def __init__(self, gui=True, testing='budget'):
super(GridWin, self).__init__()
traci.start(["sumo", "-c", Settings.sumo_config])
self.mode='default' #default, reward
self.testing = testing #if testing budget keep capacity constant, if testing capacity keep budget constant
self.env_map = Map(Settings.sumo_config)
self.rowcol_to_junction = self.env_map.complex_row_col
self.row = int(self.env_map.rows)
self.column = int(self.env_map.columns)
#self.rowcol_to_junction = self.env_map.row_col(self.row, self.column) #value is the junction id and key is row_col
self.rowcol_to_junction.update(dict((v, k) for k, v in self.rowcol_to_junction.items()))
self.player_list = {} #stores location as key, player object as value in list
self.reward_list = {} #stores location(grid) as key, reward value as value
self.random_uniform_reward_list = {} #akbas
self.min_reward = 0 #min reward value
self.gui = gui
self.setting = Settings()
#settings
self.reward_distribution_center = [(0,0), (0,self.column),(self.row, 0),(self.row, self.column)]
if gui:
self.protocol("WM_DELETE_WINDOW", self.on_closing)
w=self.winfo_screenwidth()
h=self.winfo_screenheight()
size = str(int(w/2))+'x'+str(int(h/2))
#print(size)
self.geometry(size)
self.import_frame = tk.Frame(self, bg='green')
#self.import_frame.pack(expand=True, fill='both')
self.grid_frame = tk.Frame(self, height=500, width=500, bg='red')
self.grid_frame.pack(expand=True, fill='both')
self.control_frame = tk.Frame(self)
self.control_frame.pack(expand=True, fill='both')
self.grid_list = [] #store all the buttons in grid
self.spawn_grid()
self.control_buttons()
self.mainloop()
def start_veto(ctx, users: List[discord.User], veto_type: int) -> str:
global RUNNING_VETOS
global FUCK_NUKE
num_of_players = len(users)
maps = [
Map('ancient'),
Map('anubis'),
Map('cache'),
Map('dust2'),
Map('inferno'),
Map('mirage'),
Map('overpass'),
Map('train'),
Map('vertigo'),
]
if not FUCK_NUKE:
maps.append(Map('nuke'))
for i in RUNNING_VETOS:
if i.channel == ctx.channel.name and i.server == ctx.guild.name:
return 'veto already running!'
if num_of_players == 0:
return 'gimme players'
elif num_of_players > 5:
return 'too many players'
random.shuffle(users)
players: List[Player] = []
for u in users:
players.append(Player(u))
veto = Veto(
ctx.channel.name,
ctx.guild.name,
veto_type,
maps,
players,
)
RUNNING_VETOS.append(veto)
message = veto.add_list_to_message('', maps, [])
message += veto.construct_message_veto_list()
return message
def __init__(self):
self.map_data = Map(GraphSetting.sumo_config)
self.player_list = {}
self.rewards_list = {}
self.index_counter = 0
self.poi_to_junct = {}
self.poi_list = {}
self.player_location_array = None #array storing location
def __init__(self):
self.map_data = Map(GraphSetting.sumo_config)
self.player_list = {}
self.poi_to_junct = {} #enter poi id to get junction, enter junction to get poi id
self.index_counter = 0
self.poi_list = {} #poi {poi_key:{veh_key:veh_value}} to keep track of if vehcle is being tracked by the poi to know if
#veh is leaving or entering junction
self.success_veh = [] #vehicle success arrived dest
self.poi_que = {} # {poiid:{vehid:edge, vehid:edge}} #when player arived to raidus add to this, every 10 sec this is cleared
#below are the variables for guis
self.veh_data = None #result of veh storing location
self.track_veh = None #id storing the vehicle to track
def start_best_of_veto(
ctx,
users: List[discord.User],
num_of_maps: int,
) -> str:
# system for BO1
# maps: dust2, inferno, mirage, nuke, overpass, ancient, vertigo
# ban, ban, ban, ban , ban, ban, decider
# 0 1 2 3 4 5 6
# system for BO3
# maps: dust2, inferno, mirage, nuke, overpass, ancient, vertigo
# ban, ban, pick, pick, ban, ban, decider
# 0 1 2 3 4 5 6
# system for BO5
# maps: dust2, inferno, mirage, nuke, overpass, ancient, vertigo
# ban, ban, pick, pick, pick, pick, decider
# 0 1 2 3 4 5 6
global RUNNING_VETOS
maps = [
Map('dust2'),
Map('inferno'),
Map('mirage'),
Map('nuke'),
Map('overpass'),
Map('ancient'),
Map('vertigo'),
]
num_of_players = len(users)
for i in RUNNING_VETOS:
if i.channel == ctx.channel.name and i.server == ctx.guild.name:
return 'veto already running on this channel'
if num_of_players != 2:
return 'need 2 players for veto!'
random.shuffle(users)
starter = users[0].name
second = users[1].name
players = [
Player(starter), Player(second),
Player(starter), Player(second),
Player(starter), Player(second),
]
veto = Veto(
ctx.channel.name,
ctx.guild.name,
num_of_maps,
maps,
players,
)
RUNNING_VETOS.append(veto)
message = veto.add_list_to_message('', maps, [])
message += veto.construct_message_best_of_veto_list()
return message
def add_map(self, mapveto: Map):
self.mapveto = mapveto.capitalize()
class Environment(object):
def __init__(self):
self.map_data = Map(GraphSetting.sumo_config)
self.player_list = {}
self.poi_to_junct = {} #enter poi id to get junction, enter junction to get poi id
self.index_counter = 0
self.poi_list = {} #poi {poi_key:{veh_key:veh_value}} to keep track of if vehcle is being tracked by the poi to know if
#veh is leaving or entering junction
self.success_veh = [] #vehicle success arrived dest
self.poi_que = {} # {poiid:{vehid:edge, vehid:edge}} #when player arived to raidus add to this, every 10 sec this is cleared
#below are the variables for guis
self.veh_data = None #result of veh storing location
self.track_veh = None #id storing the vehicle to track
def stop_vehicle_handle(self, t): #handle when vehicle is stopped calculate the sensing plan with buffered vehicles
if (t % GraphSetting.buffer_interval) == 0: #the number of sec passed by
#print("im checking for que ", self.t)
if self.poi_que:
for poi_key, veh_waiting_list in self.poi_que.items():
sp, number_players = self.update_veh_collection_status(veh_waiting_list, poi_key) #this is only called when veh is being handled at poi
reward_value = (self.map_data.pois[poi_key].value/pow(number_players, 2))
for veh, edge in veh_waiting_list.items():
self.adjust_sensing_plan(poi_key, veh, sp, edge, reward_value) #when veh route is adjusted does this really need to be adjusted for every veh or maybe it should be only per poi
#self.player_list[veh].reward += (self.map_data.pois[poi_key].value/pow(len(veh_waiting_list), 2))
try:
traci.vehicle.setStop(veh, edge, duration=0)
#print(f"i succeeded at resuming {veh}. routes: {traci.vehicle.getRoute(veh)}, index : {traci.vehicle.getRouteIndex(veh)}, current: {traci.vehicle.getRoute(veh)[traci.vehicle.getRouteIndex(veh)]} shouldbe {edge}")
except traci.exceptions.TraCIException as e:
print(f"i failed at resuming {veh}. routes: {traci.vehicle.getRoute(veh)}, index : {traci.vehicle.getRouteIndex(veh)}, current: {traci.vehicle.getRoute(veh)[traci.vehicle.getRouteIndex(veh)]} shouldbe {edge}")
#traci.vehicle.setStop(veh, edge, duration=0)
self.track_veh = veh
#traci.vehicle.setStop(veh, edge, duration=0)
