def main():
traci.start(sumoCmd)
agent = Agent(traci, True)
traci.close()
# start simulation
traci.start(sumoCmd)
agent.reset()
simulation_steps = 0
traffic_data = []
yell_time = 3
atom_time = 30
controlTLIds = "0"
while simulation_steps < 60000:
traffic_data.append(
measure_traffic_step(controlTLIds, simulation_steps))
agent.state = agent._getState()
action, current_phase = agent.agentAct()
if action == current_phase:
agent.setRYG(action, False)
for j in range(atom_time):
traci.simulationStep()
else:
agent.setRYG(current_phase, True)
for j in range(yell_time):
traci.simulationStep()
agent.setRYG(action, False)
for j in range(atom_time - yell_time):
traci.simulationStep()
agent._getReward() # stop training
simulation_steps += atom_time
# The END of simulationSteps loop
traci.close()
traffic_data = pd.DataFrame(traffic_data)
traffic_data.to_csv(
"/home/hjj/exp_test/Test/TestData/near_unba/a2c/interval30/a2c_near_unba_30_new.csv"
)
collect_tripinfo(
trip_file,
'/home/hjj/exp_test/Test/TestData/near_unba/a2c/interval30/a2c_near_unba_30_new'
)
def main():
# no_of_workers = 1
len_of_track = 16 # 一次MC更新的轨迹长度
# 初始化workers
workers = []
traci.start(sumoCmd)
global_agent = Agent('global', traci)
traci.close(sumoCmd)
TOTAL_EPOCH = 200
SIM_STEPS = 6000
atom_time = 30
yell_time = 3
outfile = open(out_file_path, 'w')
start = time.clock()
data = []
Cmd = []
for epoch in range(TOTAL_EPOCH):
update_record = 0
print("epoch%i========================================\n"%epoch)
traci.start(sumoCmd)
global_agent.reset(traci)
simulation_steps = 0
reward = 0
total_reward = 0
avg_wait_time = 0
# 初始化结束
# simulation loop
while simulation_steps < SIM_STEPS:
state = global_agent._getState(traci)
global_agent.state = state # 重要！！！
global_agent.buffer_s.append(state)
action, current_phase = global_agent.agentAct(traci, global_agent) # worker[i]
global_agent.buffer_a.append(action)
if action == current_phase:
global_agent.setRYG(action,False,traci)
for j in range(atom_time):
traci.simulationStep()
else:
global_agent.setRYG(current_phase,True,traci)
for j in range(yell_time):
traci.simulationStep()
global_agent.setRYG(action,False,traci)
for j in range(atom_time-yell_time):
traci.simulationStep()
next_state = global_agent._getState(traci)
global_agent.buffer_s_.append(next_state)
reward, avg_wait_time = global_agent._getReward(traci)
global_agent.buffer_r.append(reward)
total_reward += reward
print("Simulation {} : avg_wait_time_30 {}, reward_epoch {}\n"
.format(simulation_steps, avg_wait_time, total_reward))
log = {'Simulation_step': simulation_steps, 'avg_wait_time_30':avg_wait_time,'reward_30': reward,'reward_epoch': total_reward}
data.append(log)
update_record += 1
simulation_steps += atom_time
if update_record % len_of_track == 0 or update_record % (SIM_STEPS/atom_time) == 0: # 一次轨迹为16或者到每个epoch的结尾时
len_single_worker = len(global_agent.buffer_a)
bs, ba, br, bs_ = np.vstack(global_agent.buffer_s), np.vstack(global_agent.buffer_a), np.vstack(global_agent.buffer_r), np.vstack(global_agent.buffer_s_)
global_agent.learner.learn(bs, ba, br, bs_, len_single_worker, 1)
traci.close()
save_multi_agents(save_path, save_path_best, global_agent)
# The END of epoch loop
df = pd.DataFrame(data)
df.to_csv('/home/hjj/single/Train/train_data/a2c_test.csv')
