def test_evaluate_graphic():
"""
test evaluate with graphical output
"""
while input('press q to quit: ') != 'q':
tmp = np.random.randint(3, size=(19, 19))
board = np.zeros((1, 19, 19, 3), dtype=int)
for y in range(19):
for x in range(19):
if tmp[y, x] == 1:
board[0, y, x, 0] = 1
elif tmp[y, x] == 2:
board[0, y, x, 1] = 1
wpos = []
for y in range(19):
for x in range(19):
if evaluate(board, 0, [y, x]):
wpos.append([y, x])
for y in range(19):
for x in range(19):
if evaluate(board, 1, [y, x]):
wpos.append([y, x])
for y in range(19):
for x in range(19):
v = board[0, y, x, 0] + board[0, y, x, 1] * 2
if [y, x] in wpos:
print('\033[92m{}\033[0m'.format(v), end='')
else:
print(v, end='')
print()
pass
def test_against_random(model, size=100):
'''
Evaluate the model against random performance
'''
wins = {0: 0, 1: 0, 2: 0}
for _ in range(size):
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:
r_board = game.get_board_raw()
x, y, _ = evaluate(model, game, pid, tau=1)
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 basic_win():
"""
test with simple env
"""
e = np.array([[0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[2, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]])
board = conv_map(e)
player = 0
pos = (0, 0)
root = Node(0)
net = Network(-1)
return evaluate(board, player, pos)
def test_winning_move_when_one():
"""
test if network find winning move among one
return true if winning move is found
return else otherwise
"""
while True:
has_winning_move = 0
tmp = np.random.randint(3, size=(19, 19))
board = np.zeros((1, 19, 19, 3), dtype=int)
for y in range(19):
for x in range(19):
if tmp[y, x] == 1:
board[0, y, x, 0] = 1
if evaluate(board, 0, [y, x]):
board[0, y, x, 0] = 0
if has_winning_move:
has_winning_move = 1
elif tmp[y, x] == 2:
board[0, y, x, 1] = 1
if evaluate(board, 1, [y, x]):
board[0, y, x, 1] = 0
if has_winning_move:
has_winning_move = 2
if not has_winning_move:
continue
network = Network(0)
node = Node(0)
expand(node, board, has_winning_move - 1, network)
turn(board, has_winning_move - 1, node, network)
for y in range(19):
for x in range(19):
if evaluate(board, has_winning_move - 1, [y, x]):
return True
return False
def human_turn(board, node, player, net):
print("Your turn, Human")
e = 0
while (not e):
try:
x = int(input("x: "))
y = int(input("y: "))
e = 1
except:
print("wrong format, only integers required")
q = input("quit? (y/n): ")
if q == "y":
exit(0)
pos = (x, y)
put_on_board(board, pos, player, 1)
node = update_turn(board, player ^ 1, node, net, pos)
_, r = evaluate(board, player, pos)
return node, r
for it in range(ITER):
for i in range(EPOCH):
print('{}: {}/{}'.format(it, i, EPOCH), end='\r')
game = TicTacToe(3)
pid = np.random.random_integers(low=1, high=2, size=1)[0]
winner = None
inputs = {1: [], 2: []}
probs = {1 : [], 2: []}
while winner is None:
board = game.get_board(pid)
r_board = game.get_board_raw()
x, y, prob = evaluate(model, game, (pid % 2) + 1)
inputs[pid].append(copy.copy(game.get_input(pid)))
probs[pid].append(prob)
game.place(pid, x, y)
winner = game.check_win()
pid = (pid % 2) + 1
if winner != 0:
one_reward = [1 if winner == 1 else -1] * len(inputs[1])
two_reward = [1 if winner == 2 else -1] * len(inputs[2])
else:
one_reward = [0] * len(inputs[1])
two_reward = [0] * len(inputs[2])