raise traci.exceptions.TraCIException("i failed here wtf")
#traci.vehicle.setStop(veh, traci.vehicle.getRoute(veh)[traci.vehicle.getRouteIndex(veh)], duration=0)
print(f"moving {veh} at sim step {t}, processed sp for {len(veh_waiting_list)} player(s)")
self.poi_que = {}
def calculate_next_poi_new(self, veh_id, current_node, add=False):#add to show that its initializing
#this is called when player is added and is called every time a play arrived at a poi
#loops through every player and every poi to find the player prob for every poi
#print("current node is ", current_node)
player = self.player_list[veh_id]
max_eu = 0
max_eu_location=None
final_distance_cap = None
player_data = self.player_list if add else traci.vehicle.getAllSubscriptionResults()
len_player = len(player_data)
for poi_id, self.map_data.pois[poi_id] in self.map_data.pois.items():
assert len_player >0, 'something is wrong no player in system'
if not poi_id in self.player_list[veh_id].poi_potential:
#player cant go to this poi due to distance capacity
continue
else:
print(f"{veh_id} considering going to {poi_id}")
total_cost = self.map_data.pois[poi_id].player_potential[veh_id] + self.player_list[veh_id].poi_potential[poi_id]
assert total_cost <= self.player_list[veh_id].distance_capacity, f"failed {veh_id} dist cap {self.player_list[veh_id].distance_capacity} < {total_cost} going to {poi_id}"
assert poi_id in self.player_list[veh_id].current_poi_distribution, f"failed {poi_id} not in {veh_id} distribution"
if len_player <=1: #im the only one there
e_u = self.map_data.pois[poi_id].value
e_sp = 1
else:
e_u, e_sp = self.calculate_utility(veh_id, poi_id, self.map_data.pois[poi_id].value, player_data)
if (e_u > max_eu) and (e_sp <= player.capacity) and (self.map_data.pois[poi_id].junction!=player.target_poi):
max_eu = e_u
max_eu_location = self.map_data.pois[poi_id].junction
final_distance_cap = total_cost
#reset poi potential players or it might affect next time bucket generation
#self.map_data.pois[poi_id].player_potential = {}
if final_distance_cap:
self.player_list[veh_id].distance_capacity -= final_distance_cap
#self.player_list[veh_id].poi_potential = {}
return max_eu_location
def calculate_utility_new(self, veh_id, poi_id, reward, player_data):
'''
calculate the expected util for particular i veh_id to poi_id
sum up all the times
'''
#iteration through poi_combintaions and populate
#self.set_combs(poi_id)
player_data_keys = list(player_data.keys())
time_eu_dict = {} #time as key m_eu_dict as value
time_esp_dict = {}
for index, time in enumerate(self.player_list[veh_id].current_poi_distribution[poi_id][1]):
#filter out the ones that are not going there that means i have to regenerate combinations...... thats even more costly
#combinations are assumed that all players have some probalit
#even if i remove the player from the consideration list
#self.filter_participants()
for m, combs in self.player_list[veh_id].combinations.items(): #m is r {1:[10c1], 2:[10c2]} r 1:n
prob = 1
m_eu_dict = {} #number of players as key, eu as value
m_esp_dict = {}
for key, value in player_data.items(): #calculate the prob first then filter out the one that are not likely
if key in combs:
arriving_prob = self.find_probability(time, self.player_list[key].current_poi_distribution[poi_id])
if key == veh_id:
assert self.player_list[veh_id].current_poi_distribution[poi_id][0][index] == arriving_prob,\
f"time dist doesnt match {veh_id} {self.player_list[veh_id].current_poi_distribution[poi_id][0][index]} {arriving_prob}"
else:
arriving_prob = (1 - self.find_probability(time, self.player_list[key].current_poi_distribution[poi_id]))
prob *= arriving_prob
eu = prob*(reward/pow(m, 2))
esp = prob*self.compute_sensing_plan(m, (reward/pow(m, 2)), self.map_data.junctions[self.map_data.pois[poi_id].junction].cost)
m_eu_dict[m] = eu
m_esp_dict[m] = esp
time_eu_dict[time] = m_eu_dict
time_esp_dict[time] = m_esp_dict
total_eu = sum([sum(_m.values()) for _time, _m in time_eu_dict.items()])
total_esp = sum([sum(_m.values()) for _time, _m in time_esp_dict.items()])
print(f"for veh {veh_id} to poi {poi_id} eu is {total_eu} esp is {total_esp} current cap:{self.player_list[veh_id].capacity}")
return total_eu, total_esp
def calculate_utility(self, veh_id, poi_id, reward, player_data):
'''
calculate the expected util for particular i veh_id to poi_id
sum up all the times
'''
#iteration through poi_combintaions and populate
self.player_list[veh_id].combinations = defaultdict(list) #reset player combinations
self.set_combs(poi_id) #setting the combinations for all the players that are potentially going to this poi
player_data = self.map_data.pois[poi_id].player_potential
player_data_keys = list(player_data.keys())
time_eu_dict = {} #time as key m_eu_dict as value
time_esp_dict = {}
for index, time in enumerate(self.player_list[veh_id].current_poi_distribution[poi_id][1]):
#filter out the ones that are not going there that means i have to regenerate combinations...... thats even more costly
#combinations are assumed that all players have some probalit
#even if i remove the player from the consideration list
#self.filter_participants()
for m, combs in self.player_list[veh_id].combinations.items(): #m is r {1:[10c1], 2:[10c2]} r 1:n #this is set using set_combos
prob = 1
m_eu_dict = {} #number of players as key, eu as value
m_esp_dict = {}
for key, value in player_data.items(): #iteration of all players potentially going to poi
if key in combs:
arriving_prob = self.find_probability(time, self.player_list[key].current_poi_distribution[poi_id])
if key == veh_id:
assert self.player_list[veh_id].current_poi_distribution[poi_id][0][index] == arriving_prob,\
f"time dist doesnt match {veh_id} {self.player_list[veh_id].current_poi_distribution[poi_id][0][index]} {arriving_prob}"
else:
arriving_prob = (1 - self.find_probability(time, self.player_list[key].current_poi_distribution[poi_id]))
prob *= arriving_prob
eu = prob*(reward/pow(m, 2))
esp = prob*self.compute_sensing_plan(m, (reward/pow(m, 2)), self.map_data.junctions[self.map_data.pois[poi_id].junction].cost)
m_eu_dict[m] = eu
m_esp_dict[m] = esp
time_eu_dict[time] = m_eu_dict
time_esp_dict[time] = m_esp_dict
total_eu = sum([sum(_m.values()) for _time, _m in time_eu_dict.items()])
total_esp = sum([sum(_m.values()) for _time, _m in time_esp_dict.items()])
print(f"for veh {veh_id} to poi {poi_id} eu is {total_eu} esp is {total_esp} current cap:{self.player_list[veh_id].capacity}")
return total_eu, total_esp
def generate_bucket(self, veh_id=None):
#when veh arrive at destination the bucket should be changed for each veh the same combination no longer apply
#this also sets the potential players to every potential pois based on distance cap
#this is caled everytime player arrive at a poi
def set_bucket(veh_id, current_edge):
print(f"Generating buckets for {veh_id}....")
