def play():
# These four players are playing the game
if variables.montecarlo:
players = [
AdvancedPlayer(),
MonteCarloPlayer(),
MinPlayer(),
RandomPlayer()
]
else:
players = [AdvancedPlayer(), MinPlayer(), MinPlayer(), RandomPlayer()]
# We are simulating n games accumulating a total score
nr_of_matches = variables.nr_of_matches
logging.debug('We are playing {} matches in total.'.format(nr_of_matches))
winning_count = [0, 0, 0, 0]
for match_nr in range(nr_of_matches):
scores = (0, 0, 0, 0)
logging.debug("--- MATCH {} ---".format(match_nr))
for game_nr in range(1, 5):
logging.debug("--- GAME {} ---".format(game_nr))
game = Game(players, game_nr % 4)
scores = tuple(sum(x) for x in zip(scores, game.play()))
logging.debug("--- Scores: {} ---".format(scores))
max_score = max(scores)
for i in range(4):
if scores[i] == max_score:
winning_count[i] += 1
logging.debug("--- Winning count: {} ---".format(winning_count))
def process_args(argv: List[str]) -> Tuple[Player, Player]:
"""Parse command line arguments to determine the players of the game.
"""
if len(argv) != 3:
player_1 = RandomPlayer()
player_2 = RandomPlayer()
else:
str_1 = sys.argv[1]
if str_1 == "-d":
player_1 = DeepQLearningPlayer()
elif str_1 == "-h":
player_1 = HumanPlayer()
elif str_1 == "-m":
player_1 = MinimaxPlayer()
else:
player_1 = RandomPlayer()
str_2 = sys.argv[2]
if str_2 == "-d":
player_2 = DeepQLearningPlayer()
elif str_2 == "-h":
player_2 = HumanPlayer()
elif str_2 == "-m":
player_2 = MinimaxPlayer()
else:
player_2 = RandomPlayer()
return player_1, player_2
def run_game(args):
logger.info('| process {} spawned'.format(args.i_))
seed = None
p0 = Winner(id=0,
seed=seed,
timeout_seconds=args.time_seconds_,
weights=args.weights_)
p1 = ExpectimaxWeightedProbabilitiesWithFilterPlayer(
id=1, seed=seed, timeout_seconds=args.time_seconds_)
p2 = RandomPlayer(id=2, seed=seed)
p3 = RandomPlayer(id=3, seed=seed)
# p3 = ExpectimaxWeightedProbabilitiesWithFilterPlayer(id=3, seed=seed, timeout_seconds=args.time_seconds_)
state = CatanState([p0, p1, p2, p3], seed)
count_moves = 0
while not state.is_final():
state.make_move(state.get_current_player().choose_move(state))
state.make_random_move()
count_moves += 1
scores = state.get_scores_by_player()
logger.info('| done iteration {}. scores: {}'.format(
args.i_, {
'p0 (new weights)': scores[p0],
'p1': scores[p1],
'p2': scores[p2],
'p3': scores[p3]
}))
# TODO: change this block
count_moves_factor = 1 * count_moves
p0_factor = 10000 if (scores[p0] >= 10) else 0
p_others_factor = (sum(scores.values()) - scores[p0]) * 0.2
res = p0_factor - (p_others_factor * count_moves_factor)
logger.info('| process {} done. res: {}'.format(args.i_, res))
return res
def eval_state(self, game: Game, p1, p2):
from players.random_player import RandomPlayer, WEIGHT_MAP26
p1 = RandomPlayer('rr1', w=WEIGHT_MAP26)
p2 = RandomPlayer('rr2', w=WEIGHT_MAP26)
players = [p1, p2]
for o_p, p in zip(game.players, players):
p.from_player(o_p)
g = copy.copy(game)
g.players = players
winner = g.run()
return 1 if winner is p1 else -1
def choose_move(self, state: CatanState):
best_move = self.montecarlo.get_best_move(state)
if best_move is not None:
return best_move
else:
logger.warning('returning a random move')
return RandomPlayer.choose_move(self, state)
def choose_action(self, b, p_other, actions):
results = {repr(a): Result() for a in actions}
me, other = RandomPlayerWithFirstAction(
