rnn_agent.summarize(ep_r, global_counter, 'reward_900')
##开始训练
for ep in range(Epoch):
# if Inflow_rate < 600:
# Inflow_rate += 20
#每次开始训练重新注册环境
myTrafficNet = TrafficLightGridNetwork(
name='grid',
vehicles=vehicles,
net_params=get_myNetParams(),
)
env = MyMultiTrafficLightGridPOEnv(env_params=flow_params['env'],
sim_params=flow_params['sim'],
network=myTrafficNet)
# test phase
if ep >= 0:
print('测试阶段：')
print(ep)
record_line(each_line_path, "*** Epoch: {} ***\n".format(ep))
queue, speed, ret = [], [], []
for i in range(3):
ep_r, ep_q, ep_v = [], [], []
state = env.reset()
state = normalize_formation(state, Agent_NUM)
_state = [n for a in state for n in a]
_state = np.array(_state).reshape([-1, num_state * Agent_NUM])
def record_line(log_path, line):
with open(log_path, 'a') as fp:
fp.writelines(line)
fp.writelines("\n")
return True
if __name__ == "__main__":
myTrafficNet = TrafficLightGridNetwork(
name = 'grid',
vehicles = vehicles,
net_params = myNetParams,
)
env = MyMultiTrafficLightGridPOEnv(
env_params=flow_params['env'], sim_params=flow_params['sim'], network=myTrafficNet
)
# Perparations for agents
from flow.core.ppo_agent import *
# used for ploting
import matplotlib.pyplot as plt
import numpy as np
Reward_num = 0 #0代表多个rewards，1代表1个
NAME = '2x2_600_PPO_Hierarchy_SOFT_try0'
#*********************
Epoch = 100 #100 10
sub_train_epi = 5 # 20 1
steps = 210 # 210 10
def record_line(log_path, line):
with open(log_path, 'a') as fp:
fp.writelines(line)
fp.writelines("\n")
return True
if __name__ == "__main__":
myTrafficNet = TrafficLightGridNetwork(
name = 'grid',
vehicles = vehicles,
net_params = myNetParams,
)
env = MyMultiTrafficLightGridPOEnv(
env_params=flow_params['env'], sim_params=flow_params['sim'], network=myTrafficNet)
env.sim_params.render = True
env.restart_simulation(sim_params=sumoparams, render=True)
# print(env.scenario.get_edge_list())
# Perpare agent.
from flow.core.ppo_rnn_discrete import *
############################################################################
############################################################################
Agent_NUM = N_ROWS * N_COLUMNS
Reward_num = 1 #0代表多个rewards，1代表1个
write = 1 #是否将模拟哦结果写入文档。0否，1是
NAME = '3x3_400_discreteRNN_ALL_SOFT_try1_numcar15_aRL5'
num_state = 198
rnn_agent = PPO(s_dim=num_state*Agent_NUM,a_dim=Agent_NUM,name=NAME)
rnn_agent.restore_params(NAME,10)
############################################################################