def test_jouer_coup(self):
""" Test la fonction jouer_coup
Cas à tester:
- La fonction joue le bon coup automatiquement
- QuoridorError si le numéro du joueur est invalide
- QuoridorError si la partie est déjà terminée
"""
partie_terminee_etat = {
"joueurs": [{
"nom": "joueur1",
"murs": 7,
"pos": (5, 9)
}, {
"nom": "joueur2",
"murs": 3,
"pos": (6, 9)
}],
"murs": {
"horizontaux": [(4, 4), (2, 6), (3, 8), (5, 8), (7, 8)],
"verticaux": [(6, 2), (4, 4), (2, 5), (7, 5), (7, 7)]
}
}
# Tester l'erreur soulevée lorsqu'on donne un joueur invalide
jeu_nouveau = Quoridor(["joueur1", "joueur2"])
self.assertRaisesRegex(QuoridorError, "joueur invalide!",
jeu_nouveau.jouer_coup, 5)
# tester l'erreur soulevée l'orsqu'on cherche à jouer un
# coup alors que la partie est déjà terminée
jeu_termine = Quoridor(partie_terminee_etat['joueurs'],
partie_terminee_etat['murs'])
self.assertRaisesRegex(QuoridorError, "La partie est déjà terminée!",
jeu_termine.jouer_coup, 1)
def __init__(self, init_model=None):
# 棋盘参数
self.game = Quoridor()
# 训练参数
self.learn_rate = 2e-3
self.lr_multiplier = 1.0 # 适应性调节学习速率
self.temp = 1.0
self.n_playout = 400
self.c_puct = 5
self.buffer_size = 10000
self.batch_size = 128 # 取1 测试ing
self.data_buffer = deque(maxlen=self.buffer_size)
self.play_batch_size = 1
self.epochs = 5
self.kl_targ = 0.02
self.check_freq = 50
self.game_batch_num = 1500
self.best_win_ratio = 0.0
self.pure_mcts_playout_num = 1000
if init_model:
self.policy_value_net = PolicyValueNet(model_file=init_model)
else:
self.policy_value_net = PolicyValueNet()
# 设置电脑玩家信息
self.mcts_player = MCTSPlayer(self.policy_value_net.policy_value_fn,
c_puct=self.c_puct,
n_playout=self.n_playout,
is_selfplay=1)
def __init__(self, init_model=None):
self.game = Quoridor()
self.learn_rate = 2e-3
self.lr_multiplier = 1.0
self.temp = 1.0
self.n_playout = 200
self.c_puct = 5
self.buffer_size = 10000
self.data_buffer = deque(maxlen=self.buffer_size)
self.play_batch_size = 1
self.kl_targ = 0.02
self.check_freq = 10
self.game_batch_num = 1000
self.best_win_ratio = 0.0
self.pure_mcts_playout_num = 1000
self.old_probs = 0
self.new_probs = 0
self.first_trained = False
if init_model:
self.policy_value_net = PolicyValueNet(model_file=init_model)
else:
self.policy_value_net = PolicyValueNet()
self.mcts_player = MCTSPlayer(self.policy_value_net.policy_value_fn, c_puct=self.c_puct,
n_playout=self.n_playout, is_selfplay=1)
def __init__(self, game_state:Quoridor, policy_output, value_output):
# _counts is the number of times we've taken some action *from this state*. Initialized to all zeros. Stored
# as a torch tensor over all possible actions, to be later masked with the set of legal actions
self._counts = torch.zeros(3, 9, 9)
self._total_reward = torch.zeros(3, 9, 9)
self._policy = policy_output
self._value = value_output
self._legal_mask = encode_actions_to_planes(game_state.all_legal_moves(), game_state.current_player)
self._player = game_state.current_player
self._key = game_state.hash_key()
self._children = {}
self.__flagged = False
def setUp(self):
self.partie = Quoridor(
[{
"nom": "idul",
"murs": 7,
"pos": [5, 5]
}, {
"nom": "automate",
"murs": 3,
"pos": [8, 6]
}],
murs={
"horizontaux": [[4, 4], [2, 6], [3, 8], [5, 8], [7, 8]],
"verticaux": [[6, 2], [4, 4], [2, 6], [7, 5], [7, 7]]
})
def mode_automatique(idul):
"""mode automatique"""
[identifiant, état] = initialiser_partie(idul)
print(afficher_damier_ascii(état))
joueur = [état['joueurs'][0]['nom'], état['joueurs'][1]['nom']]
jeu = Quoridor(joueur)
état = jeu.état_partie()
while 1:
(coup, pos) = jeu.jouer_coup(1)
print(jeu)
état = jouer_coup(identifiant, coup, tuple(pos))
afficher_damier_ascii(état)
joueur1 = état['joueurs']
murs_j1 = état['murs']
jeu = Quoridor(joueur1, murs_j1)
def test_init_with_state(self):
# Initiation d'une partie avec une liste de dictionnaires
partie = Quoridor(
[{
"nom": "idul",
"murs": 7,
"pos": [5, 5]
}, {
"nom": "automate",
"murs": 3,
"pos": [8, 6]
}],
murs={
"horizontaux": [[4, 4], [2, 6], [3, 8], [5, 8], [7, 8]],
"verticaux": [[6, 2], [4, 4], [2, 6], [7, 5], [7, 7]]
})
self.assertEqual(partie.joueurs, [
{
'nom': 'idul',
'murs': 7,
'pos': [5, 5]
},
{
'nom': 'automate',
'murs': 3,
'pos': [8, 6]
},
])
self.assertEqual(
partie.murs, {
"horizontaux": [[4, 4], [2, 6], [3, 8], [5, 8], [7, 8]],
"verticaux": [[6, 2], [4, 4], [2, 6], [7, 5], [7, 7]]
})
def main():
"""Boucle principale."""
