def start():
config = OptimizerConfig()
tf_util.update_memory(config.gpu_mem_fraction)
util.set_high_process_priority()
AIPlayer.create_if_nonexistant(config)
models = sorted(glob.glob(config.data.model_location+"*.h5"))
ai = AIPlayer(config.buffer_size, 1, model=models[-1], compile=True)
train(ai, config)
def start():
config = OptimizerConfig()
tf_util.update_memory(config.gpu_mem_fraction)
util.set_high_process_priority()
AIPlayer.create_if_nonexistant(config)
models = sorted(glob.glob(config.data.model_location + "*.h5"))
ai = AIPlayer(config.buffer_size, 1, model=models[-1], compile=True)
train(ai, 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 create_player(player_name, config):
if player_name == "random":
model = "random"
player = RandomPlayer()
elif player_name == "newest":
model = sorted(glob.glob(config.data.model_location+"*.h5"))[-1]
#print("Loading model: %s" % model)
player = AIPlayer(0, config.game.simulation_num_per_move, train=False, model=model, tau=config.game.tau_1)
else:
model = config.data.model_location+player_name
player = AIPlayer(0, config.game.simulation_num_per_move, train=False, model=model, tau=config.game.tau_1)
return player
def run(self):
self.game_gui = Canvas(self.root, width=600, height=600, background='green')
self.game_gui.bind("<Button-1>", self.click)
self.game_gui.focus_set()
self.game_gui.bind("<Key>", self.key)
self.game_gui.pack()
for i in range(1, 8):
self.game_gui.create_line(0, i*75, 600, i*75)
self.game_gui.create_line(i*75, 0, i*75, 600)
self.pieces = []
for i in range(8):
self.pieces.append([])
for j in range(8):
self.pieces[i].append(self.game_gui.create_oval(i*75+5, j*75+5, (i+1)*75-5, (j+1)*75-5, fill="green", outline="green"))
self.root.protocol("WM_DELETE_WINDOW", self.on_closing)
self.root.resizable(0,0)
self.running = True
config = EvaluateConfig()
tf_util.update_memory(config.gpu_mem_fraction)
AIPlayer.create_if_nonexistant(config)
self.game = Othello()
if(random() > 0.5):
self.human = 1
else:
self.human = -1
ai = create_player(config.model_1, config)
#print("You are playing against", config.model_1)
#print("Playing games with %d simulations per move" % config.game.simulation_num_per_move)
self.side = -1
self.draw_board()
self.value = ai.evaluate(self.game, self.side)
while self.running and not self.game.game_over():
#play move
if self.side != self.human:
self.value = ai.evaluate(self.game, self.side)
self.root.title("Othello (Thinking of Move) Current Value: %0.2f (1 white wins, -1 black wins)" % self.value)
self.root.config(cursor="wait")
t = ai.pick_move(self.game, self.side)
self.game.play_move(t[0], t[1], self.side)
self.draw_board()
self.side *= -1
self.value = ai.evaluate(self.game, self.side)
else:
if len(self.game.possible_moves(self.side)) == 0:
self.side *= -1
continue
if self.side == -1:
color = "black"
else:
color = "white"
self.root.title("Othello (Play as %s) Current Value: %0.2f (1 white wins, -1 black wins)" % (color, self.value))
self.root.config(cursor="")
if self.update:
self.update = False
if (self.x, self.y) in self.game.possible_moves(self.side):
self.game.play_move(self.x, self.y, self.side)
self.draw_board()
self.side *= -1
time.sleep(0.01)
if self.human == self.game.get_winner():
self.root.title("Othello (You Win!)")
elif self.game.get_winner() == 0:
self.root.title("Othello (Its a draw!)")
else:
self.root.title("Othello (You Lose!)")
def start():
config = EvaluateConfig()
tf_util.update_memory(config.gpu_mem_fraction)
AIPlayer.create_if_nonexistant(config)
run_games(config)
def start():
config = EvaluateConfig()
tf_util.update_memory(config.gpu_mem_fraction)
AIPlayer.create_if_nonexistant(config)
run_games(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), 3))
win_matrix = np.zeros((len(players), len(players)))
game = Othello()
king_index = len(players) - 1
king = AIPlayer(0,
config.game.simulation_num_per_move,
train=False,
model=players[king_index],
tau=config.game.tau_1)
challenger = 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
start = time()
print("Playing king of the hill 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 range(math.ceil(total_games / (len(players) - 1))):
AIPlayer.clear()
king_index = getKingIndex(win_matrix)
if king_index == -1:
king_index = (len(players) - 1) - i % len(players)
msg = "No King Yet"
else:
msg = "King is " + os.path.basename(
players[king_index]).split(".")[0]
king.load(players[king_index])
if config.print_king:
print(msg.ljust(90))
for j in range(len(players)):
util.print_progress_bar(played_games, total_games, start=start)
if j == king_index:
continue
challenger.load(players[j])
if random.random() < 0.5:
king_side = -1
p1 = king
p2 = challenger
else:
king_side = 1
p1 = challenger
p2 = king
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() == king_side:
win_matrix[king_index, j] += 1
wtl[king_index, 0] += 1
wtl[j, 2] += 1
elif game.get_winner() == -1 * king_side:
win_matrix[j, king_index] += 1
wtl[king_index, 2] += 1
wtl[j, 0] += 1
else:
win_matrix[king_index, j] += 0.5
win_matrix[j, king_index] += 0.5
wtl[king_index, 1] += 1
wtl[j, 1] += 1
game.reset_board()
played_games += 1
if played_games == total_games:
break
util.print_progress_bar(total_games, total_games, start=start)