#for each veh need to reset their dic before generating buckets
for poi_id, poi_value in self.map_data.pois.items():
key = self.map_data.pois[poi_id].junction
value = self.map_data.pois[poi_id].value
route_value = self.map_data.find_route_reroute(current_edge, key)
route_value_todest = self.map_data.find_best_route(key, self.player_list[veh_id].dest_junc)
total_time = route_value.travelTime + route_value_todest.travelTime
if self.player_list[veh_id].distance_capacity < total_time:
continue
#the player combination to poi is not updated
print(f"considering {poi_id} total cost {total_time} cap left {self.player_list[veh_id].distance_capacity}")
self.player_list[veh_id].poi_potential[poi_id] = route_value.travelTime
self.map_data.pois[poi_id].player_potential[veh_id] = route_value_todest.travelTime
route_edges = route_value.edges
self.player_list[veh_id].temp_edges[key]=route_edges
new_mean = sum([self.map_data.edges[e].distance/self.map_data.edges[e].speed for e in route_edges if not ':' in e]) #sum of the means of edges within the route to each poi
new_std = reduce(lambda x,y:np.sqrt(x**2+y**2), [self.map_data.edges[e].std for e in route_edges if not ':' in e]) # combine the std of all the edges
#: is for junctions, when vehicle in motion, tc.roadid can return junction
route_array = generate_speed_distribution(new_mean, new_std) #distribution data generated based on new mean and std
result = generate_speed_bucket(route_array, bins_num=6) #generate histogram with bin number
self.player_list[veh_id].current_poi_distribution[poi_id] = result #save this histogram information to the player object
#current poi_distribution {poi_id:histogram}
if veh_id: #finding dict for only 1 vehicle
set_bucket(veh_id, self.player_list[veh_id].current_edge)
#self.set_combinations_new(add=True)
else: #for when 1 vehicle arrived at a poi need to evaluate the next poi thus need to update every other players bucket
for poi_id, poi_value in self.map_data.pois.items():
poi_value.player_potential = {}
for veh_id, veh_value in traci.vehicle.getAllSubscriptionResults().items():
self.player_list[veh_id].poi_potential = {}
for veh_id, veh_value in traci.vehicle.getAllSubscriptionResults().items():
set_bucket(veh_id, veh_value[tc.VAR_ROAD_ID])
#self.set_combinations_new()
#because we are in the player loop after updating the potential pois for this particular player, go ahead and generate the combination for this player
def set_combs(self, poi_id, add=False): #setting the combinations of those players who are potentially able to participate in this poi
total_players = len(self.map_data.pois[poi_id].player_potential)
print(f"{poi_id} is generating combinations for {total_players}")
for i in range(total_players):
combs = itertools.combinations(list(self.map_data.pois[poi_id].player_potential.keys()), i+1) #try optimizing using numpy or iterators instread
#combs =
self.map_data.pois[poi_id].combinations[i+1] = combs
self.set_combinations_player(i+1, combs) #setting combs for theplayers based off the comb
def set_bucket_new(self, veh_id, current_edge, add=True):
#print(f"setting histogram for {veh_id} add: {add}")
for poi_id, self.map_data.pois[poi_id] in self.map_data.pois.items():
key = self.map_data.pois[poi_id].junction
value = self.map_data.pois[poi_id].value
route_edges = self.map_data.find_route_reroute(current_edge, key).edges
self.player_list[veh_id].temp_edges[key]=route_edges
new_mean = sum([self.map_data.edges[e].distance/self.map_data.edges[e].speed for e in route_edges if not ':' in e]) #sum of the means of edges within the route to each poi
new_std = reduce(lambda x,y:np.sqrt(x**2+y**2), [self.map_data.edges[e].std for e in route_edges if not ':' in e]) # combine the std of all the edges
#: is for junctions, when vehicle in motion, tc.roadid can return junction
route_array = generate_speed_distribution(new_mean, new_std) #distribution data generated based on new mean and std
result = generate_speed_bucket(route_array, bins_num=6) #generate histogram with bin number
self.player_list[veh_id].current_poi_distribution[poi_id] = result #save this histogram information to the player object
def compute_sensing_plan(self, player_amount, reward, cost):
#print('player amount is ', player_amount)
sensing_plan = ((player_amount-1)*reward)/((player_amount**2)*cost)
#print('sensning plan value is ', sensing_plan)
return sensing_plan
def print_pc(self):
for key, value in self.player_list.items():
print(value.combinations)
#value.combinations = defaultdict(list)
def set_combinations_player(self, i, combs):
'''
np_combs = np.array(combs) #each combination is a row
print(np_combs)
for player_id, player_value in self.player_list.items():
#print(f"procssing {player_id} is at {np.apply_along_axis(lambda x: player_id in x, 1, np_combs)}")
player_value.combinations[i].append(np_combs[(np.apply_along_axis(lambda x: player_id in x, 1, np_combs)), :])
print("first combintation list")
self.print_pc()
#exit()
'''
while True:
try:
comb = next(combs)#combs.pop()
#print(f"combinations for {i} players {comb}")
for player in comb:
self.player_list[player].combinations[i].append(comb)
except StopIteration as e:
break
def set_combinations(self, add = False):
#this gets combinations for all the players after all the players has been initialized
#intialize combination
#this need to be fixed for memory error cant store all combinations for every number of vehicles
#where should the poi potential players be populated, shoudl be inside calculate next poi new, but if its populated there then combinations should be generated per poi based
'''
print(f"All players added len: {len(self.player_list)} generating combinations...")