args = analyser_commande()
if args.lister:
for partie in api.lister_parties(args.idul):
print(partie["id"])
return
id_partie, partie = api.débuter_partie(args.idul)
gagnant = False
q = None
while not gagnant:
if args.mode_graphique:
q = QuoridorX(partie["joueurs"], partie["murs"])
else:
q = Quoridor(partie["joueurs"], partie["murs"])
gagnant = q.partie_terminée()
if gagnant:
break
if args.mode_graphique:
q.afficher()
else:
print("", q, sep="\n")
partie = jouer_coup(args, q, id_partie)
if args.mode_graphique:
turtle.mainloop()
else:
print("", q, "", f'{gagnant} a gagné la partie!', "", sep="\n")
def __init__(self, init_model=None):
# params of the board and the game
self.board_width = 6
self.board_height = 6
self.n_in_row = 4
self.board = Quoridor()
self.game = Game(self.board)
# training params
self.learn_rate = 2e-3
self.lr_multiplier = 1.0 # adaptively adjust the learning rate based on KL
self.temp = 1.0 # the temperature param
self.n_playout = 1000 # num of simulations for each move
self.c_puct = 5
self.buffer_size = 10000
self.batch_size = 1000 # mini-batch size for training
self.data_buffer = deque(maxlen=self.buffer_size)
self.play_batch_size = 64
self.epochs = 5 # num of train_steps for each update
self.kl_targ = 0.02
self.check_freq = 50
self.game_batch_num = 2000
self.best_win_ratio = 0.0
# num of simulations used for the pure mcts, which is used as
# the opponent to evaluate the trained policy
self.pure_mcts_playout_num = 1000
if init_model:
# start training from an initial policy-value net
self.policy_value_net = PolicyValueNet(model_file=init_model, use_gpu=True)
else:
# start training from a new policy-value net
self.policy_value_net = PolicyValueNet()
self.mcts_player = MCTSPlayer(self.policy_value_net.policy_value_fn,
c_puct=self.c_puct,
n_playout=self.n_playout,
is_selfplay=1)
def test_partie_terminée(self):
""" Test de la fonction partie_terminée
Cas à tester:
- La fonction retourne False si la partie n'est pas terminée
- La fonction retourne le nom du joueur qui a gagné si la partie est terminée
"""
partie_terminee1_etat = {
"joueurs": [{
"nom": "joueur1",
"murs": 7,
"pos": (5, 9)
}, {
"nom": "joueur2",
"murs": 3,
"pos": (6, 9)
}],
"murs": {
"horizontaux": [(4, 4), (2, 6), (3, 8), (5, 8), (7, 8)],
"verticaux": [(6, 2), (4, 4), (2, 5), (7, 5), (7, 7)]
}
}
partie_terminee2_etat = {
"joueurs": [{
"nom": "joueur1",
"murs": 7,
"pos": (6, 1)
}, {
"nom": "joueur2",
"murs": 3,
"pos": (5, 1)
}],
"murs": {
"horizontaux": [[4, 4], [2, 6], [3, 8], [5, 8], [7, 8]],
"verticaux": [[6, 2], [4, 4], [2, 5], [7, 5], [7, 7]]
}
}
# Tester que la fonction retourne False si la partie n'est pas terminée
jeu_pas_fini = Quoridor(['joueur1', 'joueur2'])
self.assertEqual(jeu_pas_fini.partie_terminée(), False)
# Tester que la fonction retourne le nom du joueur qui a gagné quand c'est le cas
jeu_fini1 = Quoridor(partie_terminee1_etat['joueurs'],
partie_terminee1_etat['murs'])
self.assertEqual(jeu_fini1.partie_terminée(), "joueur1")
jeu_fini2 = Quoridor(partie_terminee2_etat['joueurs'],
partie_terminee2_etat['murs'])
self.assertEqual(jeu_fini2.partie_terminée(), "joueur2")
def _single_search(self, game:Quoridor, c_puct, verbose=False) -> float:
"""Recursively run a single MCTS thread out from the given state using exploration parameter 'c_puct'.
"""
node = self._node_lookup[game.hash_key()]
action = sample_action(node.upper_conf(c_puct), node._player, temperature=0.0)
if verbose:
print("\tsingle_search starting @", node, "\n\t\ttaking", action, end="")
with game.temp_move(action):
winner = game.get_winner()
if winner is not None:
if verbose:
print("--> winner is", winner)
# Case 1: 'action' ended the game. Return +1 if a win from the perspective of whoever played the move
backup_val = +1 if winner == node._player else -1
elif game.hash_key() not in self._node_lookup:
# Case 2: 'action' resulted in a state we've never seen before. Create a new node and return
pol, val = self.pol_val_fun(game)
new_node = TreeNode(game, pol, val)
self._node_lookup[game.hash_key()] = new_node
node.add_child(action, new_node)
if verbose:
print("--> leaf <{}> with value".format(str(new_node)), val)
# "val" is from the perspective of "new_node" but we're evaluating "node". Flip sign for minmax.
backup_val = -val
else:
# Case 3: we've seen this state before. But it's possible we're reaching it from a different history.
# Ensure the parent/child relationship exists then recurse, flipping the sign of the child node's value.
if verbose:
print("--> recursing to node", self._node_lookup[game.hash_key()])
node.add_child(action, self._node_lookup[game.hash_key()])
backup_val = -self._single_search(game, c_puct, verbose=verbose)
# Apply backup
node.backup(action, backup_val)
return backup_val
def create_board(self_loc, oppo_loc, self_walls, oppo_walls, walls):
q = Quoridor()
q._self_loc = self_loc
q._oppo_loc = oppo_loc
q.wall_remaining[1] = self_walls
q.wall_remaining[-1] = oppo_walls
q._walls[:] = walls[:]
return q
def pos_mur_erroné():
return Quoridor([{
'nom': 'Simon',
'murs': 10,
'pos': (10, 1)
}, 'Robot'],
murs={
'horizontaux': [(3, 5)],
'verticaux': [(4, 4)]
})
class TestQuoridorMethods(unittest.TestCase):
def setUp(self):
self.partie = Quoridor(
[{
"nom": "idul",
"murs": 7,
"pos": [5, 5]
}, {
"nom": "automate",
"murs": 3,
"pos": [8, 6]
}],
murs={
"horizontaux": [[4, 4], [2, 6], [3, 8], [5, 8], [7, 8]],
"verticaux": [[6, 2], [4, 4], [2, 6], [7, 5], [7, 7]]
})
def test_état_partie(self):
état = self.partie.état_partie()
self.assertEqual(
état, {
"joueurs": [{
"nom": "idul",
"murs": 7,
"pos": [5, 5]
}, {
"nom": "automate",
"murs": 3,
"pos": [8, 6]
}],
