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 run_game(board, player_blue, player_red, iter_max):
game = Game(player_blue, player_red, board)
while not game.is_finished(iter_max):
new_action = game.get_next_action()
game.actions.append(new_action)
game.board.apply_action(new_action)
game.switch_player()
game.board.invalidate_action_cache(
new_action) # Try to reduce memory usage
game.round += 1
return game