def test_noughts_and_crosses_instances_have_correct_size(
self, size, mocker):
mock_game = mocker.MagicMock()
rows, columns = size
NoughtsAndCrosses.__init__(mock_game, *size)
assert mock_game.rows == rows
assert mock_game.columns == columns
def test_display_function_outputs_correct_string_for_3x3(
self, state, expected_output, capsys, mocker):
mock_game = mocker.MagicMock(
rows=3, columns=3, _actions_to_binary=actions_to_binary_list[0])
NoughtsAndCrosses.display(mock_game, state)
output = capsys.readouterr().out
assert output == expected_output
def test_legal_actions_raises_exception_on_terminal_input_state(
self, mocker):
mock_game = mocker.MagicMock()
mock_game.is_terminal = mocker.MagicMock(return_value=True)
mock_state = mocker.MagicMock()
with pytest.raises(ValueError) as exception_info:
NoughtsAndCrosses.legal_actions(mock_game, mock_state)
assert str(
exception_info.value) == ("Legal actions can not be computed"
" for a terminal state.")
def test_utility_raises_exception_on_non_terminal_input_state(
self, size, state, mocker):
rows, columns = size
mock_is_terminal = mocker.MagicMock(return_value=False)
mock_game = mocker.MagicMock(rows=rows,
columns=columns,
is_terminal=mock_is_terminal)
with pytest.raises(ValueError) as exception_info:
NoughtsAndCrosses.utility(mock_game, state)
assert str(exception_info.value) == ("Utility can not be calculated "
"for a non-terminal state.")
mock_is_terminal.assert_called_once_with(state)
def test_random_noughts_and_crosses_player_gives_equal_action_probabilities():
nac = NoughtsAndCrosses()
player = RandomPlayer(game=nac)
action, action_probs = player.choose_action(nac.initial_state,
return_probabilities=True)
next_states = nac.legal_actions(nac.initial_state)
expected_action_probs = {
action: 1 / len(next_states)
for action in next_states.keys()
}
for action in expected_action_probs.keys():
np.testing.assert_almost_equal(action_probs[action],
expected_action_probs[action])
def test_is_terminal_returns_false_for_non_terminal_states(
self, size, actions_to_binary, win_bitmasks, state, mocker):
rows, columns = size
mock_game = mocker.MagicMock(rows=rows,
columns=columns,
_actions_to_binary=actions_to_binary,
_win_bitmasks=win_bitmasks)
assert NoughtsAndCrosses.is_terminal(mock_game, state) is False
def test_utility_function_returns_correct_outcomes(self, size,
win_bitmasks, state,
outcome, mocker):
rows, columns = size
mock_game = mocker.MagicMock(rows=rows,
columns=columns,
_win_bitmasks=win_bitmasks)
assert NoughtsAndCrosses.utility(mock_game, state) == outcome
def test_calculating_row_win_bitmasks(self, size, actions_to_binary,
win_bitmasks, mocker):
rows, columns = size
mock_game = mocker.MagicMock(rows=rows,
columns=columns,
_actions_to_binary=actions_to_binary)
row_win_bitmasks = NoughtsAndCrosses._calculate_row_bitmasks(mock_game)
print([(bin(x), bin(y))
for x, y in zip(row_win_bitmasks, win_bitmasks["row"])])
assert row_win_bitmasks == win_bitmasks["row"]
def test_generating_a_dict_of_all_possible_next_states(
self, size, actions_to_binary, state, player, expected_states,
mocker):
# TODO: need to split this into two tests: one testing the _next_state function and one testing legal actions
rows, columns = size
mock_game = mocker.MagicMock(rows=rows,
columns=columns,
_actions_to_binary=actions_to_binary)
mock_game.is_terminal = mocker.MagicMock(return_value=False)
mock_game.current_player = mocker.MagicMock(return_value=player)
assert NoughtsAndCrosses.legal_actions(mock_game,
state) == expected_states
def test_mcts_noughts_and_crosses_player_gives_optimal_moves(
state, optimal_actions):
# seed the random number generator.
np.random.seed(0)
nac = NoughtsAndCrosses()
estimator = create_trivial_estimator(nac)
player = MCTSPlayer(game=nac,
estimator=estimator,
mcts_iters=100,
c_puct=0.5,
tau=1)
action, action_probs = player.choose_action(state,
return_probabilities=True)
print(action_probs)
assert max(action_probs, key=action_probs.get) in optimal_actions
def test_action_to_binary_is_correct(self, size, actions_to_binary,
mocker):
mock_game = mocker.MagicMock()
NoughtsAndCrosses.__init__(mock_game, *size)
assert mock_game._actions_to_binary == actions_to_binary
def test_initial_state_is_correct(self, size, mocker):
mock_game = mocker.MagicMock()
NoughtsAndCrosses.__init__(mock_game, *size)
assert mock_game.initial_state == (0, 0, 1)
def test_current_player_returns_correct_player(self, player, state,
mocker):
mock_game = mocker.MagicMock()
assert NoughtsAndCrosses.current_player(mock_game, state) == player
"""This program plays noughts and crosses using Monte Carlo Tree Search and a
trivial evaluator. For nonterminal states, the evaluator returns the uniform
probability distribution over available actions and a value of 0. In a terminal
state, we back up the utility returned by the game.
"""
import numpy as np
from alphago.games.noughts_and_crosses import NoughtsAndCrosses
from alphago.estimator import create_trivial_estimator
from alphago.player import MCTSPlayer
if __name__ == "__main__":
nac = NoughtsAndCrosses()
evaluator = create_trivial_estimator(nac.legal_actions)
state = nac.INITIAL_STATE
computer_player_no = np.random.choice([1, 2])
computer_player = MCTSPlayer(nac,
evaluator,
mcts_iters=2000,
c_puct=0.5,
tau=0.01)
human_player_no = 1 if computer_player_no == 2 else 2
print("You are player: {}".format(human_player_no))
while not nac.is_terminal(state):
player_no = nac.current_player(state)
next_states = nac.legal_actions(state)
if player_no == computer_player_no:
action = computer_player.choose_action(state)
computer_player.update(action)
print("Taking action: {}".format(action))
