def playWeChatJump():
actions = 10
brain = BrainDQN(actions) #action采用one-hot编码
observation0 = game.init_state()
brain.setInitState(observation0)
while 1 != 0:
action = brain.getAction()
nextObservation, reward, terminal = game.frame_step(action)
nextObservation = np.reshape(nextObservation, (80, 80, 1))
brain.setPerception(nextObservation, action, reward, terminal)
noise = 3
num_sensor = 10 # N
policy = 2 # choose power change policy for PU, it should be 1(Multi-step) or 2(Single step)
brain = BrainDQN(actions, num_sensor)
com = GameState(P_1, P_2, noise, num_sensor)
terminal = True
recording = 100000
while (recording > 0):
# initialization
if (terminal == True):
com.ini()
observation0, reward0, terminal = com.frame_step(
np.zeros(actions), policy, False)
brain.setInitState(observation0)
# train
action, recording = brain.getAction()
nextObservation, reward, terminal = com.frame_step(action, policy, True)
loss = brain.setPerception(nextObservation, action, reward)
# test
if (recording + 1) % 500 == 0:
Loss.append(loss)
print "iteration : %d , loss : %f ." % (100000 - recording, loss)
success = 0.0
fre = 0
num = 1000.0