player_keys = list(self.player_list.keys())
all_combs = {}
for i in range(len(self.player_list)):
#print("generating combinations")
combs = list(itertools.combinations(player_keys, i+1))
#all_combs[i+1] = combs
#print("setting combinations")
self.set_combinations_player(i+1, combs)
'''
if add:
for player_id, player_value in self.player_list.items():
self.next_poi_reroute(player_id, player_value.start, player_value.prev_junction, add=add)
def find_probability(self, time, distribution):
'''
given time and distribution of a veh at poi, find the prob of that time for the veh
[0] is the probability, [1] is the intervals/bins
'''
buckets = distribution[1]
try:
upper_index = np.min(np.where(buckets>=time)) #index of upper boundary of time
lower_index = np.max(np.where(buckets<=time)) #index of lower boundary of time
if upper_index == lower_index and upper_index==len(distribution[0]):
lower_index -= 1
#print(f'time searching for is {time}, upper:{upper_index}, lower:{lower_index}')
#print(f'bucket is:', buckets)
#print(f'prob is:', distribution[0])
except ValueError:
lower_index = None
if not lower_index:
return 0
else:
return distribution[0][lower_index]
def add_player(self, veh_id, routes, dest_junc): #this is called before setting combinations
assert not veh_id in self.player_list, f"failed more than one player with {veh_id}"
assert self.index_counter == int(veh_id.split('_')[1]), 'player id doesnt match counter'
route_edges = routes.edges
self.player_list[veh_id] = Player(veh_id, route_edges, self.map_data.edges[route_edges[0]]._from, dest_junc)
self.player_list[veh_id].capacity = get_truncated_normal(GraphSetting.player_capacity_random[0], GraphSetting.player_capacity_random[1], 0, GraphSetting.player_capacity_random[0]*2).rvs(1)[0]
try:
print(f"{veh_id} shortest path travel time {routes.travelTime}")
if GraphSetting.distance_capacity[0] == GraphSetting.distance_capacity[1]:
self.player_list[veh_id].distance_capacity = (GraphSetting.distance_capacity[0] * routes.travelTime)
else:
self.player_list[veh_id].distance_capacity = np.random.randint(routes.travelTime * GraphSetting.distance_capacity[0], routes.travelTime * GraphSetting.distance_capacity[1])
except ValueError:
self.player_list[veh_id].distance_capacity = 0
self.generate_bucket(veh_id=veh_id)
self.index_counter+=1
print(f"Added player {veh_id}, dist_cap: {self.player_list[veh_id].distance_capacity}")
def reroute(self, veh_id, current_edge, upcome_edge, destination, add=False):
try:
print(f'{veh_id} traveling on {upcome_edge} change direction going towards {destination}({self.poi_to_junct[destination]})')
except KeyError:
print(f'{veh_id} traveling on {upcome_edge} change direction going towards {destination}(Destination)')
print()
shortest_route = self.map_data.find_route_reroute(upcome_edge, destination)
shortest_route = list(shortest_route.edges)
traci.vehicle.changeTarget(veh_id, shortest_route[-1])
return shortest_route
def update_capacity(self, veh, esp):
try:
#assert self.player_list[veh].capacity >= esp, f"CAPACITY change ERROR cap:{self.player_list[veh].capacity} esp:{esp}"
if esp > self.player_list[veh].capacity and esp == 1: #in the case of 1 player esp is set to 1 but still higher than
self.player_list[veh].capacity = 0
else:
self.player_list[veh].capacity -= esp
except KeyError:
print(veh_value, 'Error')
def update_veh_collection_status(self, veh_value, poi_key):
#iterate through all the vehicles
'''
this function have issue when the diff of esp(i) and esp(i-1)
eg. if esp(2) is 5.3 esp(1) is 1, player capacitys are 3 and 20
when calc esp of 2, 1 fit. but when calc esp of 1, 2 fits
the len(veh_value) is mostly 1, only when multiple vehicles arrives in the same radius will the length >1
'''
keys_list = list(veh_value.keys())
cap_list = []
veh_cap_list = []
counter_list = []
i_list = []
temp_veh_value = veh_value.copy()
for i in range(len(veh_value), 0, -1):
esp = self.compute_sensing_plan(i, self.map_data.pois[poi_key].value, self.map_data.junctions[self.poi_to_junct[poi_key]].cost)
cap_list.append(esp)
veh_cap_list.append(self.player_list[keys_list[i-1]].capacity)
counter = 0 #this to count how many fits the capacity
min_cap_veh_id = None #only remove one veh with smallest capacity if not fit
for new_key, new_value in veh_value.items():
if esp<=self.player_list[new_key].capacity:
counter+=1
self.player_list[new_key].participation = True
else:
self.player_list[new_key].participation = False
if not min_cap_veh_id:
min_cap_veh_id = new_key
else:
if self.player_list[new_key].capacity < self.player_list[min_cap_veh_id].capacity:
min_cap_veh_id = new_key
if min_cap_veh_id:
del temp_veh_value[min_cap_veh_id]
veh_value = temp_veh_value
if counter == i: #this line
if i==1:
esp = 1
return esp, i
counter_list.append(counter)
i_list.append(i)
print("I should not be here ")
print(f'length is {len(veh_value)}, esp list:{cap_list}, cap list:{veh_cap_list} counter {counter_list}, ilist {i_list}')
exit()
def process_destination(self):
arrived_id = traci.simulation.getArrivedIDList()
if arrived_id:
for veh_id in arrived_id:
if not veh_id in self.success_veh:
self.success_veh.append(veh_id)
print(f"vehicle {veh_id} arrived at destination")
def wait_in_radius(self, poi_key,veh_id):
#print("before stop routes ", traci.vehicle.getRoute(veh_id))
#this function is for stopping vehicles
routes = traci.vehicle.getRoute(veh_id)
route_index = traci.vehicle.getRouteIndex(veh_id)
start_edge = routes[route_index]
start_index = route_index
while True:
try:
traci.vehicle.setStop(veh_id, routes[route_index])
break
except traci.exceptions.TraCIException:
route_index += 1
except IndexError:
#because the stopping edge is determined before rerouting to he next poi, the route index might be out of range because its reached the poi and dk where to go
#print(f"oh well im out of index trying to stop {veh_id} at {poi_key} starting {start_edge} index {start_index}")
#print(f"Routes: {routes}")
#exit()
routes = self.map_data.find_route_reroute(start_edge, GraphSetting.destination).edges
route_index = 0
traci.vehicle.setRoute(veh_id,routes)
print(f"stopping.... {veh_id} at {poi_key}({routes[route_index]})")
edge = routes[route_index]
try:
if not self.poi_que[poi_key]:
self.poi_que[poi_key]= {veh_id:edge}
else:
self.poi_que[poi_key][veh_id] = edge
except KeyError as e:
self.poi_que[poi_key]= {veh_id:edge}
self.track_veh = veh_id
#print(self.poi_que[poi_key])
#print("after stop routes ", traci.vehicle.getRoute(veh_id))
def adjust_sensing_plan(self, key, veh, sp, current_edge, reward_value):
#key is the current poi key
#self.player_list[veh].target_poi = self.map_data.pois[key].junction #incase veh accidentally encounter poi, need to update
#if vehicle predetermined to go to destination but encounter a poi then update
self.generate_bucket()
if self.player_list[veh].participation:
before_capacity = self.player_list[veh].capacity
self.player_list[veh].reward += reward_value
self.update_capacity(veh, sp)
print(f"{veh} CAP_before:{before_capacity} CAP_after:{self.player_list[veh].capacity}: SP:{sp} at junction {key}({self.poi_to_junct[key]})")
self.player_list[veh].participation = False
else:
