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)
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
for ind in range(1000):
T = 0
com.ini_test()
print("iterations:%d" % Time)
actionIput = Qnetwork.getAction_1(actionNum=ACTION_NUM, stateInput=stateInput,Time= Time)
reward,actionshow = Env.getReward(stateInput=stateInput, actionInput=actionIput)
ActionShow.append(actionshow)
nextState = Env.creatSensor(Power=sig[Time + 1])
loss = Qnetwork.getLoss(currentState=stateInput, nextState=nextState, action=actionIput, reward=reward)
Time = Time + 1
R_total += reward
Reward.append(R_total)
else:
#get satate\action\reward\nextstate
stateInput = Env.creatSensor(Power=sig[Time])
print("iterations:%d" %Time)
actionIput = Qnetwork.getAction(actionNum= ACTION_NUM,stateInput= stateInput)
reward,actionshow = Env.getReward(stateInput= stateInput,actionInput= actionIput)
ActionShow.append(actionshow)
nextState = Env.creatSensor(Power= sig[Time+1])
#get loss and train Qnetwork
loss = Qnetwork.getLoss(currentState= stateInput,nextState= nextState,action=actionIput,reward= reward)
R_total += reward
Reward.append(R_total)
if not loss == 0:
Loss.append(loss)
Time = Time + 1