'me', w=WEIGHT_MAP26), RandomPlayer('other', w=WEIGHT_MAP26)
new_players = [me, other]
if b.players[0] != self:
new_players.reverse()
for _ in range(100):
a = random.choice(actions)
me.first_action = a
game = copy.copy(b)
# copy player state
for o_p, p in zip(b.players, new_players):
p.from_player(o_p)
game.players = new_players
game.verbose = False
winner = game.run()
if winner.name == 'me':
results[repr(a)].W += 1
else:
results[repr(a)].L += 1
log.info('results= %s', results)
best = max(results.items(),
key=lambda item: item[1].W / (item[1].W + item[1].L))
for a in actions:
if repr(a) == best[0]:
return a
def choose_resources_to_drop(self) -> Dict[Resource, int]:
if sum(self.resources.values()) < 8:
return {}
resources_count = sum(self.resources.values())
resources_to_drop_count = ceil(resources_count / 2)
if self.can_settle_city() and resources_count >= sum(ResourceAmounts.city.values()) * 2:
self.remove_resources_and_piece_for_city()
resources_to_drop = copy.deepcopy(self.resources)
self.add_resources_and_piece_for_city()
elif self.can_settle_settlement() and resources_count >= sum(ResourceAmounts.settlement.values()) * 2:
self.remove_resources_and_piece_for_settlement()
resources_to_drop = copy.deepcopy(self.resources)
self.add_resources_and_piece_for_settlement()
elif (self.has_resources_for_development_card() and
resources_count >= sum(ResourceAmounts.development_card.values()) * 2):
self.remove_resources_for_development_card()
resources_to_drop = copy.deepcopy(self.resources)
self.add_resources_for_development_card()
elif self.can_pave_road() and resources_count >= sum(ResourceAmounts.road.values()) * 2:
self.remove_resources_and_piece_for_road()
resources_to_drop = copy.deepcopy(self.resources)
self.add_resources_and_piece_for_road()
else:
return RandomPlayer.choose_resources_to_drop(self)
resources_to_drop = [resource for resource, count in resources_to_drop.items() for _ in range(count)]
return Counter(self._random_choice(resources_to_drop, resources_to_drop_count, replace=False))
def __init__(self,
game_state,
player_hands,
allowed_actions,
player=RandomPlayer(),
ucb_const=1):
self.root = Node(None, None, game_state, player_hands, allowed_actions)
self.player = player
self.ucb_const = ucb_const
def get_player(self, picker_value):
if picker_value == 0:
return 'human'
if picker_value == 1:
return MinimaxPlayer()
if picker_value == 2:
return LearningPlayer()
if picker_value == 3:
return RandomPlayer()
def test():
data = get_dataset(RandomPlayer(), 100000, merge=True)
print(f"data_length: {len(data)}")
# data = pd.DataFrame(X)
# data['outcome'] = y
analyze(
lambda i: tree.DecisionTreeRegressor(criterion="mae",
min_samples_leaf=i), data)
analyze(lambda i: tree.DecisionTreeRegressor(criterion="mae", max_depth=i),
data)
def choose_move(self, state: CatanState):
self.expectimax_alpha_beta.start_turn_timer()
best_move, move, depth = None, None, 1
while not self.expectimax_alpha_beta.ran_out_of_time:
best_move = move
logger.info('starting depth {}'.format(depth))
move = self.expectimax_alpha_beta.get_best_move(state,
max_depth=depth)
depth += 2
if best_move is not None:
return best_move
else:
logger.warning('did not finish depth 1, returning a random move')
return RandomPlayer.choose_move(self, state)
def simple_match():
win = {'p1': 0, 'p2': 0}
for i in range(1):
p1 = RandomPlayer('p1')
p2 = MCSimplePlayer('p2', 100)