"murs": {
"horizontaux": [[4, 4], [2, 6], [3, 8], [5, 8], [7, 8]],
"verticaux": [[6, 2], [4, 4], [2, 6], [7, 5], [7, 7]]
}
})
def test_placer_mur_exceptions(self):
def joueur_erroné():
self.partie.placer_mur(joueur=0,
position=(4, 5),
orientation='horizontal')
self.assertRaises(QuoridorError, joueur_erroné)
def __init__(self, init_model=None, first_player=1):
self.game = Quoridor()
self.temp = 1.0
self.c_puct = 5
self.play_batch_size = 1
self.alpha_playout = 100
self.pure_playout = 100
self.first = first_player
self.alpha_player = A_Player(
PolicyValueNet(model_file=init_model).policy_value_fn,
c_puct=5,
n_playout=self.alpha_playout,
is_selfplay=0)
self.pure_player = B_Player(c_puct=5, n_playout=self.pure_playout) #
self.alpha_win_total = 0
self.alpha_win_first = 0
self.alpha_draw_total = 0
self.alpha_draw_first = 0
def automatique(idul):
"""Section automatique"""
identifiant, etat = débuter_partie(idul)
partie = Quoridor(etat["joueurs"], etat['murs'])
print(partie)
while not partie.partie_terminée():
before = copy.deepcopy(partie.état_partie())
partie.jouer_coup(1)
after = copy.deepcopy(partie.état_partie())
print(partie)
if before["joueurs"][0]["pos"] != after["joueurs"][0]["pos"]:
etat = jouer_coup(identifiant, "D", after["joueurs"][0]["pos"])
elif len(after["murs"]["horizontaux"]) != len(
before["murs"]["horizontaux"]):
etat = jouer_coup(
identifiant, "MH",
after["murs"]["horizontaux"][len(after["murs"]["horizontaux"])
- 1])
elif len(after["murs"]["verticaux"]) != len(
before["murs"]["verticaux"]):
etat = jouer_coup(
identifiant, "MV",
after["murs"]["verticaux"][len(after["murs"]["verticaux"]) -
1])
partie = Quoridor(etat["joueurs"], etat['murs'])
def test_playout():
from quoridor import Quoridor
from policy_value_net import PolicyValueNet
c_puct = 5
n_playout = 400
policy_value_net = PolicyValueNet(model_file=None, use_gpu=True)
mcts = MCTS(policy_value_net.policy_value_fn,
c_puct=c_puct,
n_playout=n_playout)
q = Quoridor()
acts, act_probs = mcts.get_move_probs(q)
print(acts)
print(act_probs)
def test_init_with_strings(self):
partie = Quoridor(['Simon', 'Robot'])
self.assertEqual(
partie.joueurs,
[
{
'nom': 'Simon',
'murs': 10,
'pos': (5, 1)
},
{
'nom': 'Robot',
'murs': 10,
'pos': (5, 9)
},
],
)
self.assertEqual(partie.murs, {'horizontaux': [], 'verticaux': []})
def test_état_partie(self):
""" Test la fonction état_partie
Cas à tester:
- La fonction retourne le bon résultat
"""
nouvelle_partie_etat = {
"joueurs": [{
"nom": "joueur1",
"murs": 10,
"pos": (5, 1)
}, {
"nom": "joueur2",
"murs": 10,
"pos": (5, 9)
}],
"murs": {
"horizontaux": [],
"verticaux": []
}
}
# Tester si la fonction retourne la bonne affichage
nouvellepartie = Quoridor(["joueur1", "joueur2"])
self.assertEqual(nouvellepartie.état_partie(), nouvelle_partie_etat)
def test_mcts_player():
from quoridor import Quoridor
from policy_value_net import PolicyValueNet
policy_value_net = PolicyValueNet(model_file=None)
c_puct = 5
n_playout = 400
mcts_player = MCTSPlayer(policy_value_net.policy_value_fn,
c_puct=c_puct,
n_playout=n_playout,
is_selfplay=1)
q = Quoridor()
action = mcts_player.get_action(q)
q.take_action(action)
q.alter()
# print(q.valid_actions())
# try:
action = mcts_player.get_action(q)
# except:
# print(q.valid_actions())
print('take action : ', action)
class TrainPipeline(object):
def __init__(self, init_model=None):
# 棋盘参数
self.game = Quoridor()
# 训练参数
self.learn_rate = 2e-3
self.lr_multiplier = 1.0 # 适应性调节学习速率
self.temp = 1.0
self.n_playout = 400
self.c_puct = 5
self.buffer_size = 10000
self.batch_size = 128 # 取1 测试ing
self.data_buffer = deque(maxlen=self.buffer_size)
self.play_batch_size = 1
self.epochs = 5
self.kl_targ = 0.02
self.check_freq = 50
self.game_batch_num = 1500
self.best_win_ratio = 0.0
self.pure_mcts_playout_num = 1000
if init_model:
self.policy_value_net = PolicyValueNet(model_file=init_model)
else:
self.policy_value_net = PolicyValueNet()
# 设置电脑玩家信息
self.mcts_player = MCTSPlayer(self.policy_value_net.policy_value_fn,
c_puct=self.c_puct,
n_playout=self.n_playout,
is_selfplay=1)
# def get_equi_data(self, play_data):
# """
# 数据集增强，获取旋转后的数据，因为五子棋也是对称的
# play_data: [(state, mcts_prob, winner_z), ..., ...]"""
# extend_data = []
# for state, mcts_porb, winner in play_data:
# equi_state = np.array([np.rot90(s,2) for s in state])
# equi_mcts_prob = np.rot90(np.flipud(mcts_porb.reshape(9, 9)), 2)
# extend_data.append((equi_state, np.flipud(equi_mcts_prob).flatten(), winner))
# # flip horizontally
# equi_state = np.array([np.fliplr(s) for s in equi_state])
# equi_mcts_prob = np.fliplr(equi_mcts_prob)
# extend_data.append((equi_state, np.flipud(equi_mcts_prob).flatten(), winner))
# return extend_data
def collect_selfplay_data(self, n_games=1):
"""收集训练数据"""
for i in range(n_games):
winner, play_data = self.game.start_self_play(
self.mcts_player, temp=self.temp) # 进行自博弈
play_data = list(play_data)[:]
self.episode_len = len(play_data)
# 数据增强
# play_data = self.get_equi_data(play_data)
self.data_buffer.extend(play_data)
def policy_update(self):
"""训练策略价值网络"""
mini_batch = random.sample(self.data_buffer,
self.batch_size) # 获取mini-batch
state_batch = [data[0] for data in mini_batch] # 提取第一位的状态
mcts_probs_batch = [data[1] for data in mini_batch] # 提取第二位的概率
winner_batch = [data[2] for data in mini_batch] # 提取第三位的胜负情况
old_probs, old_v = self.policy_value_net.policy_value(
state_batch) # 输入网络计算旧的概率和胜负价值，这里为什么要计算旧的数据是因为需要计算
# 新旧之间的KL散度来控制学习速率的退火
# 开始训练epochs个轮次
for i in range(self.epochs):
loss, entropy = self.policy_value_net.train_step(
state_batch, mcts_probs_batch, winner_batch,
self.learn_rate * self.lr_multiplier)
new_probs, new_v = self.policy_value_net.policy_value(
