def test_game_end(self):
game = ZertzGame(1)
self.assertEqual(game.get_game_ended(), 0)
game.get_next_state((2, 0, 1), 'PUT')
self.assertEqual(game.get_game_ended(), 1)
game = ZertzGame(7)
self.assertEqual(game.get_game_ended(), 0)
game.board.state[11] += 4
game.board.state[5] -= 4
self.assertEqual(game.get_game_ended(), -1)
def test_take_422(self):
game = ZertzGame(19)
game.get_next_state((0, 11, 5), 'PUT')
#print('')
#print(np.sum(game.board.state[:2], axis=0))
game.get_next_state((0, 8, 24), 'PUT')
#print(np.sum(game.board.state[:2], axis=0))
game.get_next_state((0, 12, 22), 'PUT')
#print(np.sum(game.board.state[:2], axis=0))
game.get_next_state((4, 2, 1), 'CAP')
def test_init(self):
game = ZertzGame(19)
self.assertEqual(game.initial_rings, 19)
self.assertEqual(np.sum(game.board.state), 619)
self.assertEqual(game.board.get_cur_player(), 0)
game = ZertzGame(19, {'w': 2, 'g': 2, 'b': 2})
self.assertEqual(game.initial_rings, 19)
self.assertEqual(game.board.state[5, 0, 0], 2)
self.assertEqual(game.board.state[6, 0, 0], 2)
self.assertEqual(game.board.state[7, 0, 0], 2)
self.assertEqual(np.sum(game.board.state), 169)
def test_str_to_action(self):
game = ZertzGame(19)
action_strs = [
'PUT w A3 C5', 'PUT g A2 B4', 'PUT b A1 A3', 'PUT g C3',
'PUT g E1 D1'
]
actions = [('PUT', (0, 0, 2)), ('PUT', (1, 5, 1)), ('PUT', (2, 10, 0)),
('PUT', (1, 12, 25)), ('PUT', (1, 24, 23))]
for action, action_str in zip(actions, action_strs):
self.assertEqual(action, game.str_to_action(action_str))
action_strs = [
'CAP b C4 g C2', 'CAP w C1 b C3', 'CAP g C1 b A1', 'CAP b D4 w B3',
'CAP w A3 w C3', 'CAP b C2 w E2'
]
actions = [('CAP', (0, 1, 2)), ('CAP', (3, 4, 2)), ('CAP', (2, 4, 2)),
('CAP', (1, 1, 3)), ('CAP', (5, 0, 0)), ('CAP', (4, 3, 2))]
for action, action_str in zip(actions, action_strs):
self.assertEqual(action, game.str_to_action(action_str))
def test_str_to_action(self):
game = ZertzGame(19)
action_strs = [
'PUT w A3 C5', 'PUT g A2 B4', 'PUT b A1 A3', 'PUT g C3',
'PUT g E1 D1'
]
actions = [('PUT', (0, 0, 2)), ('PUT', (1, 5, 1)), ('PUT', (2, 10, 0)),
('PUT', (1, 12, 25)), ('PUT', (1, 24, 23))]
for action, action_str in zip(actions, action_strs):
self.assertEqual(action_str, game.action_to_str('PUT', action[1]))
game.get_next_state((2, 7, 25), 'PUT')
game.get_next_state((1, 12, 25), 'PUT')
# Only first capture action has actual marbels to check the marble types against.
# The other will default to 'w' marbles because they are blank rings.
action_strs = [
'CAP b C4 g C2', 'CAP w C1 w C3', 'CAP w C1 w A1', 'CAP w D4 b B3',
'CAP w A3 w C3', 'CAP w C2 w E2'
]
actions = [('CAP', (0, 1, 2)), ('CAP', (3, 4, 2)), ('CAP', (2, 4, 2)),
('CAP', (1, 1, 3)), ('CAP', (5, 0, 0)), ('CAP', (4, 3, 2))]
for action, action_str in zip(actions, action_strs):
self.assertEqual(action_str, game.action_to_str('CAP', action[1]))
def test_take_actions(self):
game = ZertzGame(19)
#(('PUT', 'w', (4, 4)), ('REM', (4, 3)))
game.get_next_state((0, 24, 23), 'PUT')
#(('PUT', 'b', (3, 4)), ('REM', (4, 2)))
game.get_next_state((2, 19, 22), 'PUT')
#(('PUT', 'g', (2, 3)), ('REM', (1, 3)))
game.get_next_state((1, 13, 8), 'PUT')
#(('PUT', 'b', (1, 1)), ('REM', (3, 1)))
game.get_next_state((2, 6, 16), 'PUT')
#(('PUT', 'b', (2, 1)), ('REM', (0, 2)))
state, player_value = game.get_next_state((2, 11, 2), 'PUT')
self.assertTrue(np.all(state[5:8, 0, 0] == [5, 7, 7]))
self.assertEqual(np.sum(state), 519)
self.assertEqual(player_value, -1)
#(('CAP', 'w', (4, 4)), ('b', (2, 4)), ('g', (2, 2)), ('b', (2, 0)))
state, player_value = game.get_next_state((3, 4, 4), 'CAP')