players = [p1, p2]
if i % 2:
players = players[::-1]
g = Game(players, seed=None)
winner = g.run()
win[winner.name] += 1
print(win)
print(win)
def receive_game_start_message(self, game_info):
self.my_model = MyModel()
nb_player = game_info['player_num']
max_round = game_info['rule']['max_round']
sb_amount = game_info['rule']['small_blind_amount']
ante_amount = game_info['rule']['ante']
self.emulator = Emulator()
self.emulator.set_game_rule(nb_player, max_round, sb_amount,
ante_amount)
#self.set_opponents_model(RandomModel())
self.set_opponents_model(RandomPlayer())
for player_info in game_info['seats']:
uuid = player_info['uuid']
player_model = self.my_model if uuid == self.uuid else self.opponents_model
self.emulator.register_player(uuid, player_model)
def raw_loop(screen):
screen.clear()
curses.curs_set(0)
init_colors()
empty_board(4).refresh()
game_end = False
env = Ludo(PLAYERS)
global agents
if agents == None:
agents = [IvanPesic(env) for i in range(PLAYERS)]
agents[0] = ReinforcePlayer(env, "players\saves\Reinforce30000-1.pth")
agents[1] = RLBasicPlayer(env, "players\saves\RLBasic30000-2.pth")
agents[3] = HumanPlayer(env)
if agents == None:
agents = [RandomPlayer() for i in range(PLAYERS)]
pstate = env.current_state()
while not game_end:
if isinstance(agents[env.current_player], HumanPlayer):
window = human_board(4, state, env, env.current_player)
window.addstr(2 * 4 + 5, 0, 'Igrac ')
window.addstr(2 * 4 + 5, 6, str(env.current_player + 1))
window.addstr(2 * 4 + 5, 8, 'je na potezu')
window.addstr(2 * 4 + 6, 0, 'Na kocki je bacen broj ')
window.addstr(2 * 4 + 6, 23, str(env.roll + 1))
while True:
window.refresh()
curses.napms(30)
c = window.getch()
action = agents[env.current_player].play(c)
if not action == -1:
break
window.addstr(2 * 4 + 7, 0, 'Morate odigrati validan potez')
else:
action = agents[env.current_player].play(pstate, TOKENS)
pstate, r, game_end = env.step(action)
state = env.current_state_as_tuple()
draw_board(4, state, env).refresh()
curses.napms(30)
curses.curs_set(1)
print('Player ', env.winning_player + 1, ' wins')
def raw_loop(screen):
env = Ludo(4)
screen.clear()
curses.curs_set(0)
init_colors()
empty_board(4).refresh()
curses.napms(100)
curses.curs_set(1)
game_end = False
agent1 = RandomPlayer()
agent2 = RandomPlayer()
state = env.current_state
while not game_end:
roll = random.randrange(1, 7)
if (env.current_player == 0): action = agent1.play(state, TOKENS)
if (env.current_player == 1): action = agent2.play(state, TOKENS)
state, r, game_end = env.step(roll, action)
draw_board(4, state, env).refresh()
curses.napms(30)
print('Player ', env.winning_player + 1, ' wins')
def main():
print("Cuda available: "+str(torch.cuda.is_available()))
#start tensorboard
tb = program.TensorBoard()
tb.configure(argv=[None, '--logdir', Settings.runs_folder])
tb.launch()
# set seed for debugging
if Settings.random_seed:
torch.manual_seed(Settings.random_seed)
#loading initial policy
hand_predictor = HandPredictor().to(Settings.device)
# take the newest generation available
i_episode = max_gen = 0
generations = [int(f[:8]) for f in listdir(Settings.checkpoint_folder) if f.endswith(".pt")]
if len(generations) > 0:
max_gen = max(generations)
hand_predictor.load_state_dict(torch.load(Settings.checkpoint_folder+"/" + str(max_gen).zfill(8) + ".pt"))
i_episode = max_gen
