def test_mcts_value_at_children_of_root():
mock_game = MockGame()
root = MCTSNode(0, player=1)
assert root.N == 0
mcts(root, mock_game, mock_game.mock_estimator, 100, 1.0)
terminal_nodes = get_terminal_nodes(root)
N_terminal_nodes = sum(node.N for node in terminal_nodes)
# Each iteration of MCTS we should add 1 to W of one of the
# children of the root.
assert sum(child.W for child in root.children.values()) == N_terminal_nodes
def test_mcts_action_count_at_root_children():
mock_game = MockGame()
root = MCTSNode(0, player=1)
action_probs = mcts(root, mock_game, mock_game.mock_estimator, 100, 1.0)
# Each iteration of MCTS we should add 1 to N at the root.
assert sum(child.N for child in root.children.values()) == 99
def test_can_run_mcts_on_fake_game():
""" This test shows that we can run MCTS using a 'next_states'
function and 'evaluator' function.
"""
mock_game = MockGame()
root = MCTSNode(0, player=1)
action_probs = mcts(root, mock_game, mock_game.mock_estimator, 100, 1.0)
assert action_probs is not None
def test_mcts_action_count_at_root():
mock_game = MockGame()
root = MCTSNode(0, player=1)
assert root.N == 0
action_probs = mcts(root, mock_game, mock_game.mock_estimator, 100, 1.0)
# Each iteration of MCTS we should add 1 to N at the root.
assert root.N == 100
def test_print_mcts():
mock_game = MockGame()
# TODO: This doesn't currently test anything.
# TODO: This next line is probably a bad idea
mock_game.TERMINAL_STATE_VALUES = [0.01 * i for i in range(12)]
root = MCTSNode(0, player=1)
action_probs = mcts(root, mock_game, mock_game.mock_estimator, 10, 1.0)
print_tree(root)
mock_game.TERMINAL_STATE_VALUES = (1, ) * 12
def test_mcts_can_play_fake_game(evaluator, expected):
mock_game = MockGame()
root = MCTSNode(0, player=1)
node = root
nodes = [node]
while not node.is_terminal:
action_probs = mcts(root, mock_game, evaluator, 100, 1)
action = max(action_probs, key=action_probs.get)
node = node.children[action]
nodes.append(node)
assert [node.game_state for node in nodes] == expected
def test_neural_net_estimator():
mock_game = MockGame()
nnet = MockNetEstimator(learning_rate=0.01)
root = MCTSNode(0, player=1)
action_probs = mcts(root, mock_game, nnet, 100, 1.0)