def play(agent_1, agent_2):
# initialize 3x3 tic tac toe board
board = Board(3)
while board.move_still_possible():
if board.player == 1: # X-player
move = agent_1(board)
else: # O-player
move = agent_2(board)
board.make_move(move)
# evaluate game state
if board.game_is_over():
# return winner
return board.player
# return 'game ended in a draw'
return 0
def learn_probabilities():
# Global counters for winning moves of X and O
winX = np.zeros((3, 3))
winO = np.zeros((3, 3))
# Learn through 50000 plays
for i in range(5000):
board = Board(3)
# Local counters
countX = np.zeros((3, 3))
countO = np.zeros((3, 3))
while board.move_still_possible():
move = get_random_move(board)
x, y = move
# Update corresponding local counter
if board.player == 1:
countX[x, y] += 1
else:
countO[x, y] += 1
board.make_move(move)
if board.move_was_winning_move(board.player):
winner = board.player
# Update winner's global counter
if winner == 1:
winX += countX
elif winner == -1:
winO += countO
break
# Collect statistics of winning moves of both players
win = winX + winO
# Normalize to obtain probabilities
win_normalized = preprocessing.normalize(win, norm='l2')
# Write probabilities to file
f = open('probabilities', 'w')
np.savetxt(f, win_normalized)
print "Learning finished!"
return win_normalized