optimizer = torch.optim.Adam(hand_predictor.parameters(),lr=Settings.lr, betas=Settings.betas, weight_decay=Settings.optimizer_weight_decay)
# training loop
for _ in range(0, 90000000):
Settings.logger.info("playing " +str(Settings.update_games)+ " games")
smart_mcts_player = HPPIMCPlayer(30, 120, RandomPlayer(), hand_predictor)
# create four players
players = [smart_mcts_player, smart_mcts_player, smart_mcts_player, smart_mcts_player]
# create a game simulation
schafkopf_env = SchafkopfEnv(Settings.random_seed)
game_statistics = GameStatistics()
memory_states = []
memory_player_hands = []
# play a bunch of games
t0 = time.time()
for _ in range(Settings.update_games):
state, reward, terminal = schafkopf_env.reset()
while not terminal:
memory_states.append(hand_predictor.preprocess(state)) #TODO: happens twice now and could be optimized
memory_player_hands.append(hand_predictor.encode_player_hands(schafkopf_env.player_cards, state["game_state"].current_player))
action, prob = players[state["game_state"].current_player].act(state)
state, reward, terminal = schafkopf_env.step(action, prob)
if state["game_state"].game_type[1] == 2:
schafkopf_env.print_game()
print("game "+str(i_episode))
i_episode += 1
game_statistics.update_statistics(state["game_state"], reward)
t1 = time.time()
#update the policy
Settings.logger.info("updating policy")
# Create dataset from collected experiences
dataset = PredictionDatasetLSTM(memory_states, memory_player_hands)
training_generator = data.DataLoader(dataset, collate_fn=dataset.custom_collate,batch_size=Settings.mini_batch_size, shuffle=True)
#logging
avg_loss = 0
count = 0
hand_predictor.train()
for epoch in range(Settings.K_epochs): # epoch
mini_batches_in_batch = int(Settings.batch_size / Settings.mini_batch_size)
optimizer.zero_grad()
for i, (states, hands) in enumerate(training_generator): # mini batch
# Transfer to GPU
states = [state.to(Settings.device) for state in states]
hands = hands.to(Settings.device)
pred = hand_predictor(states)
#loss = nn.MSELoss()(pred, hands) #TODO: replace by cross entropy
loss = nn.BCELoss()(pred, hands)
avg_loss += loss.mean().item()
count +=1
loss.mean().backward()
if (i + 1) % mini_batches_in_batch == 0:
optimizer.step()
optimizer.zero_grad()
t2 = time.time()
hand_predictor.eval()
# writing game statistics for tensorboard
Settings.logger.info("Episode: "+str(i_episode) + " game simulation (s) = "+str(t1-t0) + " update (s) = "+str(t2-t1))
schafkopf_env.print_game()
game_statistics.write_and_reset (i_episode)
Settings.summary_writer.add_scalar('Loss/MSE_Loss', avg_loss / count, i_episode)
# save and evaluate the policy
Settings.logger.info("Saving Checkpoint")
torch.save(hand_predictor.state_dict(), Settings.checkpoint_folder + "/" + str(i_episode).zfill(8) + ".pt")
Settings.logger.info("Evaluation")
else:
expected_rewards = self.q_table[new_board_hash]
expected = reward + (0.9 * max(expected_rewards.values()))
change = 0.3 * (expected - self.q_table[board_hash][move])
self.q_table[board_hash][move] += change
def swap_players(p1, p2):
return p2, p1
if __name__ == '__main__':
qplayer = QPlayer(1, 3)
rplayer = RandomPlayer(-1)
winning_length = 3
games = 10000
results = {1: 0, 0: 0, -1: 0}
import tqdm
import pickle
# for i in tqdm.tqdm(range(games)):
# board = clean_board(3)
# winner = 0
# # print(i)
#
# while True:
# move = qplayer.get_move(board)
# qplayer.learn_q(board, move)
description="play a game of poker against the computer")