print(f"{veh} not participating at {key} sp:{sp} cap:{self.player_list[veh].capacity}")
self.next_poi_reroute(veh, current_edge, self.map_data.pois[key].junction)
def next_poi_reroute(self, veh, current_edge, prev_junction, add=False): #this is called everytime we want to determine poi and reroute
next_poi = self.calculate_next_poi_new(veh, prev_junction, add=add) #maybe this function can return none for going towards dest
if not next_poi:
#this is the weighted random jumps
#right now set to go home if none fit
print(f'{veh} reached max capacity, going towards destination')
next_poi = self.player_list[veh].dest_junc
else:
self.player_list[veh].target_poi = next_poi
st = self.reroute(veh, None, current_edge, next_poi)
def process_poi(self):
#should check is the vehicle is currently on the poi adjacent edges
poi_data = traci.poi.getAllContextSubscriptionResults() #poi data is {poikey:{vehid:sub_info_veh}}
if poi_data:
for key, value in poi_data.items(): #loop through all poi list
#generate bucket should be only for one player that arrived at junction
#self.generate_bucket() #update players prob dist buckets for every reward #this is wrong should only update for that vehicle that arrived
#sp, number_players = self.update_veh_collection_status(value, key) #update all the vehicles at the junction for if they are participating or not
#print(f'esp is {esp}, {number_players}')
for veh, veh_value in value.items(): #loop through all vehicles in junctions
if not veh in self.poi_list[key] and self.player_list[veh].prev_poi != key: #this if statement for when vehicle first enter junction
#update capacity and reward first
#print('vehicle approaching poi', veh, key)
#maybe i shouldnt check for if veh at target poi if accidentally stomble then still try to collect
#current_edge = veh_value[tc.VAR_ROAD_ID]
self.track_veh = veh
self.wait_in_radius(key, veh) #this process poi function need to stop vehicle and everything else happens in the stop handle function for updating sp
self.poi_list[key][veh] = veh_value #add vahicle in poi list for removal later
elif veh in self.poi_list[key] and self.player_list[veh].prev_poi != key:
#check if it should delete vehicle
try:
if self.map_data.edges[veh_value[tc.VAR_ROAD_ID]]._to != key:
#print('vehicle left junction')
self.player_list[veh].prev_poi = key
del self.poi_list[key][veh]
self.track_veh = None
except KeyError:
#print('reached junction')
continue
class GridWin(tk.Tk):
def __init__(self, gui=True, testing='budget'):
super(GridWin, self).__init__()
traci.start(["sumo", "-c", Settings.sumo_config])
self.mode='default' #default, reward
self.testing = testing #if testing budget keep capacity constant, if testing capacity keep budget constant
self.env_map = Map(Settings.sumo_config)
self.rowcol_to_junction = self.env_map.complex_row_col
self.row = int(self.env_map.rows)
self.column = int(self.env_map.columns)
#self.rowcol_to_junction = self.env_map.row_col(self.row, self.column) #value is the junction id and key is row_col
self.rowcol_to_junction.update(dict((v, k) for k, v in self.rowcol_to_junction.items()))
self.player_list = {} #stores location as key, player object as value in list
self.reward_list = {} #stores location(grid) as key, reward value as value
self.random_uniform_reward_list = {} #akbas
self.min_reward = 0 #min reward value
self.gui = gui
self.setting = Settings()
#settings
self.reward_distribution_center = [(0,0), (0,self.column),(self.row, 0),(self.row, self.column)]
if gui:
self.protocol("WM_DELETE_WINDOW", self.on_closing)
w=self.winfo_screenwidth()
h=self.winfo_screenheight()
size = str(int(w/2))+'x'+str(int(h/2))
#print(size)
self.geometry(size)
self.import_frame = tk.Frame(self, bg='green')
#self.import_frame.pack(expand=True, fill='both')
self.grid_frame = tk.Frame(self, height=500, width=500, bg='red')
self.grid_frame.pack(expand=True, fill='both')
self.control_frame = tk.Frame(self)
self.control_frame.pack(expand=True, fill='both')
self.grid_list = [] #store all the buttons in grid
self.spawn_grid()
self.control_buttons()
self.mainloop()
def run_sim_no_gui(self, replay=False):
if (not replay) or self.testing == 'budget':
for center in self.reward_distribution_center:
self.generate_reward_spread(center[0], center[1], self.setting.reward_value_random[1], self.setting.reward_position_std, self.setting.reward_amount, mean=self.setting.reward_value_random[0])
if self.setting.percent_reward_dist:
#if we deicided to spread reward based off percentage, only need to do it once over the entire map
break
self.simulation(replay=replay)
if self.testing == 'budget':
#reset reward list if testing budget. testing budget means change budget values thus change the spread of reward every after every simulation
self.reward_list = {}
def default_mode(self):
#each grid click spawns 1 vehicle, cells black
#right click undo
for i in range(self.row):
for j in range(self.column):
string_key = str(i) + '_' + str(j)
self.grid_list[i][j].unbind('<Enter>')
self.grid_list[i][j].unbind('<Leave>')
self.grid_list[i][j].unbind('<MouseWheel>')
if string_key in self.player_list:
self.grid_list[i][j].configure(bg='black')
self.grid_list[i][j].configure(text=str(len(self.player_list[string_key])))
else:
self.grid_list[i][j].configure(bg=self.default_button_color)
self.grid_list[i][j].configure(text='')
self.grid_list[i][j].configure(command = lambda i=i, j=j: self.add_player(i, j))
self.grid_list[i][j].bind('<Button-3>', lambda event, a=i, b=j, color=self.default_button_color: self.remove_player(a, b, color))
self.grid_list[i][j].bind('<Button-2>', lambda event, a=i, b=j: self.normal_distribute_players(a,b))
self.grid_list[i][j].configure(fg='white')
def ncr(self, n, r):
r = min(r, n-r)
numer = reduce(mul, range(n, n-r, -1), 1)
denom = reduce(mul, range(1, r+1), 1)
return (numer / denom)
def spawn_reward_mode(self):
if self.mode == 'default':
self.mode='reward'
self.select_rewards.configure(text='Select Vehicles')
for i in range(self.row):
for j in range(self.column):
string_key = str(i) + '_' + str(j)
self.grid_list[i][j].configure(command=lambda i=i, j=j: self.add_reward(i,j, 1))
self.grid_list[i][j].configure(bg=self.default_button_color)
self.grid_list[i][j].configure(fg='black')
if string_key in self.reward_list:
self.grid_list[i][j].configure(text=str(self.reward_list[string_key]))
else:
self.grid_list[i][j].configure(text='0')
self.grid_list[i][j].bind('<Button-2>', lambda event, a=i, b=j: self.normal_distribute_players(a,b, False))
self.grid_list[i][j].bind('<Enter>', lambda event, i=i, j=j: self.grid_list[i][j].bind('<MouseWheel>', lambda event, i=i, j=j:self.add_reward(i,j, int(event.delta/120))))# self.reward_change(event, i, j)))
self.grid_list[i][j].bind('<Leave>', lambda event: self.grid_list[i][j].unbind('<MouseWheel>'))
self.grid_list[i][j].bind('<Button-3>', lambda event, i=i, j=j: self.reward_remove(i,j)) #this button used to clear rewards
else:
self.default_mode()
self.mode='default'
self.select_rewards.configure(text='Select Rewards')
def reward_remove(self, row, column):
string_key = str(row) + '_' + str(column)
try:
del self.reward_list[string_key]
if self.gui:
self.grid_list[row][column].configure(text='0')
