Merge pull request #7 from rapidarray1211/stockfish_eval_centipawn_loss

main measures ai v ai over 10 games and evaluates it and outputs a graph

Author: rapidarray1211

__pycache__/board.cpython-312.pyc

@@ -14,7 +14,6 @@
 #https://blog.devgenius.io/simple-interactive-chess-gui-in-python-c6d6569f7b6c
 
 engine = ChessEngine("models/movepredictorV2_24.keras", 1, 3)
-print("testing")
 
 def highlight_king_square(scrn, outcome, BOARD):
     # Find the position of the checkmated king

__pycache__/stockfish.cpython-312.pyc

@@ -1,6 +1,11 @@
+import matplotlib.pyplot as plt
 from pgn_parser import load_pgn_file, get_move_sequence
 from board import initialize_board, make_move
 from stockfish import evaluate_move_stockfish, close_engine, evaluate_game_stockfish
+from engine.chessEngine import ChessEngine
+import numpy as np
+
+engine = ChessEngine("models/movepredictorV2_24.keras", 1, 3)
 
 def play_game_pgn(pgn_file):
     # Load games from the PGN file
@@ -16,10 +21,54 @@ def play_game_pgn(pgn_file):
         for move in move_sequence:
             print(move)
             make_move(board, move)
-            evaluate_move_stockfish(board)
         evaluate_game_stockfish(move_sequence)
         close_engine()
+
+def play_ai_vs_ai(num_games=100):
+    all_accuracy_scores = []
+
+    for game_index in range(num_games):
+        print(f"Playing game {game_index + 1}/{num_games}")
+        board = initialize_board()
+        move_sequence = []
+        while not board.is_game_over():
+            move = engine.predict_best_move(board, True)
+            move_sequence.append(move.uci())
+            make_move(board, move.uci())
+        
+        accuracy_scores = evaluate_game_stockfish(move_sequence)
+        all_accuracy_scores.append(accuracy_scores)
+    
+    return all_accuracy_scores
+
+def visualize_accuracy_scores(all_accuracy_scores):
+    # Find the maximum number of moves in any game
+    max_moves = max(len(scores) for scores in all_accuracy_scores)
+    
+    # Initialize a list to hold the sum of centipawn losses for each move number
+    summed_scores = np.zeros(max_moves)
+    move_counts = np.zeros(max_moves)
+    
+    # Sum the centipawn losses for each move number across all games
+    for game_scores in all_accuracy_scores:
+        for i, score in enumerate(game_scores):
+            if score <= 8000:  # Filter out scores greater than 8000
+                capped_score = min(score, 1000)  # Cap the score at 1000
+                summed_scores[i] += capped_score
+                move_counts[i] += 1
+    
+    # Calculate the average centipawn loss for each move number
+    average_scores = summed_scores / move_counts
 
+    # Plot the average centipawn loss per move
+    plt.plot(average_scores)
+    plt.xlabel('Move Number')
+    plt.ylabel('Average Centipawn Loss')
+    plt.title('Average Centipawn Loss per Move (Capped at 1000)')
+    plt.show()
 
 if __name__ == "__main__":
-    play_game_pgn("data/lichess_dataset_1.pgn")
+    num_games = 10
+    all_accuracy_scores = play_ai_vs_ai(num_games)
+    visualize_accuracy_scores(all_accuracy_scores)
+    close_engine()

gui.py

@@ -36,7 +36,7 @@ def evaluate_game_stockfish(move_sequence):
                 board.push(move) 
 
                 evaluation = engine.analyse(board, chess.engine.Limit(depth=stockfish_depth))
-                best_move = engine.play(board, chess.engine.Limit(time=0.5)) 
+                best_move = engine.play(board, chess.engine.Limit(time=0.2)) 
                 # If increase time, it increases accuracy but decreases speed (obviously)
 
                 eval_score = evaluation['score'].relative.score(mate_score=10000)
@@ -45,7 +45,7 @@ def evaluate_game_stockfish(move_sequence):
                 best_move_score = best_move_evaluation['score'].relative.score(mate_score=10000)
                 # We compare the centipawn loss between the best move, and the move given to get a delta. Average the delta for average centipawn loss
 
-                print(f"Move: {move_str} -> Eval: {eval_score}, Best Move: {best_move_score}")
+                # print(f"Move: {move_str} -> Eval: {eval_score}, Best Move: {best_move_score}")
 
                 accuracy_score = abs(eval_score - best_move_score)
                 accuracy_scores.append(accuracy_score)
@@ -57,6 +57,7 @@ def evaluate_game_stockfish(move_sequence):
             print(f"Average centipawn loss: {average_accuracy:.2f}")
         else:
             print("No valid moves to evaluate.")
+    return accuracy_scores
 
 def generate_move(board: chess.Board): 
     result = engine.play(board, chess.engine.Limit(time=2.0))