state_batch) # 计算新的概率和价值
kl = np.mean(
np.sum(old_probs *
(np.log(old_probs + 1e-10) - np.log(new_probs + 1e-10)),
axis=1))
if kl > self.kl_targ * 4: # 如果KL散度发散的很不好，就提前结束训练
break
# 根据KL散度，适应性调节学习速率
if kl > self.kl_targ * 2 and self.lr_multiplier > 0.1:
self.lr_multiplier /= 1.5
elif kl < self.kl_targ / 2 and self.lr_multiplier < 10:
self.lr_multiplier *= 1.5
explained_var_old = 1 - np.var(
np.array(winner_batch) - old_v.flatten()) / np.var(
np.array(winner_batch))
explained_var_new = 1 - np.var(
np.array(winner_batch) - new_v.flatten()) / np.var(
np.array(winner_batch))
print(
"kl:{:.5f},lr_multiplier:{:.3f},loss:{},entropy:{},explained_var_old:{:.3f},explained_var_new:{:.3f}"
.format(kl, self.lr_multiplier, loss, entropy, explained_var_old,
explained_var_new))
return loss, entropy
def run(self):
"""训练"""
try:
for i in range(self.game_batch_num):
self.collect_selfplay_data(self.play_batch_size)
print("batch i:{}, episode_len:{}".format(
i + 1, self.episode_len))
if len(self.data_buffer) > self.batch_size:
loss, entropy = self.policy_update()
print("LOSS:", loss)
# 保存loss
with open('loss.txt', 'a') as f:
f.writelines(str(loss) + '\n')
if (i + 1) % self.check_freq == 0:
print("current self-play batch: {}".format(i + 1))
# win_ratio = self.policy_evaluate()
self.policy_value_net.save_model('current_policy') # 保存模型
except KeyboardInterrupt:
print('\n\rquit')
k][4 * x - 1::]
# On afficher le damier au complet et à l'endroit
dam.reverse()
NOMA = et['joueurs'][0]['nom']
NOMB = et['joueurs'][1]['nom']
return f'Légende: 1={NOMA}, 2={NOMB}\n -----------------------------------\n' + ''.join(
dam
) + '--|-----------------------------------\n | 1 2 3 4 5 6 7 8 9'
c = analyser_commande()
#Mode manuel
if not (c.a or c.x or c.ax):
jeu = Quoridor((c.idul, 'robot'))
ID, etat = debuter_partie(c.idul)
while True:
TYPE_COUP = input('Quel type de coup voulez-vous jouer? (D/MH/MV) ')
POSI = input(
'À quelle position (x,y) voulez-vous jouer ce coup ? ').replace(
' ', '')
x, y = int(POSI[1]), int(POSI[3])
if TYPE_COUP.lower() == 'd':
jeu.déplacer_jeton(1, (x, y))
print(jeu)
time.sleep(0.6)
reponse = jouer_coup(ID, 'D', (x, y))
elif TYPE_COUP.lower() == 'mh':
else:
return encode_loc(flip_y_perspective(row, current_player, True),
col) + "v"
if temperature < 1e-6:
# Do max operation instead of unstable low-temperature manipulations
idx = torch.argmax(policy_planes)
else:
idx = torch.multinomial(policy_planes.flatten()**temperature,
num_samples=1)
return _idx_to_action(idx.item())
if __name__ == '__main__':
# mini test
q = Quoridor()
legal_moves = q.all_legal_moves(partial_check=False)
print("INITIAL STATE LEGAL MOVES ({} of them):".format(len(legal_moves)))
print(legal_moves)
for mv in legal_moves:
planes = encode_actions_to_planes(mv, q.current_player)
print("=========== {} ============".format(mv))
print(planes)
mv2 = sample_action(planes, 0)
print(mv2)
assert mv2 == mv, "Failed to encode/decode {}".format(mv)
# Test that just sampling random moves leads to some illegal moves getting selected (this is expected)
random_actions, masked_random_actions = [''] * 100, [''] * 100
class TrainPipeline(object):
def __init__(self, init_model=None):
self.game = Quoridor()
self.learn_rate = 2e-3
self.lr_multiplier = 1.0
self.temp = 1.0
self.n_playout = 200
self.c_puct = 5
self.buffer_size = 10000
self.data_buffer = deque(maxlen=self.buffer_size)
self.play_batch_size = 1
self.kl_targ = 0.02
self.check_freq = 10
self.game_batch_num = 1000
self.best_win_ratio = 0.0
self.pure_mcts_playout_num = 1000
self.old_probs = 0
self.new_probs = 0
self.first_trained = False
if init_model:
self.policy_value_net = PolicyValueNet(model_file=init_model)
else:
self.policy_value_net = PolicyValueNet()
self.mcts_player = MCTSPlayer(self.policy_value_net.policy_value_fn, c_puct=self.c_puct,
n_playout=self.n_playout, is_selfplay=1)
def get_equi_data(self, play_data):
extend_data = []
for i, (state, mcts_prob, winner) in enumerate(play_data):
wall_state = state[:3,:BOARD_SIZE - 1,:BOARD_SIZE - 1]
dist_state1 = np.reshape(state[(6 + (WALL_NUM + 1) * 2), :BOARD_SIZE, :BOARD_SIZE], (1, BOARD_SIZE, BOARD_SIZE))
dist_state2 = np.reshape(state[(7 + (WALL_NUM + 1) * 2), :BOARD_SIZE, :BOARD_SIZE], (1, BOARD_SIZE, BOARD_SIZE))
# horizontally flipped game
flipped_wall_state = []
for i in range(3):
wall_padded = np.fliplr(wall_state[i])
wall_padded = np.pad(wall_padded, (0,1), mode='constant', constant_values=0)
flipped_wall_state.append(wall_padded)
flipped_wall_state = np.array(flipped_wall_state)
player_position = state[3:5, :,:]
flipped_player_position = []
for i in range(2):
flipped_player_position.append(np.fliplr(player_position[i]))
flipped_player_position = np.array(flipped_player_position)
h_equi_state = np.vstack([flipped_wall_state, flipped_player_position, state[5:, :,:]])
h_equi_mcts_prob = np.copy(mcts_prob)
h_equi_mcts_prob[11] = mcts_prob[10] # SE to SW
h_equi_mcts_prob[10] = mcts_prob[11] # SW to SE
h_equi_mcts_prob[9] = mcts_prob[8] # NE to NW
h_equi_mcts_prob[8] = mcts_prob[9] # NW to NE
h_equi_mcts_prob[7] = mcts_prob[6] # EE to WW
h_equi_mcts_prob[6] = mcts_prob[7] # WW to EE
h_equi_mcts_prob[3] = mcts_prob[2] # E to W
h_equi_mcts_prob[2] = mcts_prob[3] # W to E
h_wall_actions = h_equi_mcts_prob[12:12 + (BOARD_SIZE-1) ** 2].reshape(BOARD_SIZE-1, BOARD_SIZE-1)
v_wall_actions = h_equi_mcts_prob[12 + (BOARD_SIZE-1) ** 2:].reshape(BOARD_SIZE-1, BOARD_SIZE -1)
flipped_h_wall_actions = np.fliplr(h_wall_actions)
flipped_v_wall_actions = np.fliplr(v_wall_actions)
h_equi_mcts_prob[12:] = np.hstack([flipped_h_wall_actions.flatten(), flipped_v_wall_actions.flatten()])