except KeyError:
print('no rewards at this cell')
def add_reward(self, row, column, scroll_value):
string_key = str(row) + '_' + str(column)
if string_key in self.reward_list:
self.reward_list[string_key] += scroll_value
else:
self.reward_list[string_key] = scroll_value
if self.reward_list[string_key] < 0:
self.reward_list[string_key]=0
if self.min_reward == 0 or (self.reward_list[string_key] < self.min_reward):
self.min_reward = self.reward_list[string_key]
print(f'min reward value is {self.min_reward} at location {string_key}')
if self.gui:
self.grid_list[row][column].configure(text=str(self.reward_list[string_key]))
def normal_distribute_players(self, a,b, dist_player=True):
second_level = tk.Toplevel(self)
second_level.title('normal distribution')
std_label = tk.Label(second_level, text='STD: ').grid(row=0, column=0)
number_label = tk.Label(second_level, text='Amount: ').grid(row=1, column=0)
std_var = tk.StringVar()
num_var = tk.StringVar()
L1 = tk.Entry(second_level, justify='left', relief='ridge', background='#6699ff',textvariable=std_var)
L1.grid(row=0, column=1)
L2 = tk.Entry(second_level, justify='left', relief='ridge', background='#6699ff', textvariable=num_var)
L2.grid(row=1, column=1)
enter_button = tk.Button(second_level,text='Enter', command=lambda a=a, b=b, second_level=second_level,std=std_var, num_veh=num_var, dist_player=dist_player:self.generate_players(a,b, second_level ,std, num_var, dist_player))
enter_button.grid(row=2, column=0)
second_level.bind('<Return>', lambda event, a=a, b=b, second_level=second_level,std=std_var, num_veh=num_var:self.generate_players(a,b, second_level ,std, num_var))
def generate_players(self, a,b, root, std=None, num_var=None, dist_player=True):
if std and num_var:
std_value = float(std.get())
num_var_value = int(num_var.get())
else:
print('no std and number chosen')
#randomly generate
x = self.get_truncated_normal(a, std_value, 0, 10).rvs(num_var_value+10000).round().astype(int)
y = self.get_truncated_normal(b, std_value, 0, 10).rvs(num_var_value+10000).round().astype(int)
xPo = np.random.choice(x, num_var_value)
yPo = np.random.choice(y, num_var_value)
for x_points, y_points in zip(xPo, yPo):
if dist_player:
self.add_player(x_points, y_points)
else:
self.add_reward(x_points, y_points, 1)
root.destroy()
def generate_reward_spread(self, x, y, value_std, position_std, amount, mean=None):
if self.setting.percent_reward_dist:
print(f'length is {len(self.reward_list)}')
#if defined percentage is true then spread reward based off the percentage value randomly over the map
junction_list = np.array(list(self.env_map.junctions.keys()))
choice_junctions = [self.rowcol_to_junction[x] for x in np.random.choice(junction_list, int(len(junction_list)*self.setting.percent_reward_dist), replace=False)]
for junct_grid in choice_junctions:
row, column = int(junct_grid.split('_')[0]), int(junct_grid.split('_')[1])
if mean:
mean_dist = self.get_truncated_normal(mean, value_std, 0, 10*value_std*mean).rvs(amount+10000).astype(float)
self.add_reward(row, column, np.random.choice(mean_dist, 1)[0])
assert len(self.reward_list) <= len(junction_list), f'Error reward list greater than junction'
assert len(self.reward_list) == int(len(junction_list)*self.setting.percent_reward_dist), f'Error reward and expcted per not match expected {int(len(junction_list)*self.setting.percent_reward_dist)}, got {len(self.reward_list)} chocie {len(choice_junctions)}'
return
#if mean is not none generate distribution based off mean to add to rewards
x, y = self.find_closest_road(x, y)
x_dist = self.get_truncated_normal(x, position_std, 0, self.row).rvs(amount+10000).round().astype(int)
y_dist = self.get_truncated_normal(y, position_std, 0, self.column).rvs(amount+10000).round().astype(int)
xPo = np.random.choice(x_dist, amount)
yPo = np.random.choice(y_dist, amount)
zip_po = list(zip(xPo, yPo))
i = 0
while i < len(zip_po):
x_points, y_points = zip_po[i]
string_key = str(x_points) + '_' + str(y_points)
if not string_key in self.rowcol_to_junction:
#print(string_key, 'not in')
zip_po[i] = (np.random.choice(x_dist, 1)[0], np.random.choice(y_dist, 1)[0])
continue
if mean:
mean_dist = self.get_truncated_normal(mean, value_std, 0, 10*value_std*mean).rvs(amount+10000).astype(float)
self.add_reward(x_points, y_points, np.random.choice(mean_dist, 1)[0])
else:
self.add_reward(x_points, y_points, 1)
i+=1
#assert len(self.reward_list) == amount, f'len of rewrd amount does not match {len(self.reward_list)}'
def find_closest_road(self, x, y):
closest = None
for key, value in self.env_map.junctions.items():
key = self.rowcol_to_junction[key]
x1, y1 = int(key.split('_')[0]), int(key.split('_')[1])
dist = np.linalg.norm(np.array([x,y])-np.array([x1, y1]))
if not closest:
closest = (key, dist)
else:
if dist < closest[1]:
closest = (key, dist)
return int(closest[0].split('_')[0]), int(closest[0].split('_')[1])
def get_truncated_normal(self, mean=0, sd=1, low=0, upp=10):
return truncnorm((low - mean) / sd, (upp - mean) / sd, loc=mean, scale=sd)
def control_buttons(self):
for row in range(2):
for column in range(3):
self.control_frame.grid_columnconfigure(column, weight=1)
self.control_frame.grid_rowconfigure(row, weight=1)
start_sim_button = tk.Button(self.control_frame, text='START', command=self.simulation)
start_sim_button.grid(row=0, column=1, sticky=tk.N+tk.S+tk.W+tk.E)
self.select_rewards = tk.Button(self.control_frame, text='Select Rewards', command = self.spawn_reward_mode)
self.select_rewards.grid(row=0, column=0, sticky=tk.N+tk.S+tk.W+tk.E)
clear_button = tk.Button(self.control_frame, text='CLEAR', command=self.clear)
clear_button.grid(row=0, column=2,sticky=tk.N+tk.S+tk.W+tk.E)
self.replay_button = tk.Button(self.control_frame, text='Replay Simulation', command=self.replay_simulation)
self.replay_button.grid(row=1, column=1,sticky=tk.N+tk.S+tk.W+tk.E)
self.save_button = tk.Button(self.control_frame, text='Save Simulation', command=self.save_simulation)
self.save_button.grid(row=1, column=0,sticky=tk.N+tk.S+tk.W+tk.E)
self.load_button = tk.Button(self.control_frame, text='Load Simulation', command=self.load_simulation)
self.load_button.grid(row=1, column=2,sticky=tk.N+tk.S+tk.W+tk.E)
def load_simulation(self):
with open(Settings.sim_save, 'rb') as config_dictionary_file:
self.cap = pickle.load(config_dictionary_file)
print('simulation load sucess...')
self.replay_simulation(load=True)
def replay_simulation(self, algo, load=False):
print('player numbers total(10):',len(self.cap.player_list))
for value in self.cap.player_list:
new_player = GridPlayer(value.start, value.destination)
row, column = self.rowcol_to_junction[value.start].split('_')
if load:
new_player.node_path = value.node_hit
self.add_player(int(row), int(column), new_player)
else:
self.add_player(int(row), int(column))
self.cap.player_list=[]
return self.start_sim(algo=algo, replay=True, load=load)
def save_simulation(self):
if not self.cap:
print('No recent simulation, please run sim before save')
return
with open(Settings.sim_save, 'wb') as config_dictionary_file:
pickle.dump(self.cap, config_dictionary_file)
print('simulation saved success...')
