def saveWTL(config, players, wtl): for i in range(wtl.shape[0]): for j in range(wtl.shape[1]): util.saveWTL(config, util.getPlayerName(players[i]), util.getPlayerName(players[j]), wtl[i, j, 0], wtl[i, j, 1], wtl[i, j, 2]) util.mergeStagedWTL(config)
def run_games(config): game = Othello() model = "" x = config.iterations while(x != 0): x -= 1 models = sorted(glob.glob(config.data.model_location+"*.h5")) if model == "": model = models[-1] print("Loading new model: %s" % util.getPlayerName(model)) ai = AIPlayer(config.buffer_size, config.game.simulation_num_per_move, model=model) elif models[-1] != model: model = models[-1] print("Loading new model: %s" % util.getPlayerName(model)) ai.load(model) start=time() for j in range(config.nb_game_in_file): util.print_progress_bar(j, config.nb_game_in_file, start=start) side = -1 turn = 1 while not game.game_over(): ai.tau = config.game.tau_1 if config.game.tau_swap < turn: ai.tau = config.game.tau_2 t = ai.pick_move(game, side) game.play_move(t[0], t[1], side) side *= -1 turn += 1 ai.update_buffer(game.get_winner()) game.reset_board() #print("Average Game Time: ", (time()-start)/(config.nb_game_in_file)) util.print_progress_bar(config.nb_game_in_file, config.nb_game_in_file, start=start) save_games(config, ai.buffer) t.join()
def savePerformance(config, model_1, model_2, wins, ties, losses): m1 = config.model_1 if m1 == "newest": m1 = util.getPlayerName(model_1) m2 = config.model_2 if m2 == "newest": m2 = util.getPlayerName(model_2) util.saveWTL(config, m1, m2, wins, ties, losses) util.mergeStagedWTL(config)
def load_player(player, player_name, current, config): if player_name == "newest": model = sorted(glob.glob(config.data.model_location+"*.h5"))[-1] if model != current: print("Loading new model: %s" % util.getPlayerName(model)) player.load(model) return model else: return current
def load_player(player, player_name, current, config): if player_name == "newest": model = sorted(glob.glob(config.data.model_location + "*.h5"))[-1] if model != current: print("Loading new model: %s" % util.getPlayerName(model)) player.load(model) return model else: return current
def calc_ranking(config): models = sorted(glob.glob(config.data.model_location+"*.h5")) players = [] for i, model in enumerate(models): if i % config.model_skip == 0 or i == len(models): players.append(model) wtl = np.zeros((len(players), len(players), 3)) win_matrix = np.zeros((len(players),len(players))) game = Othello() challenger1 = AIPlayer(0, config.game.simulation_num_per_move, train=False, model=players[-1], tau=config.game.tau_1) challenger2 = AIPlayer(0, config.game.simulation_num_per_move, train=False, model=players[0], tau=config.game.tau_1) total_games = (config.game_num_per_model * (len(players)))//2 played_games = 0 finished = False start = time() print("Ranking with %d players and %d games per player" % (len(players), config.game_num_per_model)) if config.game_num_per_model < len(players): print("We suggest that you increase games per player to be greater than players") for i in itertools.count(): ranks = getRankings(win_matrix) if len(ranks) == 0: msg = "No Clear Best Yet" else: msg = "Current Best is "+util.getPlayerName(players[ranks[-1]]) if config.print_best: print(msg.ljust(90)) for j in range(len(players)): util.print_progress_bar(played_games, total_games, start=start) challenger1_index = getLeastPlayed(win_matrix, j) AIPlayer.clear() challenger1.load(players[challenger1_index]) challenger2.load(players[j]) if random.random() < 0.5: challenger1_side = -1 p1 = challenger1 p2 = challenger2 else: challenger1_side = 1 p1 = challenger2 p2 = challenger1 side = -1 turn = 1 while not game.game_over(): tau = config.game.tau_1 if config.game.tau_swap < turn: tau = config.game.tau_2 p1.tau = tau p2.tau = tau if side == -1: t = p1.pick_move(game, side) else: t = p2.pick_move(game, side) game.play_move(t[0], t[1], side) side *= -1 turn += 1 if game.get_winner() == challenger1_side: win_matrix[challenger1_index,j] += 1 wtl[challenger1_index, j,0] += 1 elif game.get_winner() == -1*challenger1_side: win_matrix[j, challenger1_index] += 1 wtl[challenger1_index, j,2] += 1 else: win_matrix[challenger1_index,j] += 0.5 win_matrix[j, challenger1_index] += 0.5 wtl[challenger1_index, j, 1] += 1 game.reset_board() played_games += 1 if played_games >= total_games: finished = True break saveWTL(config, players, wtl) wtl = np.zeros((len(players), len(players), 3)) if finished: break util.print_progress_bar(total_games, total_games, start=start) print("\n",[util.getPlayerName(player) for player in players]) print("\nWin Matrix(row beat column):") print(win_matrix) try: with np.errstate(divide='ignore', invalid='ignore'): params = choix.ilsr_pairwise_dense(win_matrix) print("\nRankings:") for i, player in enumerate(np.argsort(params)[::-1]): print("%d. %s (expected %d) with %0.2f rating"% (i+1, util.getPlayerName(players[player]), len(players)-player, params[player])) print("\n(Rating Diff, Winrate) -> (0.5, 62%), (1, 73%), (2, 88%), (3, 95%), (5, 99%)") except Exception: print("\nNot Enough data to calculate rankings")
def saveWTL(config, players, wtl): for i in range(wtl.shape[0]): for j in range(wtl.shape[1]): util.saveWTL(config, util.getPlayerName(players[i]), util.getPlayerName(players[j]), wtl[i,j,0], wtl[i,j,1], wtl[i,j,2]) util.mergeStagedWTL(config)
def calc_ranking(config): models = sorted(glob.glob(config.data.model_location + "*.h5")) players = [] for i, model in enumerate(models): if i % config.model_skip == 0 or i == len(models): players.append(model) wtl = np.zeros((len(players), len(players), 3)) win_matrix = np.zeros((len(players), len(players))) game = Othello() challenger1 = AIPlayer(0, config.game.simulation_num_per_move, train=False, model=players[-1], tau=config.game.tau_1) challenger2 = AIPlayer(0, config.game.simulation_num_per_move, train=False, model=players[0], tau=config.game.tau_1) total_games = (config.game_num_per_model * (len(players))) // 2 played_games = 0 finished = False start = time() print("Ranking with %d players and %d games per player" % (len(players), config.game_num_per_model)) if config.game_num_per_model < len(players): print( "We suggest that you increase games per player to be greater than players" ) for i in itertools.count(): ranks = getRankings(win_matrix) if len(ranks) == 0: msg = "No Clear Best Yet" else: msg = "Current Best is " + util.getPlayerName(players[ranks[-1]]) if config.print_best: print(msg.ljust(90)) for j in range(len(players)): util.print_progress_bar(played_games, total_games, start=start) challenger1_index = getLeastPlayed(win_matrix, j) AIPlayer.clear() challenger1.load(players[challenger1_index]) challenger2.load(players[j]) if random.random() < 0.5: challenger1_side = -1 p1 = challenger1 p2 = challenger2 else: challenger1_side = 1 p1 = challenger2 p2 = challenger1 side = -1 turn = 1 while not game.game_over(): tau = config.game.tau_1 if config.game.tau_swap < turn: tau = config.game.tau_2 p1.tau = tau p2.tau = tau if side == -1: t = p1.pick_move(game, side) else: t = p2.pick_move(game, side) game.play_move(t[0], t[1], side) side *= -1 turn += 1 if game.get_winner() == challenger1_side: win_matrix[challenger1_index, j] += 1 wtl[challenger1_index, j, 0] += 1 elif game.get_winner() == -1 * challenger1_side: win_matrix[j, challenger1_index] += 1 wtl[challenger1_index, j, 2] += 1 else: win_matrix[challenger1_index, j] += 0.5 win_matrix[j, challenger1_index] += 0.5 wtl[challenger1_index, j, 1] += 1 game.reset_board() played_games += 1 if played_games >= total_games: finished = True break saveWTL(config, players, wtl) wtl = np.zeros((len(players), len(players), 3)) if finished: break util.print_progress_bar(total_games, total_games, start=start) print("\n", [util.getPlayerName(player) for player in players]) print("\nWin Matrix(row beat column):") print(win_matrix) try: with np.errstate(divide='ignore', invalid='ignore'): params = choix.ilsr_pairwise_dense(win_matrix) print("\nRankings:") for i, player in enumerate(np.argsort(params)[::-1]): print("%d. %s (expected %d) with %0.2f rating" % (i + 1, util.getPlayerName( players[player]), len(players) - player, params[player])) print( "\n(Rating Diff, Winrate) -> (0.5, 62%), (1, 73%), (2, 88%), (3, 95%), (5, 99%)" ) except Exception: print("\nNot Enough data to calculate rankings")