# Vertically flipped game
flipped_wall_state = []
for i in range(3):
wall_padded = np.flipud(wall_state[i])
wall_padded = np.pad(wall_padded, (0,1), mode='constant', constant_values=0)
flipped_wall_state.append(wall_padded)
flipped_wall_state = np.array(flipped_wall_state)
flipped_player_position = []
for i in range(2):
flipped_player_position.append(np.flipud(player_position[1-i]))
flipped_player_position = np.array(flipped_player_position)
cur_player = (np.ones((BOARD_SIZE, BOARD_SIZE)) - state[5 + 2* (WALL_NUM+1),:,:]).reshape(-1,BOARD_SIZE, BOARD_SIZE)
v_equi_state = np.vstack([flipped_wall_state, flipped_player_position, state[5+(WALL_NUM+1):5 + 2*(WALL_NUM+1), :,:], state[5:5+(WALL_NUM+1),:,:], cur_player, dist_state2, dist_state1])
# v_equi_state = np.vstack([flipped_wall_state, flipped_player_position, state[5:(5 + (WALL_NUM+1) * 2), :, :], cur_player, state[:(6 + (WALL_NUM + 1) * 2), :, :]])
v_equi_mcts_prob = np.copy(mcts_prob)
v_equi_mcts_prob[11] = mcts_prob[9] # SE to NE
v_equi_mcts_prob[10] = mcts_prob[8] # SW to NW
v_equi_mcts_prob[9] = mcts_prob[11] # NE to SE
v_equi_mcts_prob[8] = mcts_prob[10] # NW to SW
v_equi_mcts_prob[5] = mcts_prob[4] # NN to SS
v_equi_mcts_prob[4] = mcts_prob[5] # SS to NN
v_equi_mcts_prob[1] = mcts_prob[0] # N to S
v_equi_mcts_prob[0] = mcts_prob[1] # S to N
h_wall_actions = v_equi_mcts_prob[12:12 + (BOARD_SIZE-1) ** 2].reshape(BOARD_SIZE-1, BOARD_SIZE-1)
v_wall_actions = v_equi_mcts_prob[12 + (BOARD_SIZE-1) ** 2:].reshape(BOARD_SIZE-1, BOARD_SIZE -1)
flipped_h_wall_actions = np.flipud(h_wall_actions)
flipped_v_wall_actions = np.flipud(v_wall_actions)
v_equi_mcts_prob[12:] = np.hstack([flipped_h_wall_actions.flatten(), flipped_v_wall_actions.flatten()])
## Horizontally-vertically flipped game
wall_state = state[:3,:BOARD_SIZE - 1,:BOARD_SIZE - 1]
flipped_wall_state = []
for i in range(3):
wall_padded = np.fliplr(np.flipud(wall_state[i]))
wall_padded = np.pad(wall_padded, (0,1), mode='constant', constant_values=0)
flipped_wall_state.append(wall_padded)
flipped_wall_state = np.array(flipped_wall_state)
flipped_player_position = []
for i in range(2):
flipped_player_position.append(np.fliplr(np.flipud(player_position[1-i])))
flipped_player_position = np.array(flipped_player_position)
cur_player = (np.ones((BOARD_SIZE, BOARD_SIZE)) - state[5 + 2*(WALL_NUM+1),:,:]).reshape(-1,BOARD_SIZE, BOARD_SIZE)
hv_equi_state = np.vstack([flipped_wall_state, flipped_player_position, state[5 + (WALL_NUM+1):5 + 2*(WALL_NUM+1), :,:], state[5:5+(WALL_NUM+1),:,:], cur_player, dist_state2, dist_state1])
# hv_equi_state = np.vstack([flipped_wall_state, flipped_player_position, state[5:(5 + (WALL_NUM+1) * 2), :, :], cur_player, state[(6 + (WALL_NUM + 1) * 2):, :, :]])
hv_equi_mcts_prob = np.copy(mcts_prob)
hv_equi_mcts_prob[11] = mcts_prob[8] # SE to NW
hv_equi_mcts_prob[10] = mcts_prob[9] # SW to NE
hv_equi_mcts_prob[9] = mcts_prob[10] # NE to SW
hv_equi_mcts_prob[8] = mcts_prob[11] # NW to SE
hv_equi_mcts_prob[7] = mcts_prob[6] # EE to WW
hv_equi_mcts_prob[6] = mcts_prob[7] # WW to EE
hv_equi_mcts_prob[5] = mcts_prob[4] # NN to SS
hv_equi_mcts_prob[4] = mcts_prob[5] # SS to NN
hv_equi_mcts_prob[3] = mcts_prob[2] # E to W
hv_equi_mcts_prob[2] = mcts_prob[3] # W to E
hv_equi_mcts_prob[1] = mcts_prob[0] # N to S
hv_equi_mcts_prob[0] = mcts_prob[1] # S to N
h_wall_actions = hv_equi_mcts_prob[12:12 + (BOARD_SIZE-1) ** 2].reshape(BOARD_SIZE-1, BOARD_SIZE-1)
v_wall_actions = hv_equi_mcts_prob[12 + (BOARD_SIZE-1) ** 2:].reshape(BOARD_SIZE-1, BOARD_SIZE -1)
flipped_h_wall_actions = np.fliplr(np.flipud(h_wall_actions))
flipped_v_wall_actions = np.fliplr(np.flipud(v_wall_actions))
hv_equi_mcts_prob[12:] = np.hstack([flipped_h_wall_actions.flatten(), flipped_v_wall_actions.flatten()])
###########
extend_data.append((state, mcts_prob, winner))
extend_data.append((h_equi_state, h_equi_mcts_prob, winner))
extend_data.append((v_equi_state, v_equi_mcts_prob, winner * -1))
extend_data.append((hv_equi_state, hv_equi_mcts_prob, winner * -1))
return extend_data
def collect_selfplay_data(self, n_games=1):
for i in range(n_games):
winner, play_data = self.game.start_self_play(self.mcts_player, temp=self.temp)
play_data = list(play_data)[:]
self.episode_len = len(play_data)
play_data = self.get_equi_data(play_data)
self.data_buffer.extend(play_data)
print("{}th game finished. Current episode length: {}, Length of data buffer: {}".format(i, self.episode_len, len(self.data_buffer)))
def policy_update(self):
dataloader = DataLoader(self.data_buffer, batch_size=BATCH_SIZE, shuffle=False, pin_memory=True)
valloss_acc = 0
polloss_acc = 0
entropy_acc = 0
for i in range(NUM_EPOCHS):
self.old_probs = self.new_probs
if self.first_trained:
kl = np.mean(np.sum(self.old_probs * (np.log(self.old_probs + 1e-10) - np.log(self.new_probs + 1e-10)), axis=1))
if kl > self.kl_targ * 4:
break
if kl > self.kl_targ * 2 and self.lr_multiplier > 0.1:
self.lr_multiplier /= 1.5
elif kl < self.kl_targ / 2 and self.lr_multiplier < 10:
self.lr_multiplier *= 1.5
for i, (state, mcts_prob, winner) in enumerate(dataloader):
valloss, polloss, entropy = self.policy_value_net.train_step(state, mcts_prob, winner, self.learn_rate * self.lr_multiplier)
self.new_probs, new_v = self.policy_value_net.policy_value(state)
global iter_count
writer.add_scalar("Val Loss/train", valloss.item(), iter_count)
writer.add_scalar("Policy Loss/train", polloss.item(), iter_count)
writer.add_scalar("Entropy/train", entropy, iter_count)
writer.add_scalar("LR Multiplier", self.lr_multiplier, iter_count)
iter_count += 1
valloss_acc += valloss.item()
polloss_acc += polloss.item()
