finished simulated annealing

01_sudoku_hillclimber.py

@@ -34,6 +34,6 @@ while -np.sum([len(set(board[:, i])) != 9 for i in range(9)]):
                 last_fitness = current_fitness
             else:
                 board[row, rand_col], board[row, col], = board[row, col], board[row, rand_col] # swap back
-    # print(last_fitness) # debugging
+    print(last_fitness) # debugging
 
 print(board)        

02_sudoku_sa.py

@@ -7,6 +7,7 @@ erfüllen.
 
 import numpy as np
 import random
+import math
 
 board = np.array([
     [1, 2, 3, 4, 5, 6, 7, 8, 9],
@@ -21,19 +22,44 @@ board = np.array([
 ])
 
 board_size = len(board) # Board is always quadratic
-last_fitness = -np.sum([len(set(board[:, i])) != 9 for i in range(9)]) # -9
 
-while -np.sum([len(set(board[:, i])) != 9 for i in range(9)]):
+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
+
+last_fitness = calculate_fitness(board)
+T = 10
+
+while calculate_fitness(board) < 0:  # Continue until no violations
     for row in range(board_size):
         for col in range(board_size):
             # swap col in row with random other col
             rand_col = random.randrange(board_size)
             board[row, col], board[row, rand_col] = board[row, rand_col], board[row, col]
-            current_fitness = -np.sum([len(set(board[:, i])) != 9 for i in range(9)])
+            current_fitness = calculate_fitness(board)
+            
             if current_fitness >= last_fitness:
                 last_fitness = current_fitness
             else:
-                board[row, rand_col], board[row, col], = board[row, col], board[row, rand_col] # swap back
-    # print(last_fitness) # debugging
+                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[row, rand_col], board[row, col] = board[row, col], board[row, rand_col] # swap back
+            T = max(T - 0.1, 0.1)  # Decrease T more slowly and don't let it reach 0
+    
+    print(last_fitness) # debugging
 
 print(board)