parser.add_argument("--name",
required=True,
help="your name to be displayed throughout the game")
parser.add_argument("--num-hands",
required=True,
help="number of hands you want to play",
type=int)
args = parser.parse_args()
deck = Deck()
computer_name = "Maniac"
human = HumanPlayer(args.name)
computer = RandomPlayer(computer_name)
starting_chips = 100
blind = 1
name_width = max(len(args.name), len(computer_name))
rank_width = max(len(str(x)) for x in HandRank)
action_width = max(len(str(x)) for x in Action)
winner_template = "{{:{}}} wins the pot of: {{}}".format(name_width)
show_template = "{{:{}}} shows cards: {{}}, hand: {{:{}}}, {{}}".format(
name_width, rank_width)
action_template = "Pot: {{:3}}, Player: {{:{}}}, Action: {{:{}}}, Size: {{}}".format(
name_width, action_width)
human_results = 0
def execute_game(plot_map=True):
seed = None
p0 = MonteCarloWithFilterPlayer(seed)
p1 = RandomPlayer(seed)
players = [p0, p1]
state = CatanState(players, seed)
turn_count = 0
score_by_player = state.get_scores_by_player()
if plot_map:
state.board.plot_map('turn_{}_scores_{}.png'.format(
turn_count,
''.join('{}_'.format(v) for v in score_by_player.values())))
while not state.is_final():
# noinspection PyProtectedMember
logger.info(
'----------------------p{}\'s turn----------------------'.format(
state._current_player_index))
turn_count += 1
robber_placement = state.board.get_robber_land()
move = state.get_current_player().choose_move(state)
assert not scores_changed(state, score_by_player,
state.get_scores_by_player())
state.make_move(move)
state.make_random_move()
score_by_player = state.get_scores_by_player()
move_data = {
k: v
for k, v in move.__dict__.items()
if (v and k != 'resources_updates')
and not (k == 'robber_placement_land' and v == robber_placement)
and not (isinstance(v, dict) and sum(v.values()) == 0)
}
logger.info('| {}| turn: {:3} | move:{} |'.format(
''.join('{} '.format(v) for v in score_by_player.values()),
turn_count, move_data))
if plot_map and (turn_count == 4 or turn_count % 50 == 0):
image_name = 'turn_4_scores_{}.png'.format(
turn_count,
''.join('{}_'.format(v) for v in score_by_player.values()))
state.board.plot_map(image_name, state.current_dice_number)
if plot_map:
state.board.plot_map('turn_{}_scores_{}.png'.format(
turn_count,
''.join('{}_'.format(v) for v in score_by_player.values())))
players_scores_by_names = {(k, v.__class__): score_by_player[v]
for k, v in locals().items() if v in players}
fileLogger.info('\n' + '\n'.join(
' {:150} : {} '.format(str(name), score)
for name, score in players_scores_by_names.items()) +
'\n turns it took: {}\n'.format(turn_count) + ('-' * 156))
p0_type = type(p0).__name__
p_others_type = type(p1).__name__
results = defaultdict(int)
for _ in tqdm(range(0, repetitions)):
result = play(playerX, playerO)
# result_text = 'x' if result == 1 else ('o' if result == -1 else 'draw')
results[result] += 1
all = sum(results.values())
x = results[1]
o = results[-1]
d = results[0]
print(
f"\n{playerX.__class__.__name__} as X vs. {playerO.__class__.__name__} as O"
)
print(f"X won {x} times {(x / all) * 100}%")
print(f"O won {o} times {(o / all) * 100}%")
print(f"draw {d} times {(d / all) * 100}%")
random_player = RandomPlayer()
win_selecting_player = WinSelectingPlayer()
q_player = QPlayer(win_selecting_player)
cart_player = CartPlayer(100000, win_selecting_player)
print("TESTING...")