entropy_acc += entropy.item()
self.first_trained = True
valloss_mean = valloss_acc / (len(dataloader) * NUM_EPOCHS)
polloss_mean = polloss_acc / (len(dataloader) * NUM_EPOCHS)
entropy_mean = entropy_acc / (len(dataloader) * NUM_EPOCHS)
#explained_var_old = 1 - np.var(np.array(winner_batch) - old_v.flatten()) / np.var(np.array(winner_batch))
#explained_var_new = 1 - np.var(np.array(winner_batch) - new_v.flatten()) / np.var(np.array(winner_batch))
#print( "kl:{:.5f}, lr_multiplier:{:.3f}, value loss:{}, policy loss:[], entropy:{}".format(
# kl, self.lr_multiplier, valloss, polloss, entropy, explained_var_old, explained_var_new))
return valloss_mean, polloss_mean, entropy_mean
def run(self):
try:
self.collect_selfplay_data(3)
count = 0
for i in range(self.game_batch_num):
self.collect_selfplay_data(self.play_batch_size) # collect_s
print("batch i:{}, episode_len:{}".format(i + 1, self.episode_len))
if len(self.data_buffer) > BATCH_SIZE:
valloss, polloss, entropy = self.policy_update()
print("VALUE LOSS: %0.3f " % valloss, "POLICY LOSS: %0.3f " % polloss, "ENTROPY: %0.3f" % entropy)
#writer.add_scalar("Val Loss/train", valloss.item(), i)
#writer.add_scalar("Policy Loss/train", polloss.item(), i)
#writer.add_scalar("Entory/train", entropy, i)
if (i + 1) % self.check_freq == 0:
count += 1
print("current self-play batch: {}".format(i + 1))
# win_ratio = self.policy_evaluate()
# Add generation to filename
self.policy_value_net.save_model('model_7x7_' + str(count) + '_' + str("%0.3f_" % (valloss+polloss) + str(time.strftime('%Y-%m-%d', time.localtime(time.time())))))
except KeyboardInterrupt:
print('\n\rquit')
def test__init__(self):
"""test la fonction __init
Cas à tester:
- Création d'une partie nouvelle
- Création d'une partie existante
- QuoridorError si 'joueur' n'est pas de longueur 2
- QuoridorError si le nombre de murs plaçable est 0 > n >10
- QuoridorError si la position d'un joueur est invalide
- QuoridorError si l'argument 'mur' n'est pas un dictionnaire si présent
- QuoridorError si le total des murs placés et plaçables n'est pas 20
- QuoridorError si la position d'un mur est invalide
"""
# Dresser des rootaux connus pour des constructions connues
nouveau_jeu = ("légende: 1=foo 2=bar\n" +
" -----------------------------------\n" +
"9 | . . . . 2 . . . . |\n" +
" | |\n" +
"8 | . . . . . . . . . |\n" +
" | |\n" +
"7 | . . . . . . . . . |\n" +
" | |\n" +
"6 | . . . . . . . . . |\n" +
" | |\n" +
"5 | . . . . . . . . . |\n" +
" | |\n" +
"4 | . . . . . . . . . |\n" +
" | |\n" +
"3 | . . . . . . . . . |\n" +
" | |\n" +
"2 | . . . . . . . . . |\n" +
" | |\n" +
"1 | . . . . 1 . . . . |\n" +
"--|-----------------------------------\n" +
" | 1 2 3 4 5 6 7 8 9\n")
partie_existante_rootau = (
"légende: 1=foo 2=bar\n" +
" -----------------------------------\n" +
"9 | . . . . . . . . . |\n" +
" | |\n" +
"8 | . . . . . . | . . . |\n" +
" | ------- -------|------- |\n" +
"7 | . . . . 2 . | . . . |\n" +
" | |\n" +
"6 | . | . . . 1 . | . . . |\n" +
" | |------- | |\n" +
"5 | . | . . | . . . | . . . |\n" +
" | | |\n" +
"4 | . . . | . . . . . . |\n" +
" | ------- |\n" +
"3 | . . . . . | . . . . |\n" +
" | | |\n" +
"2 | . . . . . | . . . . |\n" +
" | |\n" +
"1 | . . . . . . . . . |\n" +
"--|-----------------------------------\n" +
" | 1 2 3 4 5 6 7 8 9\n")
partie_existante_etat = {
"joueurs": [{
"nom": "foo",
"murs": 7,
"pos": [5, 6]
}, {
"nom": "bar",
"murs": 3,
"pos": [5, 7]
}],
"murs": {
"horizontaux": [[4, 4], [2, 6], [3, 8], [5, 8], [7, 8]],
"verticaux": [[6, 2], [4, 4], [2, 5], [7, 5], [7, 7]]
}
}
# Test de création d'une partie nouvelle
self.assertEqual(str(Quoridor(["foo", "bar"])), nouveau_jeu)
# Test de création d'une partie déjà existante
self.assertEqual(
str(
Quoridor(partie_existante_etat['joueurs'],
partie_existante_etat['murs'])),
partie_existante_rootau)
# Test de l'erreur soulevée si l'argument 'joueur' n'est pas itérable
self.assertRaisesRegex(QuoridorError, "joueurs n'est pas iterable!",
Quoridor, 2)
# Test de l'erreur soulevée si l'argument 'joueur' n'est pas de longueur 2
self.assertRaisesRegex(QuoridorError,
"Il n'y a pas exactement 2 joueurs!", Quoridor,
["joueur1"])
self.assertRaisesRegex(QuoridorError,
"Il n'y a pas exactement 2 joueurs!", Quoridor,
["joueur1", "joueur2", "joueur3"])
# Test de l'erreur soulevée si le nombre de murs qu'un joueur peut
# placer est > 10 ou négatif
self.assertRaisesRegex(QuoridorError, "mauvais nombre de murs!",
Quoridor, [{
"nom": "foo",
"murs": 11,
"pos": (5, 6)
}, {
"nom": "bar",
"murs": 10,
"pos": (5, 7)
}])
self.assertRaisesRegex(QuoridorError, "mauvais nombre de murs!",
Quoridor, [{
"nom": "joueur1",
"murs": 10,
"pos": (5, 6)
}, {
"nom": "joueur2",
"murs": -1,
"pos": (5, 7)
}])
# Test de l'erreur soulevée si la position d'un joueur est invalide
self.assertRaisesRegex(QuoridorError, "position du joueur invalide!",
Quoridor, [{
"nom": "foo",
"murs": 10,
"pos": (5, 10)
}, {
"nom": "bar",
"murs": 10,
"pos": (5, 5)
}])
self.assertRaisesRegex(QuoridorError, "position du joueur invalide!",
Quoridor, [{
"nom": "foo",
"murs": 10,
"pos": (5, 10)
}, {
"nom": "bar",
"murs": 10,
"pos": (5, 5)
}])
# Test de l'erreur soulevée si l'argument "mur" n'est pas un dictionnaire lorsque présent
self.assertRaisesRegex(QuoridorError,
"murs n'est pas un dictionnaire!", Quoridor,
["joueur1", "joueur2"], [(5, 5)])
# Test de l'erreur soulevée si le total des murs placés et plaçables n'est pas égal à 20
self.assertRaisesRegex(QuoridorError,
"mauvaise quantité totale de murs!", Quoridor,
[{
"nom": "foo",
"murs": 5,
"pos": (5, 6)
}, {
"nom": "bar",
"murs": 10,
"pos": (5, 7)
}])
self.assertRaisesRegex(
