def get_buffer(config, game) -> (ChessEnv, list): """ Gets data to load into the buffer by playing a game using PGN data. :param Config config: config to use to play the game :param pgn.Game game: game to play :return list(str,list(float)): data from this game for the SupervisedLearningWorker.buffer """ env = ChessEnv().reset() white = ChessPlayer(config, dummy=True) black = ChessPlayer(config, dummy=True) result = game.headers["Result"] white_elo, black_elo = int(game.headers["WhiteElo"]), int(game.headers["BlackElo"]) white_weight = clip_elo_policy(config, white_elo) black_weight = clip_elo_policy(config, black_elo) actions = [] while not game.is_end(): game = game.variation(0) actions.append(game.move.uci()) k = 0 while not env.done and k < len(actions): if env.white_to_move: action = white.sl_action(env.observation, actions[k], weight=white_weight) #ignore=True else: action = black.sl_action(env.observation, actions[k], weight=black_weight) #ignore=True env.step(action, False) k += 1 if not env.board.is_game_over() and result != '1/2-1/2': env.resigned = True if result == '1-0': env.winner = Winner.white black_win = -1 elif result == '0-1': env.winner = Winner.black black_win = 1 else: env.winner = Winner.draw black_win = 0 black.finish_game(black_win) white.finish_game(-black_win) data = [] for i in range(len(white.moves)): data.append(white.moves[i]) if i < len(black.moves): data.append(black.moves[i]) return env, data
def get_buffer(config, game) -> (ChessEnv, list): """ Gets data to load into the buffer by playing a game using PGN data. :param Config config: config to use to play the game :param pgn.Game game: game to play :return list(str,list(float)): data from this game for the SupervisedLearningWorker.buffer """ env = ChessEnv().reset() white = ChessPlayer(config, dummy=True) black = ChessPlayer(config, dummy=True) result = game.headers["Result"] # Rare cases where elo ratings are not in the headers if "WhiteElo" not in game.headers or "BlackElo" not in game.headers: return None, None, None, None, False white_elo, black_elo = int(game.headers["WhiteElo"]), int( game.headers["BlackElo"]) white_weight = clip_elo_policy(config, white_elo) black_weight = clip_elo_policy(config, black_elo) actions = [] while not game.is_end(): game = game.variation(0) actions.append(game.move.uci()) k = 0 while not env.done and k < len(actions): if env.white_to_move: action = white.sl_action(env.observation, actions[k], weight=white_weight) # ignore=True else: action = black.sl_action(env.observation, actions[k], weight=black_weight) # ignore=True env.step(action, False) k += 1 if not env.board.is_game_over() and result != '1/2-1/2': env.resigned = True if result == '1-0': env.winner = Winner.white black_win = -1 elif result == '0-1': env.winner = Winner.black black_win = 1 else: env.winner = Winner.draw black_win = 0 black.finish_game(black_win) white.finish_game(-black_win) fen_data = [] moves_array = np.zeros( (len(white.moves) + len(black.moves), white.labels_n), dtype=np.float16) scores = np.zeros((len(white.moves) + len(black.moves)), dtype=np.int8) for i in range(len(white.moves)): fen_data.append(white.moves[i][0]) moves_array[i * 2] = white.moves[i][1] scores[i * 2] = white.moves[i][2] if i < len(black.moves): fen_data.append(black.moves[i][0]) moves_array[i * 2 + 1] = black.moves[i][1] scores[i * 2 + 1] = black.moves[i][2] return env, fen_data, moves_array, scores, True