test(win_selecting_player, cart_player)
test(win_selecting_player, q_player)
test(cart_player, random_player)
test(random_player, cart_player)
test(q_player, RandomPlayer())
test(RandomPlayer(), q_player)
def execute_game(plot_map=True):
seed = None
timeout_seconds = TIME_OUT
p0 = Winner(id=0, seed=seed, timeout_seconds=timeout_seconds)
p1 = ExpectimaxWeightedProbabilitiesWithFilterPlayer(id=1, seed=seed, timeout_seconds=timeout_seconds)
p2 = RandomPlayer(id=2)
p3 = RandomPlayer(id=3)
players = [p0, p1, p2, p3]
state = CatanState(players, seed)
turn_count = 0
score_by_player = state.get_scores_by_player_indexed()
while not state.is_final():
# noinspection PyProtectedMember
logger.info('----------------------p{}\'s turn----------------------'.format(state._current_player_index))
turn_count += 1
robber_placement = state.board.get_robber_land()
move = state.get_current_player().choose_move(state)
assert not scores_changed(state, score_by_player, state.get_scores_by_player_indexed())
state.make_move(move)
state.make_random_move()
score_by_player = state.get_scores_by_player_indexed()
move_data = {k: v for k, v in move.__dict__.items() if (v and k != 'resources_updates') and not
(k == 'robber_placement_land' and v == robber_placement) and not
(isinstance(v, dict) and sum(v.values()) == 0)}
logger.info('| {}| turn: {:3} | move:{} |'.format(''.join('{} '.format(v) for v in score_by_player),
turn_count, move_data))
if plot_map:
image_name = 'turn_{}_scores_{}.png'.format(
turn_count, ''.join('{}_'.format(v) for v in score_by_player))
state.board.plot_map(image_name, state.current_dice_number)
players_scores_by_names = {(k, v.__class__, v.expectimax_alpha_beta.evaluate_heuristic_value.__name__ if (
isinstance(v, ExpectimaxBaselinePlayer)) else None): score_by_player[v.get_id()]
for k, v in locals().items() if v in players
}
names = list(players_scores_by_names.keys())
names.sort()
fileLogger.info('\n' + '\n'.join(' {:80} : {} '.format(str(name), players_scores_by_names[name])
for name in names) +
'\n turns it took: {}\n'.format(turn_count) + ('-' * 156))
p0_type = type(p0).__name__
p_others_type = type(p1).__name__
global excel_file_name
excel_file_name = '{}_vs_{}_timeout_{}_seed_{}.xlsx'.format(p0_type, p_others_type, timeout_seconds, seed,
int(time.time()))
excel_data_grabber(score_by_player[0], score_by_player[1], score_by_player[2], score_by_player[3], turn_count,
p0_type, p_others_type)
excel_output = ""
for i in range(len(players)):
player_output = str(players[i]) + "@" + str(score_by_player[i])
excel_output += player_output + "\n"
# fileLogger.info("|\n#" + str(turn_count) + "\n" + excel_output)
if score_by_player[0] >= 10:
return 1
else:
return 0
from pypokerengine.api.game import setup_config, start_poker from players.fish_player import FishPlayer from players.console_player import ConsolePlayer from players.random_player import RandomPlayer config = setup_config(max_round=100, initial_stack=1000, small_blind_amount=20) config.register_player(name="f1", algorithm=FishPlayer()) config.register_player(name="r1", algorithm=RandomPlayer()) config.register_player(name="c1", algorithm=ConsolePlayer()) game_result = start_poker(config, verbose=1)
def execute_game(i, iterations, plot_map=True):
seed = None
timeout_seconds = 5
p0 = MCTSPlayer(0, iterations=iterations)
p1 = RandomPlayer(1)
p2 = RandomPlayer(2)
p3 = RandomPlayer(3)
players = [p0, p1, p2, p3]
state = CatanState(players, seed)
turn_count = 0
score_by_player = state.get_scores_by_player_indexed()
# if plot_map:
# state.board.plot_map('turn_{}_scores_{}.png'
# .format(turn_count, ''.join('{}_'.format(v) for v in score_by_player.values())))
while not state.is_final():
# noinspection PyProtectedMember
logger.info(