QuoridorError, "mauvaise quantité totale de murs!", Quoridor,
[{
"nom": "foo",
"murs": 8,
"pos": (5, 6)
}, {
"nom": "bar",
"murs": 3,
"pos": (5, 7)
}], {
"horizontaux": [(4, 4), (2, 6), (3, 8), (5, 8), (7, 8)],
"verticaux": [(6, 2), (4, 4), (2, 5), (7, 5), (7, 7)]
})
# Test de l'erreur soulevée si la position d'un mur est invalide
self.assertRaisesRegex(QuoridorError, "position du mur non-valide!",
Quoridor, [{
"nom": "foo",
"murs": 9,
"pos": (3, 3)
}, {
"nom": "bar",
"murs": 9,
"pos": (7, 7)
}], {
"horizontaux": [(0, 5)],
"verticaux": [(5, 5)]
})
self.assertRaisesRegex(QuoridorError, "position du mur non-valide!",
Quoridor, [{
"nom": "foo",
"murs": 9,
"pos": (3, 3)
}, {
"nom": "bar",
"murs": 9,
"pos": (7, 7)
}], {
"horizontaux": [(9, 5)],
"verticaux": [(5, 5)]
})
self.assertRaisesRegex(QuoridorError, "position du mur non-valide!",
Quoridor, [{
"nom": "foo",
"murs": 9,
"pos": (3, 3)
}, {
"nom": "bar",
"murs": 9,
"pos": (7, 7)
}], {
"horizontaux": [(5, 1)],
"verticaux": [(5, 5)]
})
self.assertRaisesRegex(QuoridorError, "position du mur non-valide!",
Quoridor, [{
"nom": "foo",
"murs": 9,
"pos": (3, 3)
}, {
"nom": "bar",
"murs": 9,
"pos": (7, 7)
}], {
"horizontaux": [(5, 10)],
"verticaux": [(5, 5)]
})
self.assertRaisesRegex(QuoridorError, "position du mur non-valide!",
Quoridor, [{
"nom": "foo",
"murs": 9,
"pos": (3, 3)
}, {
"nom": "bar",
"murs": 9,
"pos": (7, 7)
}], {
"horizontaux": [(5, 5)],
"verticaux": [(1, 5)]
})
self.assertRaisesRegex(QuoridorError, "position du mur non-valide!",
Quoridor, [{
"nom": "foo",
"murs": 9,
"pos": (3, 3)
}, {
"nom": "bar",
"murs": 9,
"pos": (7, 7)
}], {
"horizontaux": [(5, 5)],
"verticaux": [(10, 5)]
})
self.assertRaisesRegex(QuoridorError, "position du mur non-valide!",
Quoridor, [{
"nom": "foo",
"murs": 9,
"pos": (3, 3)
}, {
"nom": "bar",
"murs": 9,
"pos": (7, 7)
}], {
"horizontaux": [(5, 5)],
"verticaux": [(5, 0)]
})
self.assertRaisesRegex(QuoridorError, "position du mur non-valide!",
Quoridor, [{
"nom": "foo",
"murs": 9,
"pos": (3, 3)
}, {
"nom": "bar",
"murs": 9,
"pos": (7, 7)
}], {
"horizontaux": [(5, 5)],
"verticaux": [(5, 9)]
})
def test_déplacer_jeton(self):
""" Test de la fonction déplacer_jeton
Cas à tester:
- déplacement des deux joueurs fonctionne bien
- QuoridorError si le joueur indiqué est invalide
- QuoridorError si la position est hors des limites du jeu
- QuoridorError si la position n'est pas accessible au joueur
"""
etat_partie = {
"joueurs": [{
"nom": "joueur1",
"murs": 7,
"pos": (5, 6)
}, {
"nom": "joueur2",
"murs": 3,
"pos": (5, 7)
}],
"murs": {
"horizontaux": [(4, 4), (2, 6), (3, 8), (5, 8), (7, 8)],
"verticaux": [(6, 2), (4, 4), (2, 5), (7, 5), (7, 7)]
}
}
etat_partie2 = {
"joueurs": [{
"nom": "joueur1",
"murs": 7,
"pos": (5, 6)
}, {
"nom": "joueur2",
"murs": 3,
"pos": (5, 5)
}],
"murs": {
"horizontaux": [(4, 4), (2, 6), (3, 8), (5, 8), (7, 8)],
"verticaux": [(6, 2), (4, 4), (2, 5), (7, 5), (7, 7)]
}
}
etat_partie3 = {
"joueurs": [{
"nom": "joueur1",
"murs": 7,
"pos": (5, 6)
}, {
"nom": "joueur2",
"murs": 3,
"pos": (6, 5)
}],
"murs": {
"horizontaux": [(4, 4), (2, 6), (3, 8), (5, 8), (7, 8)],
"verticaux": [(6, 2), (4, 4), (2, 5), (7, 5), (7, 7)]
}
}
nouveaujeu = Quoridor(["joueur1", "joueur2"])
# Tester l'erreur soulevée si le joueur indiqué est invalide
self.assertRaisesRegex(QuoridorError, "joueur invalide!",
nouveaujeu.déplacer_jeton, 5, (5, 2))
# Tester l'erreur soulevée si la position demandée est hors des limites du jeu
self.assertRaisesRegex(QuoridorError, "position invalide!",
nouveaujeu.déplacer_jeton, 1, (0, 5))
self.assertRaisesRegex(QuoridorError, "position invalide!",
nouveaujeu.déplacer_jeton, 1, (10, 5))
self.assertRaisesRegex(QuoridorError, "position invalide!",
nouveaujeu.déplacer_jeton, 1, (5, 0))
self.assertRaisesRegex(QuoridorError, "position invalide!",
nouveaujeu.déplacer_jeton, 1, (5, 10))
# Tester l'erreur soulevée si la position demandée n'est pas accessible au joueur
jeu = Quoridor(etat_partie['joueurs'], etat_partie['murs'])
self.assertRaisesRegex(QuoridorError, "mouvement invalide!",
jeu.déplacer_jeton, 2, (5, 8))
self.assertRaisesRegex(QuoridorError, "mouvement invalide!",
jeu.déplacer_jeton, 1, (5, 8))
self.assertRaisesRegex(QuoridorError, "mouvement invalide!",
jeu.déplacer_jeton, 2, (3, 7))
self.assertRaisesRegex(QuoridorError, "mouvement invalide!",
jeu.déplacer_jeton, 2, (4, 6))
self.assertRaisesRegex(QuoridorError, "mouvement invalide!",
jeu.déplacer_jeton, 2, (6, 6))
# Tester des déplacements qui fonctionnent
jeu.déplacer_jeton(2, (5, 5))
self.assertEqual(jeu.état_partie(), etat_partie2)
jeu.déplacer_jeton(2, (6, 5))
self.assertEqual(jeu.état_partie(), etat_partie3)
def main():
parser = argparse.ArgumentParser()
parser.add_argument(
"--player_type",
type=int,
default=1,
help="palyer type you want to fight,1 is human,2 is computer")
parser.add_argument("--computer_type",
type=int,
default=0,
help="computer type,1 is Alpha MCTS,2 is pure MCTS")
args = parser.parse_args()
game = Quoridor()
human1 = ManualPygameAgent('Kurumi')
human2 = ManualPygameAgent('Cryer')
MCTS_Alpha = A_Player(PolicyValueNet().policy_value_fn,
c_puct=5,
n_playout=30,
is_selfplay=0)
MCTS_Pure = B_Player(c_puct=5, n_playout=50) # 50层400秒
if args.player_type == 1:
player_types = {1: 'human', 2: 'human'}
players = {1: human1, 2: human2}
if args.computer_type == 0:
pass
elif args.player_type == 2:
player_types = {1: 'human', 2: 'computer'}
if args.computer_type == 1:
players = {1: human1, 2: MCTS_Alpha}
elif args.computer_type == 2:
players = {1: human1, 2: MCTS_Pure}
elif args.computer_type == 0:
print("Set computer type to 1 or 2 for choosing computer!")
