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)
import numpy as np
import matplotlib.pyplot as plt
P_1 = [round(0.1 * i / 2.0, 2) for i in range(1, 9)]
P_2 = [round(0.1 * i / 2.0, 2) for i in range(1, 9)]
actions = len(P_2)
Loss = []
Success = []
Fre = []
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 = [] SENSOR_NUM = 10 ACTION_NUM = 26 record = 95000 Time = 10 #构造状态生成函数A*sin(omega*t) N = 10000 t = np.linspace(0, 10000, num=N) #步进为1，共采样10000个数据 omega = math.pi / 25 A = 10 x1 = A * np.sin(omega * t) DQN1 = BrainDQN(SENSOR_NUM, ACTION_NUM) ENV = ENVIRNOMENT(ACTION_NUM) state = x1[Time] stateInput = ENV.creat_sensor(Power=x1[Time], sensor_num=SENSOR_NUM) #通过 action_input = DQN1.getAction_1(action_num=ACTION_NUM, stateInput=stateInput) print(action_input) # reward = ENV.get_reward(stateInput= stateInput,actionInput= action_input) # nextState = ENV.creat_sensor(Power= x1[time+1],sensor_num= SENSOR_NUM) # # #将stateinput、actioninput、reward、nextState放入replaymemory中 # LOSS.append(DQN1.get_loss(currentState= stateInput,nextState= nextState,action=action_input,reward= reward)) # time = time + 1 # record = record -1
LOSS = []
SENSOR_NUM = 10
ACTION_NUM = 26
record = 95000
time = 0
#构造状态生成函数A*sin(omega*t)
N = 100000
t = np.linspace(0, 100000, num=N) #步进为1，共采样10000个数据
omega = math.pi / 25
A = 10
x1 = A * np.sin(omega * t)
DQN = BrainDQN(SENSOR_NUM, ACTION_NUM)
ENV = ENVIRNOMENT(ACTION_NUM)
R = []
R_total = 0
while not record == 0:
if time > 90000:
break
state = x1[time]
stateInput = ENV.creat_sensor(Power=x1[time], sensor_num=SENSOR_NUM)
action_input = DQN.getAction(action_num=ACTION_NUM, stateInput=stateInput)
reward = ENV.get_reward(stateInput=stateInput, actionInput=action_input)
#print(reward)
R_total += reward
R.append(R_total)
nextState = ENV.creat_sensor(Power=x1[time + 1], sensor_num=SENSOR_NUM)
t2 = np.linspace(0,8,num= 9)
omega = math.pi/9
A = 10
sig = A * np.sin(omega*t1)
#get actionChoice matrix
def actionChoice(sig):
actionChoice = []
for i in range(5):
actionChoice.append(sig[i])
for i in range(10,14):
actionChoice.append(sig[i])
return actionChoice
actionChoice = actionChoice(sig)
Qnetwork = BrainDQN(SENSOR_NUM,ACTION_NUM)
Env = ENVIRONMENT(SensorNum= SENSOR_NUM,ActionNum=ACTION_NUM,ActionChoice=actionChoice)
Reward = []
R_total = 0
Loss = []
ActionShow = []
while Time < N-1:
if Time < 2000: #使用先验经验训练3个周期
stateInput = Env.creatSensor(Power=sig[Time])
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)
SU_powers = [round(i / 10.0, 2) for i in range(1, 15)]
actions = len(SU_powers)
PU_num = 2
# rd.randint(1, 10)
SU_num = 5
# rd.randint(1, 10)
Loss = []
Success = []
Fre = []
noise = 3
policy = 1 # choose power change policy for PU, it should be 1(Multi-step) or 2(Single step)
brain = BrainDQN(actions, PU_num, SU_num)
com = GameState(PU_powers, SU_powers, noise, PU_num, SU_num)
terminal = True
recording = 1000
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)