def run_one_episode_on_STL(self, episode, total_episodes):
# load fixed route:
simu_type = genorator.gen_route(episode, is_random = False) # load fixed_route.
print('Run fixed_route on mode: ', self.get_simu_type_str(simu_type))
self._epsilon = 0.0 # all actions by model.
traci.start(self._sumoCmd) # then, start sumo
# INIT some vars:
self._steps = 0
self._sum_intersection_queue = 0 # increases every step/seconds
tot_neg_reward = 0 # total negative reward
pre_wait_time = 0 #
# INIT my vars:
action = 0 # initial action
old_action = 0
state = self._get_state(self.I)
# run 1 simulation (maxsteps)
while self._steps < self._max_steps:
# reset current_wait_time:
current_wait_time = 0
# select action (select index of action, then edit action_time)
action = self._choose_action(state)
# ================================ Take action ====================================================================
if self._steps != 0 and old_action != action:
# just set traffic_light in sumo
self._set_yellow_phase(old_action)
# ver 2: dynamic yellow.
yellow_duration_new = self.cal_yellow_duration(old_action)
current_wait_time = self._simulate(yellow_duration_new)
self._set_green_phase(action)
current_wait_time = self._simulate(self._green_duration)
# =================================================================================================================
# get next_state and reward
next_state = self._get_state(self.I)
reward = pre_wait_time - current_wait_time
if reward < 0:
tot_neg_reward += reward
# reassign:
pre_wait_time = current_wait_time
state = next_state
old_action = action
# print
print('step: ', self._steps, '|',self._max_steps,' || action: ', self.get_action_name(action), ': || negative reward: ', tot_neg_reward)
# append data
self._save_fixed_status(simu_type)
# close gui.
traci.close(wait = False)
def run_one_episode(self, simu_type, total_episodes):
traffic_code_mode = genorator.gen_route(simu_type, is_random = False) # gen route file.
print('Mode: ', self.get_simu_type_str(traffic_code_mode))
traci.start(self._sumoCmd) # then, start sumo
traci.gui.setZoom('View #0',1500)
traci.gui.setOffset('View #0',595,916)
self.reset()
# INIT some vars:
self._steps = 0
self._sum_intersection_queue = 0 # increases every step/seconds
tot_neg_reward = 0 # total negative reward
pre_wait_time = 0 #
# INIT my vars:
new_action = 0 # initial action
state = self._get_state(self.I)
# test
actions_count = [0,0] # count numb_times of picked actions.
# run 1 simulation (1h30m)
# while self._steps < self._max_steps:
while (traci.simulation.getMinExpectedNumber() > 0):
current_wait_time = 0
# select action (select index of action, then edit action_time)
new_action = self._choose_action(state)
# for writing log:
actions_count[new_action] += 1
action_name = self.get_action_name(new_action)
# ================================ Take new_action ====================================================================
if self._steps != 0 and old_action != new_action:
self._set_yellow_phase(old_action)
current_wait_time = self._simulate(self._yellow_duration)
# take action:
self._set_green_phase(new_action)
current_wait_time = self._simulate(self._green_duration)
# ======================================================================================================================
# get next_state and reward
next_state = self._get_state(self.I)
reward = pre_wait_time - current_wait_time
self._neg_pos_reward_store.append(reward)
if reward < 0:
tot_neg_reward += reward
# reassign:
pre_wait_time = current_wait_time
state = next_state
old_action = new_action
# print
# print('step: ', self._steps, ' || action: ', action_name, ' || negative reward: ', tot_neg_reward)
self._save_stats(traffic_code_mode, tot_neg_reward) # mode LOW + total neg-REWARD
print("Total negative reward: {}, Eps: {}".format(tot_neg_reward, self._epsilon))
mode = self.get_simu_type_str(traffic_code_mode)
log_path = constants.plot_path_trained_model + mode + '/'
os.makedirs(os.path.dirname(log_path), exist_ok=True)
log = open(log_path +'tog_neg_reward.txt', 'a') # open file text.
