array(action.delta_array(), dtype=np.uint64)

Distance Heuristics

# Manhattan distance for evaluation functions
sq1 = kish.Square.D4
sq2 = kish.Square.H8
distance = sq1.manhattan(sq2)  # 8

# Distance to promotion row
def distance_to_promotion(square, team):
    """Distance to back row for promotion."""
    if team == kish.Team.White:
        target_row = 7  # Row 8
    else:
        target_row = 0  # Row 1
    return abs(square.row() - target_row)

Performance Testing

import kish

board = kish.Board()

# Count positions at depth (perft)
nodes = board.perft(6)
print(f"Positions at depth 6: {nodes}")  # ~450 million nodes/sec

Examples

The examples/ directory contains runnable examples:

• basic_game.py - Simple game loop
• custom_position.py - Setting up custom board positions
• ml_features.py - Extracting features for machine learning
• perft.py - Performance testing with perft
• random_playout.py - Random game simulation

API Reference

Types

Type	Description
Team	Enum: White, Black
Square	Enum: A1 through H8 (64 squares)
GameStatus	Game state with query methods
Action	Move with notation and bitboard access
Board	Immutable game board
Game	Mutable game with history tracking

Board Methods

Method	Description
Board()	Standard starting position
Board.from_squares(...)	Custom position from square lists
Board.from_bitboards(...)	Custom position from bitboards
board.actions()	Get legal moves
board.apply(action)	Make move (returns new board)
board.status()	Get game status
board.perft(depth)	Performance test

Board Bitboard Methods (ML)

Method	Description
board.white_bitboard()	White pieces as u64
board.black_bitboard()	Black pieces as u64
board.kings_bitboard()	Kings as u64
board.bitboards()	Tuple: (white, black, kings, turn)
board.to_array()	Array: [white, black, kings, turn]

Action Methods

Method	Description
action.source()	Source square
action.destination()	Destination square
action.notation()	Algebraic notation (e.g., "d4xd6")
action.is_capture()	Is this a capture?
action.is_promotion()	Does piece promote?
action.path()	Full path of squares

Action Bitboard Methods (ML)

Method	Description
action.captured_pieces()	Captured squares as list
action.captured_bitboard()	Captured pieces as u64
action.delta()	Tuple: (white_delta, black_delta, kings_delta)
action.delta_array()	Array: [white_delta, black_delta, kings_delta]

Game Methods

Method	Description
Game()	New game
Game.from_board(board)	From existing position
game.make_move(action)	Make move (mutates)
game.undo_move()	Undo last move
game.is_threefold_repetition()	Check repetition draw
game.halfmove_clock	Moves since last capture
game.move_count	Total moves made

Square Methods

Method	Description
Square.from_notation("d4")	Parse notation
Square.from_row_col(row, col)	From indices
Square.from_mask(u64)	From bitboard
square.notation()	To string (e.g., "D4")
square.row()	Row index (0-7)
square.col()	Column index (0-7)
square.to_mask()	To bitboard
square.manhattan(other)	Distance to other square

Bitboard Layout

Bit index = row * 8 + col

    A   B   C   D   E   F   G   H
  +---+---+---+---+---+---+---+---+
8 |56 |57 |58 |59 |60 |61 |62 |63 |  <- White promotes here
  +---+---+---+---+---+---+---+---+
7 |48 |49 |50 |51 |52 |53 |54 |55 |
  +---+---+---+---+---+---+---+---+
6 |40 |41 |42 |43 |44 |45 |46 |47 |
  +---+---+---+---+---+---+---+---+
5 |32 |33 |34 |35 |36 |37 |38 |39 |
  +---+---+---+---+---+---+---+---+
4 |24 |25 |26 |27 |28 |29 |30 |31 |
  +---+---+---+---+---+---+---+---+
3 |16 |17 |18 |19 |20 |21 |22 |23 |
  +---+---+---+---+---+---+---+---+
2 | 8 | 9 |10 |11 |12 |13 |14 |15 |
  +---+---+---+---+---+---+---+---+
1 | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 |  <- Black promotes here
  +---+---+---+---+---+---+---+---+
    A   B   C   D   E   F   G   H

License

Apache-2.0