try:
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 and results of %d-%d-%d"
% (i + 1, os.path.basename(players[player]).split(".")[0],
len(players) - player, params[player], wtl[player, 0],
wtl[player, 1], wtl[player, 2]))
print(
"\n(Rating Diff, Winrate) -> (0.5, 62%), (1, 73%), (2, 88%), (3, 95%), (5, 99%)"
)
except Exception:
print("\n Not Enough data to calculate rankings")
print("\nWin Matrix:")
print(win_matrix)
print("\nResults:")
for player in range(win_matrix.shape[0]):
print("%s results of %d-%d-%d" %
(os.path.basename(players[player]).split(".")[0],
wtl[player, 0], wtl[player, 1], wtl[player, 2]))
def start():
config = RankingConfig()
tf_util.update_memory(config.gpu_mem_fraction)
AIPlayer.create_if_nonexistant(config)
calc_ranking(config)
def run(self):
self.game_gui = Canvas(self.root, width=600, height=600, background='green')
self.game_gui.bind("<Button-1>", self.click)
self.game_gui.focus_set()
self.game_gui.bind("<Key>", self.key)
self.game_gui.pack()
for i in range(1, 8):
self.game_gui.create_line(0, i*75, 600, i*75)
self.game_gui.create_line(i*75, 0, i*75, 600)
self.pieces = []
for i in range(8):
self.pieces.append([])
for j in range(8):
self.pieces[i].append(self.game_gui.create_oval(i*75+5, j*75+5, (i+1)*75-5, (j+1)*75-5, fill="green", outline="green"))
self.root.protocol("WM_DELETE_WINDOW", self.on_closing)
self.root.resizable(0,0)
self.running = True
config = EvaluateConfig()
tf_util.update_memory(config.gpu_mem_fraction)
AIPlayer.create_if_nonexistant(config)
self.game = Othello()
if(random() > 0.5):
self.human = 1
else:
self.human = -1
ai = create_player(config.model_1, config)
#print("You are playing against", config.model_1)
#print("Playing games with %d simulations per move" % config.game.simulation_num_per_move)
self.side = -1
self.draw_board()
self.value = ai.evaluate(self.game, self.side)
while self.running and not self.game.game_over():
#play move
if self.side != self.human:
self.value = ai.evaluate(self.game, self.side)
self.root.title("Othello (Thinking of Move) Current Value: %0.2f (1 white wins, -1 black wins)" % self.value)
self.root.config(cursor="wait")
t = ai.pick_move(self.game, self.side)
self.game.play_move(t[0], t[1], self.side)
self.draw_board()
self.side *= -1
self.value = ai.evaluate(self.game, self.side)
else:
if len(self.game.possible_moves(self.side)) == 0:
self.side *= -1
continue
if self.side == -1:
color = "black"
else:
color = "white"
self.root.title("Othello (Play as %s) Current Value: %0.2f (1 white wins, -1 black wins)" % (color, self.value))
self.root.config(cursor="")
if self.update:
self.update = False
if (self.x, self.y) in self.game.possible_moves(self.side):
self.game.play_move(self.x, self.y, self.side)
self.draw_board()
self.side *= -1
time.sleep(0.01)
self.root.config(cursor="")
if self.human == self.game.get_winner():
self.root.title("Othello (You Win!)")
elif self.game.get_winner() == 0:
self.root.title("Othello (Its a draw!)")
else:
self.root.title("Othello (You Lose!)")
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), 3))
win_matrix = np.zeros((len(players), len(players)))
game = Othello()
##give every player a random order to play games against opponents
order = []
for i in range(len(players)):
nums = [x for x in range(len(players))]
nums.remove(i)
random.shuffle(nums)
order.append(nums)
p1 = AIPlayer(1, config.game.simulation_num_per_move, model=players[0])
p2 = AIPlayer(1,
config.game.simulation_num_per_move,
model=players[order[0][0]])
start = time()
print(
"Playing random round robin with %d players and %d games per player" %
(len(players), config.game_num_per_model))
for i in range(config.game_num_per_model // 2):
util.print_progress_bar(i, config.game_num_per_model // 2, start=start)
ordering = [x for x in range(len(players))]
random.shuffle(ordering)
for j in ordering:
AIPlayer.clear()
x = i
if x >= len(order[j]):
x %= len(order[j])
if x == 0:
random.shuffle(order[j])
p1.load(players[j])
p2.load(players[order[j][x]])
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() == -1:
win_matrix[j, order[j][x]] += 1
wtl[j, 0] += 1
wtl[order[j][x], 2] += 1
elif game.get_winner() == 1:
win_matrix[order[j][x], j] += 1
wtl[j, 2] += 1
wtl[order[j][x], 0] += 1
else:
win_matrix[j, order[j][x]] += 0.5
win_matrix[order[j][x], j] += 0.5
wtl[j, 1] += 1
wtl[order[j][x], 1] += 1
game.reset_board()
util.print_progress_bar(config.game_num_per_model // 2,
config.game_num_per_model // 2,
start=start)
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 and results of %d-%d-%d" %
(i + 1, os.path.basename(players[player]), len(players) - player,
params[player], wtl[player, 0], wtl[player, 1], wtl[player, 2]))
print(
"\n(Rating Diff, Winrate) -> (0.5, 62%), (1, 73%), (2, 88%), (3, 95%), (5, 99%)"
)
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 start():
config = RankingConfig()
tf_util.update_memory(config.gpu_mem_fraction)
AIPlayer.create_if_nonexistant(config)
calc_ranking(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")
def start():
config = SelfPlayConfig()
tf_util.update_memory(config.gpu_mem_fraction)
util.set_low_process_priority()
AIPlayer.create_if_nonexistant(config)
run_games(config)