MLE/02_simulated_annealing/sudoku_sa.py

67 lines
2.2 KiB
Python

"""
Implementiere einen Hill-Climbing-Algorithmus, der die Zahlen in
einem SUDOKU-Feld durch vertauschen innerhalb einer Zeile so
umsortiert, dass sie die SUDOKU-Bedingung auch für die Spalten
erfüllen.
"""
import numpy as np
import random
import math
def calculate_fitness(board):
# Previous fitness
column_violations = np.sum([len(set(board[:, i])) != 9 for i in range(9)])
# plus checking 3x3 sub-grids
grid_violations = 0
for block_row in range(0, 9, 3):
for block_col in range(0, 9, 3):
# Extract the 3x3 block
block = board[block_row:block_row + 3, block_col:block_col + 3].flatten()
if len(set(block)) != 9:
grid_violations += 1
return -(column_violations + grid_violations) # Negative because we want to maximize
board = np.array([
[1, 2, 3, 4, 5, 6, 7, 8, 9],
[1, 2, 3, 4, 5, 6, 7, 8, 9],
[1, 2, 3, 4, 5, 6, 7, 8, 9],
[1, 2, 3, 4, 5, 6, 7, 8, 9],
[1, 2, 3, 4, 5, 6, 7, 8, 9],
[1, 2, 3, 4, 5, 6, 7, 8, 9],
[1, 2, 3, 4, 5, 6, 7, 8, 9],
[1, 2, 3, 4, 5, 6, 7, 8, 9],
[1, 2, 3, 4, 5, 6, 7, 8, 9]
])
board_size = len(board) # Board is always quadratic
last_fitness = calculate_fitness(board)
T = 100
print("Working...")
while calculate_fitness(board) < 0:
# swap col in row with random other col
rand_row = random.randrange(board_size)
rand_col = random.randrange(board_size)
swap_col = (rand_col + random.randrange(1, board_size)) % board_size
board[rand_row, rand_col], board[rand_row, swap_col] = board[rand_row, swap_col], board[rand_row, rand_col]
current_fitness = calculate_fitness(board)
if current_fitness >= last_fitness:
last_fitness = current_fitness
else:
p = math.e ** (-(last_fitness - current_fitness) / T) # adjusted formula
print(p) # debugging
if p > random.random(): # if probability occurs
last_fitness = current_fitness
else:
board[rand_row, swap_col], board[rand_row, rand_col] = board[rand_row, rand_col], board[rand_row, swap_col]
T = max(T - 0.01, 0.1) # Decrease T more slowly and don't let it reach 0
# print(last_fitness) # debugging
print(board)