def spawn_grid(self):
for i in range(self.row):
temp_list = []
for j in range(self.column):
b= tk.Button(self.grid_frame)
b.grid(row=i, column=j, sticky=tk.N+tk.S+tk.W+tk.E, columnspan=1)
self.default_button_color = b.cget('bg')
self.grid_frame.grid_columnconfigure(j, weight=2)
self.grid_frame.grid_rowconfigure(i, weight=2)
temp_list.append(b)
self.grid_list.append(temp_list)
self.default_mode()
def find_adjacent_cells(self, x_y, param='to'): #this function need to change to be determined using sumo
adjacent_list_sumo = [self.rowcol_to_junction[x] for x in self.env_map.find_adjacent_cells(self.rowcol_to_junction[x_y], param=param)]
return adjacent_list_sumo
def find_adjacent_players(self, x_y, list_players=False): #check for capacity
player_list = []
player_num = 0
adjacent_list = self.find_adjacent_cells(x_y, param='from')
for value in adjacent_list:
try:
player_num +=len(self.player_list[value])
for player in self.player_list[value]:
player_list.append(player)
except KeyError:
continue
if list_players:
return player_num, player_list
return player_num
def remove_player(self, row, column, color):
try:
string_key = str(row) + '_' + str(column)
if len(self.player_list[string_key]) == 1:
del self.player_list[string_key]
else:
self.player_list[string_key].pop(0)
self.env_map.junctions[self.rowcol_to_junction[string_key]].number_players -= 1
if self.env_map.junctions[self.rowcol_to_junction[string_key]].number_players == 0:
self.grid_list[row][column].configure(text='')
self.grid_list[row][column].configure(bg=color)
else:
self.grid_list[row][column].configure(text=self.env_map.junctions[self.rowcol_to_junction[string_key]].number_players)
except KeyError:
#try to delete vehicle that doesnt exist
pass
def add_player(self, row, column, player=None): # add player to dictionary and return a dict
string_key = str(row) + '_' + str(column)
destination = self.setting.destination
if destination =='random':
destination = str(randrange(self.row))+'_'+str(randrange(self.column))
if string_key == destination:
print('start is destination')
return False
try:
if player:
player_instance = player
else:
player_instance = GridPlayer(self.rowcol_to_junction[string_key], self.rowcol_to_junction[destination])
player_instance.path = self.env_map.find_best_route(player_instance.start, player_instance.destination)
if not (player_instance.path and player_instance.path.edges):
print('no path edges')
return False
player_instance.node_path = [self.env_map.edges[x]._to for x in player_instance.path.edges]
player_instance.capacity = self.get_truncated_normal(self.setting.player_capacity_random[0], self.setting.player_capacity_random[1], 0, self.setting.player_capacity_random[0]*2).rvs(1)[0]
#print('plyer capacity is: ', player_instance.capacity)
except KeyError as e:
print(f'no key in dict {e}')
return False
print(f'vehicle start at {string_key} end at {destination} capacity value is {player_instance.capacity}, reward list len is {len(self.reward_list)}')
if string_key in self.player_list:
self.player_list[string_key].append(player_instance)
else:
self.player_list[string_key] = [player_instance]
self.env_map.junctions[self.rowcol_to_junction[string_key]].number_players += 1
if self.gui:
self.grid_list[row][column].configure(bg='black')
self.grid_list[row][column].configure(text=self.env_map.junctions[self.rowcol_to_junction[string_key]].number_players)
return True
def redirect_route(self, player_instance, new_route):
if new_route[0] != player_instance.destination:
player_instance.path = new_route # self.env_map.find_best_route(next_node, player_instance.destination)
player_instance.node_path = [self.env_map.edges[x]._to for x in player_instance.path.edges]
player_instance.node_index=0
return player_instance
def GTA_next_node(self, location, player_instance, current_node):
#takes in location, player, and next node return new next node and player instance
cells = self.find_adjacent_cells(location) #return grid junction
assert cells, f'Junction has no where else to go {location}'
new_route = player_instance.path
#print('new route is ', new_route)
max_utility = 0
max_utility_cell = None
for cell in cells:
#print(f'cells value is {cells}')
adjacent_players, player_list = self.find_adjacent_players(cell, list_players=True)
assert adjacent_players > 0, f'player is {adjacent_players} failed, location: {location}, surrounding cells: {cells}, current: {cell}, player: {self.player_list[location]}'
cell_utility = 0 #total utility at that cell
try:
cell_utility = self.reward_list[cell]
if cell_utility == 0:
continue
except KeyError:
continue
expected_utility = 0
expected_sensing_plan = 0
if adjacent_players == 1: #no one is around me
expected_utility = cell_utility
#expected_sensing_plan = self.env_map.junctions[self.rowcol_to_junction[cell]].cost
expected_sensing_plan = 1
else:
#sum the denomenator of the combination, store ncr in list to reuse later
ncr_list = [self.ncr(adjacent_players-1, player_number) for player_number in range(0, adjacent_players)] #consists of a list from 0
denom = reduce(lambda x,y: x+y, ncr_list) #from 0 to second to last in list
for current_player in range(1, i+1):
numerator = ncr_list[current_player-1] #retrieve ncr value from list
prM = numerator/denom
expected_utility += (prM*(cell_utility/pow(current_player, 2)))
expected_sensing_plan += (prM*(self.compute_sensing_plan(current_player, self.reward_list[cell], self.env_map.junctions[self.rowcol_to_junction[cell]].cost)))
if (expected_utility > max_utility) and (expected_sensing_plan <= player_instance.capacity) and (not cell in player_instance.past_recent_nodes):
#choose highest utility, with capacity taken into consideration, as well as no repeating within last 3 visited
max_utility = expected_utility
max_utility_cell = cell
#this part to generate the weighted random path
if not max_utility_cell:
#if adj cells no rewards
#if adj cells rewards dont fit capacity
#weighted random need to priotize coverage good
weight_dict, best_route = self.env_map.find_best_route(current_node, player_instance.destination, weights=True)
#weight_dict = {key:value.travelTime*self.visited_cell_list[key]*len(value.edges) if key in self.visited_cell_list else value.travelTime for key, value in weight_dict.items()}
#travel time multiple by cell visited number , multiple by number of edges before reaching destination
try:
total_sum = reduce(lambda x,y:x+y,[exp(self.setting.theta_random/x.travelTime) for x in weight_dict.values()])
prob_distribute = [exp(self.setting.theta_random/x.travelTime)/total_sum for x in weight_dict.values()]
selected_index = np.random.choice(len(cells), 1, p=prob_distribute)
next_node = list(weight_dict.keys())[selected_index[0]]
except OverflowError:
#when theta random value is too large just take the best route node
next_node = self.env_map.edges[best_route.edges[0]]._to
else:
#when max utility cells is found its in terms of grid junction need to convert to sumo junction
next_node = self.rowcol_to_junction[max_utility_cell]
player_instance = self.redirect_route(player_instance, new_route)
player_instance = self.add_stm(player_instance, next_node)
return next_node, player_instance
def add_stm(self, player_instance, next_node):
