def run_episode(self):
examples = []
board = get_random_board()
initial_node = MCTSNode(is_initial=True)
player_blue = NNPlayer(Color.BLUE,
n_simulations=self.n_simulations,
current_node=initial_node,
janggi_net=self.predictor,
temperature_start=1,
temperature_threshold=30,
temperature_end=0.01)
player_red = NNPlayer(Color.RED,
n_simulations=self.n_simulations,
current_node=initial_node,
janggi_net=self.predictor,
temperature_start=1,
temperature_threshold=30,
temperature_end=0.01)
game = Game(player_blue, player_red, board)
while not game.is_finished(self.iter_max):
new_action = game.get_next_action()
game.actions.append(new_action)
if game.current_player == Color.BLUE:
examples.append([
board.get_features(game.current_player, game.round),
player_blue.current_node.get_policy(game.current_player),
Color.BLUE
])
examples.append([
board.get_features(game.current_player,
game.round,
data_augmentation=True),
player_blue.current_node.get_policy(
game.current_player, data_augmentation=True),
Color.BLUE
])
else:
examples.append([
board.get_features(game.current_player,
game.round,
data_augmentation=True),
player_red.current_node.get_policy(game.current_player,
data_augmentation=True),
Color.RED
])
examples.append([
board.get_features(game.current_player, game.round),
player_red.current_node.get_policy(game.current_player),
Color.RED
])
game.board.apply_action(new_action)
game.switch_player()
game.board.invalidate_action_cache(
new_action) # Try to reduce memory usage
game.round += 1
winner = game.get_winner()
set_winner(examples, winner)
return examples
def test_stockfish(self):
board = get_random_board()
process = get_process_stockfish(board, "level 40 5 0")
player_blue = StockfishPlayer(Color.BLUE, process, think_time=-1)
player_red = StockfishPlayer(Color.RED, process, think_time=-1)
game = Game(player_blue, player_red, board)
winner = game.run_game(200)
print(winner)
print(repr(game.board))
def fight(player_blue, player_red, iter_max, print_board=False):
board = get_random_board()
game = Game(player_blue, player_red, board)
winner = game.run_game(iter_max, print_board=print_board)
print("Winner:", winner)
print("Score BLUE:", board.get_score(Color.BLUE))
print("Score RED:", board.get_score(Color.RED))
print(repr(board))
print(board)
print(game.to_uci_usi())
return winner
def run_episode_stockfish(args):
print("Starting episode", current_process().name)
begin_time = time.time()
iter_max = args
board = get_random_board()
process = get_process_stockfish(board)
player_blue = StockfishPlayer(Color.BLUE, process, think_time=2)
player_red = StockfishPlayer(Color.RED, process, think_time=2)
game = run_game(board, player_blue, player_red, iter_max)
print("Time Episode: ", time.time() - begin_time)
return game.dumps()
def test_single_action_random(self):
n_simulations = 800
node = MCTSNode()
player_blue = RandomMCTSPlayer(Color.BLUE,
n_simulations=n_simulations,
current_node=node)
player_red = RandomMCTSPlayer(Color.RED,
n_simulations=n_simulations,
current_node=node)
board = get_random_board()
game = Game(player_blue, player_red, board)
game.get_next_action()
def test_random_vs_random(self):
n_simulations = 400
node = MCTSNode()
player_blue = RandomMCTSPlayer(Color.BLUE,
n_simulations=n_simulations,
current_node=node)
player_red = RandomMCTSPlayer(Color.RED,
n_simulations=n_simulations,
current_node=node)
# winner = fight(player_blue, player_red, 200)
board = get_random_board()
game = Game(player_blue, player_red, board)
winner = game.run_game(200)
self.assertIn(winner, [Color.BLUE, Color.RED])
print(game.to_json(node))
def test_single_action_nn(self):
n_simulations = 800
player_blue = NNPlayer(Color.BLUE,
n_simulations=n_simulations,
janggi_net=JanggiNetwork(),
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01)
player_red = NNPlayer(Color.RED,
n_simulations=n_simulations,
janggi_net=JanggiNetwork(),
temperature_start=0.01,
temperature_threshold=30,
temperature_end=0.01)
board = get_random_board()
game = Game(player_blue, player_red, board)
game.get_next_action()
def run_episode_raw_not_nn(args):
print("Starting episode", current_process().name)
begin_time = time.time()
n_simulations, iter_max = args
board = get_random_board()
initial_node = MCTSNode(is_initial=True)
player_blue = RandomMCTSPlayer(Color.BLUE,
n_simulations=n_simulations,
current_node=initial_node,
temperature_start=1,
temperature_threshold=30,
temperature_end=0.01)
player_red = RandomMCTSPlayer(Color.RED,
n_simulations=n_simulations,
current_node=initial_node,
temperature_start=1,
temperature_threshold=30,
temperature_end=0.01)
game = run_game(board, player_blue, player_red, iter_max)
print("Time Episode: ", time.time() - begin_time)
return game.to_json(initial_node)