'----------------------p{}\'s turn----------------------'.format(
state._current_player_index))
turn_count += 1
robber_placement = state.board.get_robber_land()
move = state.get_current_player().choose_move(state)
assert not scores_changed(state, score_by_player,
state.get_scores_by_player_indexed())
state.make_move(move)
state.make_random_move()
score_by_player = state.get_scores_by_player_indexed()
move_data = {
k: v
for k, v in move.__dict__.items()
if (v and k != 'resources_updates')
and not (k == 'robber_placement_land' and v == robber_placement)
and not (isinstance(v, dict) and sum(v.values()) == 0)
}
logger.info('| {}| turn: {:3} | move:{} |'.format(
''.join('{} '.format(v) for v in score_by_player), turn_count,
move_data))
# if plot_map:
# image_name = 'turn_{}_scores_{}.png'.format(
# turn_count, ''.join('{}_'.format(v) for v in score_by_player))
# state.board.plot_map(image_name, state.current_dice_number)
players_scores_by_names = {
(k, v.__class__,
v.expectimax_alpha_beta.evaluate_heuristic_value.__name__ if
(isinstance(v, ExpectimaxBaselinePlayer)) else None):
score_by_player[v.get_id()]
for k, v in locals().items() if v in players
}
fileLogger.info('\n' + '\n'.join(
' {:80} : {} '.format(str(name), score)
for name, score in players_scores_by_names.items()) +
'\n turns it took: {}\n'.format(turn_count) +
'game num: {}, num iterations: {}'.format(i, iterations) +
'\n' + ('-' * 156))
p0_type = type(p0).__name__
p_others_type = type(p1).__name__
from core import Game from players.trivial_player import TrivialPlayer from players.random_player import RandomPlayer players = [TrivialPlayer(), TrivialPlayer(), RandomPlayer()] result = Game(players).play(with_prints=True)
def main():
pimc_player = PIMCPlayer(10, 40, RandomPlayer())
policy = ActorCriticNetworkLSTM().to(Settings.device)
policy.load_state_dict(torch.load("../policies/pretrained/lstm-policy.pt"))
rl_player = RlPlayer(policy, action_shaping=False, eval=True)
hp = HandPredictor().to(Settings.device)
hp.load_state_dict(torch.load("../policies/pretrained/hand-predictor.pt"))
smart_pimc_player = HPPIMCPlayer(10, 40, RandomPlayer(),
HandPredictor().to(Settings.device))
ip = ImmitationPolicy().to(Settings.device)
ip.load_state_dict(torch.load("../policies/00010340.pt"))
immitation_player = RlPlayer(ip, action_shaping=False, eval=True)
participants = [
rl_player,
immitation_player,
smart_pimc_player,
pimc_player,
RuleBasedPlayer(),
RandomCowardPlayer(),
RandomPlayer(),
]
number_of_games = 1000
for i in range(len(participants)):
for j in range(i + 1, len(participants)):
p1 = participants[i]
p2 = participants[j]
cummulative_reward = [0, 0, 0, 0]
for k in range(
2
): #run the same tournament twice with differen positions of players
print(' ')
schafkopf_env = SchafkopfEnv(seed=1)
if k == 0:
players = [p1, p1, p2, p2]
else:
players = [p2, p2, p1, p1]
cummulative_reward.reverse()
# tournament loop
for game_nr in range(1, number_of_games + 1):
state, reward, terminal = schafkopf_env.reset()
while not terminal:
action, prob = players[
state["game_state"].current_player].act(state)
state, reward, terminal = schafkopf_env.step(
action, prob)
cummulative_reward = [
cummulative_reward[m] + reward[m] for m in range(4)
]
if game_nr % 100 == 0:
print('.', end='')
#schafkopf_env.print_game()
print("player " + str(i) + " vs. player " + str(j) + " = " +
str((cummulative_reward[2] + cummulative_reward[3]) /
(2 * 2 * number_of_games)) + " to " +
str((cummulative_reward[0] + cummulative_reward[1]) /
(2 * 2 * number_of_games)))
def parse_args():
parser = argparse.ArgumentParser(
description="Play different players against each other.")
parser.add_argument("--infile",
type=str,
required=True,
help="Input file name.")