self.assertEqual(player_value, -1)
state, player_value = game.get_next_state((1, 2, 4), 'CAP')
self.assertEqual(player_value, -1)
state, player_value = game.get_next_state((1, 2, 2), 'CAP')
self.assertTrue(np.all(state[11:14, 0, 0] == [0, 1, 2]))
self.assertEqual(np.sum(state), 566)
self.assertTrue(np.all(state[:4, 2, 0] == [1, 1, 0, 0]))
self.assertTrue(np.all(state[:4, 4, 4] == [1, 0, 0, 0]))
self.assertTrue(np.all(state[:4, 2, 3] == [1, 0, 0, 0]))
self.assertTrue(np.all(state[:4, 1, 1] == [1, 0, 0, 1]))
self.assertEqual(np.sum(state[game.board._CAPTURE_LAYER]), 0)
self.assertEqual(player_value, 1)
def test_get_actions(self):
game = ZertzGame(19)
placement, capture = game.get_valid_actions()
self.assertEqual(np.sum(placement), 648)
self.assertTrue(np.all(capture == False))
game = ZertzGame(37)
placement, capture = game.get_valid_actions()
self.assertEqual(np.sum(placement), 1944)
self.assertTrue(np.all(capture == False))
#(('PUT', 'w', (4, 4)), ('REM', (4, 3)))
state = game.get_next_state((0, 32, 31), 'PUT')
#(('PUT', 'b', (3, 4)), ('REM', (4, 2)))
game.get_next_state((2, 25, 30), 'PUT')
#(('PUT', 'g', (2, 3)), ('REM', (1, 3)))
game.get_next_state((1, 17, 10), 'PUT')
#(('PUT', 'b', (1, 1)), ('REM', (3, 1)))
game.get_next_state((2, 8, 22), 'PUT')
#(('PUT', 'b', (2, 1)), ('REM', (0, 2)))
game.get_next_state((2, 15, 2), 'PUT')
placement, capture = game.get_valid_actions()
self.assertEqual(np.sum(capture), 5)
self.assertTrue(np.all(placement == False))
game = ZertzGame(1)
placement, capture = game.get_valid_actions()
self.assertEqual(np.sum(placement), 3)
self.assertTrue(np.all(placement[:, :, 1]))
self.assertTrue(np.all(capture == False))
def test_mcts2(self):
# set up
rings = 19
marbles = {'w': 10, 'g': 10, 'b': 10}
win_con = [{'w': 2}, {'g': 2}, {'b': 2}, {'w': 1, 'g': 1, 'b': 1}]
t = 3
game = Game(rings, marbles, win_con, t)
nnet = DumbNN(game)
# take some actions
#Human: PUT g B1 B4
game.get_next_state((1, 16, 1), 'PUT')
#AI: PUT b D3 C5
game.get_next_state((2, 13, 2), 'PUT')
#Human: PUT b E1 C4
game.get_next_state((2, 24, 7), 'PUT')
#AI: PUT w B2 D1
game.get_next_state((0, 11, 23), 'PUT')
#Human: CAP g B1 w B3
game.get_next_state((3, 3, 1), 'CAP')
#AI: PUT g A3 D4
#game.get_next_state((1, 0, 8), 'PUT')
#Human: CAP g A3 g C3
#game.get_next_state((5, 0, 0), 'CAP')
#Human: CAP g C3 b E3
# do MCTS
board_state, player_value = game.get_current_state()
print(board_state[0] + board_state[1] + board_state[2]*2 + board_state[3]*3)
print(board_state[-1])
ai = MCTS(game, nnet, 1, 6)
ai.reset(player_value)
ai.get_action_prob(board_state, temp=0)
def test_mcts(self):
# set up
rings = 19
marbles = {'w': 10, 'g': 10, 'b': 10}
win_con = [{'w': 2}, {'w': 1, 'g': 1, 'b': 1}]
t = 3
game = Game(rings, marbles, win_con, t)
nnet = DumbNN(game)
# take some actions
#(('PUT', 'w', (4, 4)), ('REM', (4, 3)))
game.get_next_state((0, 24, 23), 'PUT')
#(('PUT', 'b', (3, 4)), ('REM', (4, 2)))
game.get_next_state((2, 19, 22), 'PUT')
#(('PUT', 'g', (2, 3)), ('REM', (1, 3)))
game.get_next_state((1, 13, 8), 'PUT')
#(('PUT', 'b', (1, 1)), ('REM', (3, 1)))
game.get_next_state((1, 6, 16), 'PUT')
#(('PUT', 'b', (2, 1)), ('REM', (0, 2)))
game.get_next_state((2, 11, 2), 'PUT')
#(('PUT', 'w', (3, 3)), ('REM', (0, 0)))
game.get_next_state((0, 18, 0), 'PUT')
# do MCTS
board_state, player_value = game.get_current_state()
print(board_state[0] + board_state[1] + board_state[2]*2 + board_state[3]*3)
ai = MCTS(game, nnet, 1, 50)
ai.reset(player_value)
ai.get_action_prob(board_state, temp=0)
def __init__(self):
ZertzGame.__init__(self)