# pygame.quit()
# game.load(player1, player2)
pygame.init()
WINDOW_SIZE = [SCREEN_WIDTH, SCREEN_HEIGHT]
screen = pygame.display.set_mode(WINDOW_SIZE)
pygame.display.set_caption("QUORIDOR")
clock = pygame.time.Clock()
# valid_actions = game.valid_actions 11
valid_actions = game.actions()
done = False
winner = None
t1 = time.time()
while not done:
player_moved = False
# 定义落子历史
# move_history = []
pawn_moves, walls = draw_game(game, screen, valid_actions)
# text(screen, "player1 move:", position1=2, position2=0.8, color=BLUE)
valid_walls = [wall for wall in walls if wall[2] in valid_actions]
if player_types[game.current_player] == 'human':
touch = pygame.mouse.get_pos()
for wall, collides, _ in valid_walls:
for collide in collides:
if collide.collidepoint(touch):
pygame.draw.rect(screen, LIGHTBROWN, wall)
break
for event in pygame.event.get():
if event.type == pygame.QUIT:
done = True
elif event.type == pygame.MOUSEBUTTONDOWN:
touch = pygame.mouse.get_pos()
# This is messy - fix later
for rect, action in pawn_moves:
if rect.collidepoint(touch):
players[game.current_player].receive_action(action)
player_moved = True
break
if player_moved:
break
# if player_moved: break
# 添加
if player_moved:
real_action = players[
game.current_player].choose_action()
# move_history.append(real_action)
done, winner = game.step(real_action)
render(game, screen) # 渲染游戏
break
for rect, collide_points, action in valid_walls:
for collides in collide_points:
if collides.collidepoint(touch):
players[game.current_player].receive_action(
action)
player_moved = True
break
# 修改
if player_moved == True:
real_action = players[
game.current_player].choose_action()
# move_history.append(real_action)
done, winner = game.step(real_action)
render(game, screen) # 渲染游戏
break
clock.tick(30)
pygame.display.flip()
valid_actions = game.actions()
# 待改
if player_types[game.current_player] == 'computer':
print("computer %s thinking..." % str(game.current_player))
tic = time.time()
# real_action = np.random.choice(valid_actions)
real_action = players[game.current_player].choose_action(game)
# move_history.append(real_action)
toc = time.time()
print("MCTS choose action:", real_action,
" ,spend %s seconds" % str(toc - tic))
done, winner = game.step(real_action)
# render(game, screen)
# valid_actions = game.valid_actions
# if game.current_player == 1:
# text(screen, text, position1=2, position2=0.8, color=BLUE)
if done:
print("game over! winner is %s player:%s" %
(player_types[winner], winner))
break
t2 = time.time()
print("total time :", t2 - t1)
pygame.quit()
from quoridor import Quoridor
partie = Quoridor(["rod", "jimbo"])
print(partie)
print(partie.placer_mur(1, [1, 8], "horizontal"))
print(partie)
try:
partie.placer_mur(1, [2, 8], "horizontal")
except:
print("aaaaaaaaaaaaaaaaaaaaa")
print(partie)
partie.placer_mur(1, [3, 8], "horizontal")
print(partie)
partie.placer_mur(1, [3, 4], "vertical")
print(partie)
try:
partie.placer_mur(1, [3, 5], "vertical")
except:
print(".awemc3eic43icm3ocmn3rkco ")
partie.placer_mur(1, [3, 6], "vertical")
partie.placer_mur(1, (2, 1), "qmce")
print(partie)
partie.placer_mur_auto(1, partie.etat["joueurs"][1]["pos"], "horizontal")
q = QuoridorX(DAMIER['joueurs'], DAMIER['murs'])
except RuntimeError as err:
print(err)
CHOIX = input(
"Voulez-vous continuer à jouer, oui ou non? ")
if CHOIX.lower() == 'non':
break
except StopIteration as err:
q.window.clearscreen()
q = QuoridorX(DAMIER['joueurs'], DAMIER['murs'])
print(f'Le grand gagnant est le joueur {err} !\n')
break
elif ARGS.automatique:
#automatique et ascii
print('automatique et ascii')
q = Quoridor(PARTIE[1]['joueurs'], PARTIE[1]['murs'])
print(q)
while True:
try:
TYPE_COUP, POSITION = q.jouer_coup(1)
DAMIER = jouer_coup(ID_PARTIE, TYPE_COUP, POSITION)
q = Quoridor(DAMIER['joueurs'], DAMIER['murs'])
print(q)
except RuntimeError as err:
print(err)
CHOIX = input("Voulez-vous continuer à jouer, oui ou non? ")
if CHOIX.lower() == 'non':
break
except StopIteration as err:
print(q)
print(f'Le grand gagnant est le joueur {err} !\n')
def test_placer_mur(self):
""" Test de la fonction placer_mur
Cas à tester:
- Les murs horizontaux et verticaux sont placés correctement
- QuoridorError si le numéro du joueur n'est pas bon
- QuoridorError si un mur occupe déjà la position
- QuoridorError si la position est invalide pour l'horientation
- QuoridorError si le joueur a déjà placé tous ses murs
"""
jeu1_etat = {
"joueurs": [{
"nom": "joueur1",
"murs": 9,
"pos": (5, 1)
}, {
"nom": "joueur2",
"murs": 9,
"pos": (5, 9)
}],
"murs": {
"horizontaux": [(4, 4)],
"verticaux": [(6, 6)]
}
}
jeu2_etat = {
"joueurs": [{
"nom": "joueur1",
"murs": 8,
"pos": (5, 1)
}, {
"nom": "joueur2",
"murs": 8,
"pos": (5, 9)
}],
"murs": {
"horizontaux": [(4, 4), (5, 5)],
"verticaux": [(6, 6), (7, 7)]
}
}
jeu3_etat = {
"joueurs": [{
"nom": "joueur1",
"murs": 7,
"pos": (5, 3)
}, {
"nom": "joueur2",
"murs": 0,
"pos": (3, 5)
}],
"murs": {
"horizontaux": [(4, 4), (2, 6), (4, 2), (5, 8), (7, 8)],
"verticaux": [(6, 2), (4, 4), (2, 5), (7, 5), (7, 7), (2, 2),
(2, 3), (2, 4)]
}
}
jeu1 = Quoridor(jeu1_etat['joueurs'], jeu1_etat['murs'])
# Tester si le mur est bien placé avec les 2 joueurs
jeu1.placer_mur(1, (5, 5), 'horizontal')
jeu1.placer_mur(2, (7, 7), 'vertical')
self.assertEqual(jeu1.état_partie(), jeu2_etat)
# Tester l'erreur si le numéro du joueur n'est pas bon
self.assertRaisesRegex(QuoridorError, "joueur invalide!",
jeu1.placer_mur, 5, (2, 2), 'horizontal')
# Tester l'erreur si le joueur ne peut plus placer de murs
jeu3 = Quoridor(jeu3_etat['joueurs'], jeu3_etat['murs'])
self.assertRaisesRegex(QuoridorError,
"le joueur ne peut plus placer de murs!",
jeu3.placer_mur, 2, (2, 2), 'horizontal')
# Tester l'erreur si l'emplacement est déjà occupé pour un mur horizontal -->
# position exacte
self.assertRaisesRegex(QuoridorError, "Il y a déjà un mur!",
jeu3.placer_mur, 1, (4, 4), 'horizontal')
# Position décallée
self.assertRaisesRegex(QuoridorError, "Il y a déjà un mur!",
jeu3.placer_mur, 1, (5, 4), 'horizontal')
# Tester l'erreur si l'emplacement est déjà occupé pour un mur vertical --> position exacte
self.assertRaisesRegex(QuoridorError, "Il y a déjà un mur!",
jeu3.placer_mur, 1, (4, 4), 'vertical')
# Position décallée
self.assertRaisesRegex(QuoridorError, "Il y a déjà un mur!",
jeu3.placer_mur, 1, (4, 5), 'vertical')
# Tester l'erreur si l'orientation n'est pas valide
self.assertRaisesRegex(QuoridorError, "orientation invalide!",
jeu3.placer_mur, 1, (4, 5), 'diagonale')
# Tester l'erreur si la position est hors des limites du jeu pour un mur horizontal
self.assertRaisesRegex(QuoridorError, "position du mur invalide!",
jeu1.placer_mur, 1, (0, 5), 'horizontal')
self.assertRaisesRegex(QuoridorError, "position du mur invalide!",
jeu1.placer_mur, 1, (9, 5), 'horizontal')
self.assertRaisesRegex(QuoridorError, "position du mur invalide!",
jeu1.placer_mur, 1, (5, 1), 'horizontal')
self.assertRaisesRegex(QuoridorError, "position du mur invalide!",
jeu1.placer_mur, 1, (5, 10), 'horizontal')
# Tester l'erreur si la position est hors des limites du jeu pour un mur vertical
self.assertRaisesRegex(QuoridorError, "position du mur invalide!",
jeu1.placer_mur, 1, (1, 5), 'vertical')
self.assertRaisesRegex(QuoridorError, "position du mur invalide!",
jeu1.placer_mur, 1, (10, 5), 'vertical')
self.assertRaisesRegex(QuoridorError, "position du mur invalide!",
jeu1.placer_mur, 1, (5, 0), 'vertical')
self.assertRaisesRegex(QuoridorError, "position du mur invalide!",
jeu1.placer_mur, 1, (5, 9), 'vertical')
# tester l'erreur si le coup enfermerait le joueur
self.assertRaisesRegex(nx.exception.NetworkXError, "", jeu3.placer_mur,
1, (3, 3), 'horizontal')
self.assertRaisesRegex(nx.exception.NetworkXError, "", jeu3.placer_mur,
1, (4, 2), 'vertical')