def __init__(self, env):
self.temp = 1
self.start_temp = 1
self.end_temp = 0.2
self.action_space = card.get_action_space()
self.name = 'global'
self.env = env
self.a_dim = 8310
self.gamma = 0.99
self.sess = None
self.train_intervals = 30
self.trainer = tf.train.AdamOptimizer(learning_rate=0.001)
self.episode_rewards = [[] for i in range(2)]
self.episode_length = [[] for i in range(2)]
self.episode_mean_values = [[] for i in range(2)]
self.summary_writers = [
tf.summary.FileWriter("train_agent%d" % i) for i in range(2)
]
self.agents = [
CardAgent('agent%d' % i, self.trainer) for i in range(2)
]
self.global_episodes = tf.Variable(0,
dtype=tf.int32,
name='global_episodes',
trainable=False)
self.increment = self.global_episodes.assign_add(1)
def __init__(self):
self.deck = None
self.players = []
self.last_player = None
self.last_cards = None
self.history = []
self.extra_cards = []
self.action_space = card.get_action_space()
self.next_turn = 0
self.reset()
def collect_data():
cnt = 0
action_space = card.get_action_space()
# print(action_space)
# print("a : ", len(action_space))
while cnt < N:
gameID = random.randint(0, N - 1)
demoGame = demoGames[gameID]
lordID = demoGame.lordID
gameLen = len(demoGame.actions)
# while True:
th = random.randint(0, gameLen - 1)
# if not(th % 3 == lordID):
# break
# if actions[th] == 0:
# continue
handcards = to_color_handcards(demoGame.handcards)
# print("demo handcards", demoGame.handcards)
# print("hand ", handcards)
# print("lord ", lordID)
extracards = to_color_extracards(handcards[lordID], demoGame.extracards)
# print("demo extracards", demoGame.extracards)
# print("hand ", extracards)
# print("demo actions", demoGame.actions)
acts = [action_space[a] for a in demoGame.actions]
# print("demo actions", acts)
# print("th = ", th)
outCardList = [[] for i in range(3)]
ind = lordID
last_cards = []
last_ID = lordID
for i in range(th):
put_list = to_color_putlist(action_space[demoGame.actions[i]], handcards[ind])
if not(put_list == []):
last_cards = copy.deepcopy(put_list)
last_ID = ind
outCardList[ind] += put_list
# print(put_list)
for c in put_list:
handcards[ind].remove(c)
ind = int(ind + 1) % 3
if (last_ID == (th + lordID) % 3):
last_cards = []
# print("last cards ", last_cards, [to_card(x) for x in last_cards])
state = []
total = [1 for i in range(54)]
# print(handcards[th % 3])
self_cards = to_one_hot(handcards[(th + lordID) % 3])
remains = subtract(total, self_cards);
history = [to_one_hot(outCardList[i]) for i in range(3)]
# for i in range(3):
# print("out card ", i, outCardList[i], [to_card(x) for x in outCardList[i]])
# print("history ", history)
for i in range(3):
remains = subtract(remains, history[i])
extra_cards = to_one_hot(extracards)
state += self_cards;
state += remains;
state += history[0];
state += history[1];
state += history[2];
state += extra_cards;
# numOfFeasibleActs = 0
# print("feasible actions : ")
action = []
mask = get_mask([to_card(color) for color in handcards[(th + lordID) % 3]], action_space, [to_card(color) for color in last_cards])
for a in range(len(action_space)):
if mask[a]:
if a == demoGame.actions[th]:
action.append(1)
else:
action.append(0)
# print(action_space[a])
# numOfFeasibleActs += 1
# print("num of fea : ", numOfFeasibleActs)
# X.append(state)
# Y.append(action)
printf(state)
printf(action)
# print("state ", cnt, " ", [(i % 54, state[i]) for i in range(54 * 6)])
# print("action ", cnt, " ", action, len(action))
return
cnt += 1