def debug_run(model):
'''
Shows the board and value for each step in a game
'''
game = TicTacToe(3)
pid = np.random.random_integers(low=1, high=2, size=1)[0]
winner = None
while winner is None:
board = game.get_board(pid)
val = model.evaluate(game.get_input(pid))
print(board)
print(val)
x, y = q_select(pid, board, model, game)
game.place(pid, x, y)
winner = game.check_win()
pid = (pid % 2) + 1
print(
model.sess.run(model.probs,
feed_dict={model.states: game.get_input(pid)})[0])
def test_against_random(model):
'''
Evaluate the model against random performance
'''
wins = {0: 0, 1: 0, 2: 0}
for _ in range(1000):
game = TicTacToe(3)
pid = np.random.random_integers(low=1, high=2, size=1)[0]
winner = None
while winner is None:
board = game.get_board(pid)
if pid == 1:
x, y = random_choice(board)
else:
x, y = q_select(pid, board, model, game)
game.place(pid, x, y)
winner = game.check_win()
pid = (pid % 2) + 1
wins[winner] += 1
print('Wins: %d Ties: %d Losses: %d' % (wins[2], wins[0], wins[1]))
return (wins[2] / (wins[0] + wins[1] + wins[2]))
def play_user(model):
'''
Test the model against human skill level
'''
game = TicTacToe(3)
pid = np.random.random_integers(low=1, high=2, size=1)[0]
winner = None
while winner is None:
board = game.get_board(pid)
print(board)
if pid == 2:
x, y, prob = evaluate(model, game, pid, tau=.1)
print(prob)
print(model.evaluate(game.get_input(pid)))
else:
x = int(input('x: '))
y = int(input('y: '))
game.place(pid, x, y)
winner = game.check_win()
pid = (pid % 2) + 1
print(game.get_input(1))
def chess_worker(connection, args):
while True:
game = TicTacToe()
rewards = []
pid = 1
boards = [copy.deepcopy(game)]
i = 0
while game.check_win() is None:
if random.random() > args.gamma:
connection.send({'type': 'board', 'boards': boards[-8:]})
move = connection.recv()
else:
move = random.choice(game.legal_moves)
game = game.push(move)
boards.append(copy.deepcopy(game))
pid = (pid % 2 + 1)
i += 1
res = game.check_win()
size = len(boards) - 1
if res != 0:
rewards = [1 if (i % 2) == 0 else 0 for i in range(size)]
rewards = rewards[::-1]
else:
rewards = [0] * (size)
rewards = np.expand_dims(np.array(rewards), 1)
in_boards, in_targets = build_input(boards, rewards, args.history)
connection.send({
'type': 'data',
'inp': in_boards,
'rewards': in_targets
})
connection.send({'type': 'end'})
def debug_run(model):
'''
Shows the board and value for each step in a game
'''
game = TicTacToe()
pid = np.random.random_integers(low=1, high=2, size=1)[0]
winner = None
while winner is None:
board = game.get_board(pid)
val = model.evaluate(board.reshape(1, 3, 3))
print(board)
print(val)
x, y = q_select(board, model)
game.place(pid, x, y)
winner = game.check_win()
pid = (pid % 2) + 1
def play_user(model):
'''
Test the model against human skill level
'''
game = TicTacToe()
pid = np.random.random_integers(low=1, high=2, size=1)[0]
winner = None
while winner is None:
board = game.get_board(pid)
val = model.evaluate(board.reshape(1, 3, 3))
print(board)
if pid == 2:
x, y = q_select(pid, board, model, game)
else:
x = int(input('x: '))
y = int(input('y: '))
game.place(pid, x, y)
winner = game.check_win()
pid = (pid % 2) + 1
pid = 1
opp_pid = 2
winner = None
while winner is None:
board = game.get_board(pid)
if random.random() < gamma:
x, y = random_choice(board)
else:
x, y = q_select(board, model)
game.place(pid, x, y)
board = game.get_board(pid)
boards[pid].append(board)
winner = game.check_win()
pid = (pid % 2) + 1
board = game.get_board(pid)
boards[pid].append(board)
if winner != 0:
loser = (winner % 2) + 1
#winner_rewards = decay_reward(1, len(boards[winner]))
#loser_rewards = decay_reward(-1, len(boards[loser]))
winner_rewards = [1] * len(boards[winner])
loser_rewards = [-1] * len(boards[loser])
rewards = np.array(winner_rewards + loser_rewards)