return parser.parse_args()
if __name__ == "__main__":
args = parse_args()
games = []
with gzip.open(args.infile, 'r') as f:
for line in f:
games.append(json.loads(line.decode()))
player_1 = HeuristicPlayer()
player_2 = HeuristicPlayer()
player_3 = RandomPlayer()
score = {0: 0, 1: 0, 2: 0}
for init_game in games:
g = Game(init_game, [player_1, player_2, player_3])
winners = g.play()
for w in winners:
score[w] += 1
print(score)
points += len(diagonal)
# 2nd diagonal
new_y = (BOARD_SIZE - 1) - y
k = new_y - x
diagonal = np.diag(np.fliplr(board), k)
if np.all(diagonal) and len(diagonal) > 1:
points += len(diagonal)
return points
if __name__ == '__main__':
p1 = HumanPlayer('p1')
# p2 = HumanPlayer('p2')
p2 = RandomPlayer('p2')
g = Game(3, p1, p2)
g.play()
# player1_wins = []
# # print(0, end='')
# for i in range(1000):
# # print('\r', (i+1)/1000, end='')
# player1_wins.append(g.play())
# # time.sleep(0.1)
# sys.stdout.write("\r%d%%" % (i / 1000 * 100))
# sys.stdout.flush()
#
# print()
# print(sum(player1_wins) / len(player1_wins) * 100, '%')
boards_batch, moves_batch, rewards_batch = [], [], []
if episode_number % log_every == 0:
print("episode: %s win_rate: %s" %
(episode_number, _win_rate(log_every, results)))
def _win_rate(print_results_every, results):
i = sum(results)
every___ = (print_results_every * 2.)
return 0.5 + i / every___
def normalize_rewards(rewards_batch):
normalized_rewards = rewards_batch - np.mean(rewards_batch)
rewards_std = np.std(normalized_rewards)
if rewards_std != 0:
normalized_rewards /= rewards_std
else:
print("warning: got mini batch std of 0.")
return normalized_rewards
if __name__ == '__main__':
train_policy_gradients(layers=[9, 100, 100, 100, 9],
learning_rate=1e-4,
batch_size=100,
games=100000,
log_every=1000,
opponent=RandomPlayer(-1),
winning_length=3)
"""Connect Games"""
from players.random_player import RandomPlayer
from connect_games.gomoku import Gomoku
from connect_games.tictactoe import TicTacToe
from game_manager import GameManager
if __name__ == '__main__':
play = "TicTacToe"
game_dict = {"Gomoku": Gomoku, "TicTacToe": TicTacToe}
if play in game_dict:
game = game_dict[play]
else:
print("Game not available!")
exit()
manager = GameManager(game)
manager.run_experiment(2000, RandomPlayer(), RandomPlayer(), render=False)
from players.honest_player_og import HonestPlayer
from players.qlearner import RLPlayer
from players.random_player import RandomPlayer
from players.risky_player import RiskyPlayer
from players.call_player import CallPlayer
from players.bluff_player import BluffPlayer
import pandas as pd
# 1. Set game settings on emulator
n_players = 4
j=0
emulator = Emulator()
#quuid = "uuid-q"
p_uuid=["A1","A2","A3","A4"]
#qlearner_player = RLPlayer(n_players, quuid)
monte_carlos_tries=[[RandomPlayer(p_uuid[0]),RiskyPlayer()],[RiskyPlayer(),CallPlayer()],[RiskyPlayer(),BluffPlayer()]]
names=[['Rand','Risk'],['Call','Risk'],['Bluff','Risk']]
for tryM in monte_carlos_tries:
df = pd.DataFrame(columns = ['uuid', 'stack', 'game'])
for i in range(0,1):
print("starting game " + str(i) + " from try " + str(j))
df1 = pd.DataFrame(columns = ['uuid', 'stack', 'round'])
df2 = pd.DataFrame(columns = ['uuid', 'stack', 'round'])
df3 = pd.DataFrame(columns = ['uuid', 'stack', 'round'])
df4 = pd.DataFrame(columns = ['uuid', 'stack', 'round'])
emulator.register_player(uuid=p_uuid[0], player=tryM[0])
emulator.register_player(uuid=p_uuid[1], player=tryM[1])
emulator.set_game_rule(player_num=2, max_round=1000, small_blind_amount=20, ante_amount=0)
# 2. Setup GameState object
players_info = {
p_uuid[0]: { "name": "player1", "stack": 10000 },