text = mode + ' reward: ' + str(tot_neg_reward) + ' _sum_intersection_queue: ' + str(self._sum_intersection_queue)
log.write(text)
log.close()
# close gui.
traci.close(wait = False)
def run_one_episode_training(self, episode, total_episodes):
# gen 3 way random route.
simu_type = genorator.gen_route(episode, is_random = True)
print('Training on random route with type: ', self.get_simu_type_str(simu_type))
self._epsilon = 1.0 - (episode / constants.total_ep_for_epislon) # setup epsilon
traci.start(self._sumoCmd) # then, start sumo
# INIT some vars:
self._steps = 0
self._sum_intersection_queue = 0 # increases every step/seconds
tot_neg_reward = 0 # total negative reward
pre_wait_time = 0 #
# INIT my vars:
action = 0 # initial action
old_action = 0
state = self._get_state(self.I)
action_count = [0,0] # cal percent of actions
good_bad_count = [0,0] # count good bad actions
# run 1 simulation (maxsteps)
while self._steps < self._max_steps:
# reset current_wait_time:
current_wait_time = 0
# select action (select index of action, then edit action_time)
action = self._choose_action(state)
# just count numb of taken actions.
action_count[action] += 1
# ================================ Take action ====================================================================
if self._steps != 0 and old_action != action:
# just set traffic_light in sumo
self._set_yellow_phase(old_action)
# ver 1: fixed yellow.
# current_wait_time = self._simulate(self._yellow_duration)
# ver 2: dynamic yellow.
yellow_duration_new = self.cal_yellow_duration(old_action)
current_wait_time = self._simulate(yellow_duration_new)
self._set_green_phase(action)
current_wait_time = self._simulate(self._green_duration)
# =================================================================================================================
# get next_state and reward
next_state = self._get_state(self.I)
reward = pre_wait_time - current_wait_time
if reward < 0:
tot_neg_reward += reward
good_bad_count[1] += 1
else:
good_bad_count[0] += 1
# save tuple:
self._memory.add_sample((state, action, reward, next_state))
# training:
self._replay()
# reassign:
pre_wait_time = current_wait_time
state = next_state
old_action = action
# print
eval_this_action = 'Good action' if (reward>=0) else 'Bad Action'
print('step: ', self._steps, '|',self._max_steps,' || action: ', self.get_action_name(action), ': ',eval_this_action,' || negative reward: ', tot_neg_reward)
print('percent of actions: ', np.array(action_count)/sum(action_count))
print("Total negative reward: {}, Eps: {}".format(tot_neg_reward, self._epsilon))
# append data
self._save_stats(simu_type,tot_neg_reward)
# save array to disk:
self.save_all_arrays()
# close gui.
traci.close(wait = False)
def run_fixed_duration(self, simu_type, durations_of_phases):
traffic_code_mode = genorator.gen_route(simu_type, is_random = False) # gen route file.
print('Fixed route, mode: ', self.get_simu_type_str(simu_type))
traci.start(self._sumoCmd) # then, start sumo
# reset everything:
self.reset()
self._steps = 0
self._sum_intersection_queue = 0
tot_neg_reward = 0
pre_wait_time = 0
action = 0 # initial action
# run 1 simulation (1h30m)
while self._steps < self._max_steps:
# reset current_wait_time:
current_wait_time = 0
# ============================================================ Perform action ======================
# 0: EW green 1: EW yellow
# 2: NS green 3: NS Yellow
if action == 0:
self._set_green_phase(action)
current_wait_time = self._simulate(durations_of_phases[0])
elif action == 1:
self._set_yellow_phase(0)
current_wait_time = self._simulate(self._yellow_duration)
elif action == 2:
self._set_green_phase(1)
current_wait_time = self._simulate(durations_of_phases[1])
elif action == 3:
self._set_yellow_phase(1)
current_wait_time = self._simulate(self._yellow_duration)
# =================================================================================================================
reward = pre_wait_time - current_wait_time
self._neg_pos_reward_store.append(reward)
if reward < 0:
tot_neg_reward += reward
# reassign:
pre_wait_time = current_wait_time
# next action (phase):
action = self.select_fixed_action(action)
# print every step:
# print('step: ', self._steps, ' || action: ', self.get_action_name(action), ' || negative reward: ', tot_neg_reward)
self._save_stats(traffic_code_mode, tot_neg_reward) # mode LOW + total neg-REWARD
print("Total negative reward: {}, Total_waiting_time: {}, AWT: {}".format(tot_neg_reward, self._sum_intersection_queue, self._avg_waiting_time_per_veh))
mode = self.get_simu_type_str(traffic_code_mode)
log_path = constants.plot_path_fixed_sys + mode + '/'
os.makedirs(os.path.dirname(log_path), exist_ok=True)
log = open(log_path +'tog_neg_reward.txt', 'a') # open file text.
text = mode + ' reward: ' + str(tot_neg_reward) + ' _sum_intersection_queue: ' + str(self._sum_intersection_queue) + ' || AWT: ' + str(self._avg_waiting_time_per_veh) + '\n'
log.write(text)
log.close()
# close gui.
traci.close(wait = False)
