def __init__ (self, bidding, player, p, learning_p, active_P_count):
self.player = player
self.learning_p = learning_p
if active_P_count == 2:
if player != 0 and player != 2:
self.legal_bids = [possible_bids.index ("P")]
else:
self.legal_bids = self.getLegalBid (bidding)
else:
self.legal_bids = self.getLegalBid (bidding)
self.W = []
self.N = []
self.Q = []
self.P = []
self.children = []
self.width = len (self.legal_bids)
for i in range (self.width):
self.W.append (0)
self.N.append (0)
self.Q.append (0)
self.P.append (0)
self.children.append (None)
if len (self.legal_bids) > 1 and self.player != learning_p:
self.P = self.addNoise (p [0])
def getLegalBid(bidding):
ret = []
legal_bids = generateBiddings(bidding, "Competitive")
for bid in legal_bids:
ret.append(possible_bids.index(bid))
return ret
def update (self, value, bid): index = self.legal_bids.index (possible_bids.index (bid)) self.N [index] += 1 self.W [index] += value self.Q [index] = self.W [index] / self.N [index] return
def one_hot_encode (self, bids): indices = [] res = np.zeros (POSSIBLE_BID_COUNT * 3) for bid in bids [-3:]: indices.append (possible_bids.index (bid)) base = 3 - len (indices) for i in range (len (indices)): res [indices [i] + POSSIBLE_BID_COUNT * (base + i)] = 1 return list (res)
def expReplay(network_1, par, hands, resTable, mode=2):
agents = {}
counter = 0
stats = list(-1 for i in range(STAT_SIZE * 4))
temp_Agent = Agent("", {})
temp_Agent.setHand(hands[0])
for i in range(STAT_SIZE):
stats[i] = temp_Agent.stat[i]
temp_Agent.setHand(hands[2])
for i in range(STAT_SIZE):
stats[i + STAT_SIZE * 2] = temp_Agent.stat[i]
while counter < 10000:
ended = False
bids = []
player = 0
related_agents = []
last_agent = None
while not ended:
key = str(player) + "|" + toString(bids)
if key not in agents.keys():
agents[key] = Agent_Open_Hand(network_1, {},
EXPLORE_COEFFICIENT)
agents[key].setState((stats, bids.copy(), player),
hands[player])
agent = agents[key]
related_agents.append(agent)
bid = agent.quickMove()
bids.append(bid)
bids.append("P")
if player == 0:
player = 2
else:
player = 0
# print (bids)
if bids[-3:] == ["P", "P", "P"]:
ended = True
if last_agent != None and last_agent.children[possible_bids.index(
bid)] == None:
last_agent.children[possible_bids.index(bid)] = agent
last_agent = agent
score = getScore(bids, resTable)
# print (counter)
# print (bids)
for agent in reversed(related_agents):
agent.updateValues(score - par)
counter += 1
# print ("--------------")
# print ("--------------")
# print ("--------------")
x = []
y = []
for key in agents.keys():
# print (key, max (agents [key].visit_count))
x.append(agents[key].X)
y.append(agents[key].visit_count)
return
if K.backend() == "tensorflow":
X = np.asarray(x)
Y_true = np.asarray(y)
# print ("X_SHAPE", x.shape)
# print ("Y_SHAPE", y.shape)
if _DEBUG > 2:
session = tf.Session()
print(session.run(custom_loss(network.predict(X), Y_true)))
print(network.predict(X))
print("------")
network.fit(X, Y_true, epochs=1, verbose=1)
print(session.run(custom_loss(network.predict(X), Y_true)))
print(network.predict(X))
print("------End of Exp Replay-----")