if len(player_instance.past_recent_nodes) < self.setting.max_memory_size:
player_instance.past_recent_nodes.append(self.rowcol_to_junction[next_node]) #sumo junction is added to memory
else:
player_instance.past_recent_nodes.pop(0)
player_instance.past_recent_nodes.append(self.rowcol_to_junction[next_node])
return player_instance
def simulation(self, replay=False):
for algo in self.setting.game_theory_algorithm:
multi_data = MultiCapture('Traffic Simulation')
i=0
while i < self.setting.simulation_steps:
print('iteration ', i)
if replay:
suc = self.replay_simulation(algo=algo)
else:
suc = self.start_sim(algo=algo) #first time running simulation generate random players save it to self.cap
replay=True
if suc: #if simulation is success
self.cap.setting = self.setting
self.cap.reward_list = self.reward_list
self.cap.reward_junction_ratio = len(self.reward_list)/len(self.env_map.junctions)
cov = self.cap.calculate_coverage()
test_cov = self.cap.calculate_test_coverage()
print('road utilization is ', cov)
print('coverage is ', test_cov)
multi_data.simulation_conv_list.append(cov)
multi_data.simulation_test_coverage.append(test_cov)
multi_data.simulation_list.append(self.cap)
#if i%10 ==0:
if i == (self.setting.simulation_steps-1):
if self.setting.percent_reward_dist:
multi_data.pickle_save(os.path.join(Settings.sim_save_path, f'{self.setting.percent_reward_dist}_reward{self.setting.reward_value_random[0]}_capacity{self.setting.player_capacity_random[0]}_Step{self.setting.simulation_steps}_{algo}_cluster_reward.sim'))
else:
multi_data.pickle_save(os.path.join(Settings.sim_save_path, f'reward{self.setting.reward_value_random[0]}_capacity{self.setting.player_capacity_random[0]}_Step{self.setting.simulation_steps}_{algo}_cluster_reward.sim'))
i+=1
self.player_list = {}
#changing rewards should be before this funtion
if self.testing =='budget' and not self.setting.percent_reward_dist: #when running budget distribute based off percentage no need to run it second time
temp_multi_data = MultiCapture('Traffic Simulation')
temp = self.reward_list #save the temp reward list before replacing it with uniform reward.
j = 0
while j < self.setting.simulation_steps:
#this is for random uniform reward distribution
self.reward_spread_uniform((multi_data.simulation_list[0].reward_junction_ratio)*len(self.env_map.junctions))
suc = self.replay_simulation(algo=algo)
if suc:
cov = self.cap.calculate_coverage()
test_cov = self.cap.calculate_test_coverage()
print('RU for random reward ', cov)
print('COV for random reward ', test_cov)
temp_multi_data.simulation_conv_list.append(cov)
temp_multi_data.simulation_test_coverage.append(test_cov)
temp_multi_data.simulation_list.append(self.cap)
if j == (self.setting.simulation_steps-1):
temp_multi_data.pickle_save(os.path.join(Settings.sim_save_path, f'reward{self.setting.reward_value_random[0]}_capacity{self.setting.player_capacity_random[0]}_Step{self.setting.simulation_steps}_{algo}_random_reward.sim'))
j+=1
self.player_list = {}
self.reward_list = temp
def reward_spread_uniform(self, amount_rewards):
#spread rewards uniformly based of ratio
self.reward_list = {}
if self.random_uniform_reward_list:
for key, value in self.random_uniform_reward_list.items():
self.random_uniform_reward_list[key] = randrange(self.setting.reward_value_random[0]-self.setting.reward_value_random[1], self.setting.reward_value_random[0]+self.setting.reward_value_random[1])
self.reward_list = self.random_uniform_reward_list
else:
reward_locations = np.random.choice(list(self.env_map.junctions.keys()), round(amount_rewards))
for value in reward_locations:
value = self.rowcol_to_junction[value]
x,y = int(value.split('_')[0]), int(value.split('_')[1])
self.add_reward(x,y,randrange(self.setting.reward_value_random[0]-self.setting.reward_value_random[1], self.setting.reward_value_random[0]+self.setting.reward_value_random[1]))
def greedy_next_node(self, location, player_instance, current_node):
'''
location current location in terms of sumo
player instance the player we are looking at
current node is the location in terms of grid values
next_node is predicted next node based on dex
'''
shortest_path = False
cells = self.find_adjacent_cells(location)
#cells return grid location not sumo location
#reward list returns grid location as well
max_utility_cell = None
max_utility = 0
for i, cell in enumerate(cells):
try:
self.reward_list[cell]
if not max_utility_cell and (self.reward_list[cell]<=player_instance.capacity):
max_utility_cell = cell
max_utility = self.reward_list[cell]
else:
if (self.reward_list[cell] > max_utility) and (self.reward_list[cell]<=player_instance.capacity):
max_utility_cell = cell
max_utility = self.reward_list[cell]
except KeyError as e:
#print(f'im here {i}/{len(cells)}')
assert not cell in self.reward_list, f"failed {cell} is in reward_list"
pass
#two cases, when no rewards around, or
#find smallest reward, if capacity falls below smallest reward, uses djk instead
if not max_utility_cell:
#let random jump decide on capacity that way greedy no need to implement
next_node, player_instance, shortest_path = self.random_next_node(location, player_instance, current_node) #remove random directly go towards destination shortest path
#next node for random returns sumo cells
else:
next_node = max_utility_cell
assert max_utility_cell in self.reward_list, f'something is wrong {max_utility_cell} supose in reward list'
player_instance.collected_sp_list.append(next_node)
#next node normally return grid cells
next_node = self.rowcol_to_junction[next_node]
player_instance.capacity -= max_utility
player_instance.reward += max_utility
#this calculation is done after running
#player_instance = self.add_stm(player_instance, next_node)
return next_node, player_instance, shortest_path
def random_next_node(self, location, player_instance, current_node):
# capacity for random
shortest_path = False
cells = self.find_adjacent_cells(location)
weight_dict, best_route = self.env_map.find_best_route(current_node, player_instance.destination, weights=True)
try:
total_sum = reduce(lambda x,y:x+y,[exp(self.setting.theta_random/x.travelTime) for x in weight_dict.values()])
prob_distribute = [exp(self.setting.theta_random/x.travelTime)/total_sum for x in weight_dict.values()]
#print('prob_distribute ', prob_distribute)
#print('max value is {}, index value is {}, the next cell is {}, current cell is {}'.format(max(prob_distribute), prob_distribute.index(max(prob_distribute)), self.rowcol_to_junction[list(weight_dict.keys())[prob_distribute.index(max(prob_distribute))]], self.rowcol_to_junction[current_node]))
selected_index = np.random.choice(len(cells), 1, p=prob_distribute)
next_node = list(weight_dict.keys())[selected_index[0]]
except OverflowError:
#when theta random value is too large just take the best route node
next_node = self.env_map.edges[best_route.edges[0]]._to
if self.rowcol_to_junction[next_node] in self.reward_list: #if next node contains rewards
if self.reward_list[self.rowcol_to_junction[next_node]] <= player_instance.capacity: #if reward is less than capacity then collect
player_instance.collected_sp_list.append(self.rowcol_to_junction[next_node])
player_instance.capacity -= self.reward_list[self.rowcol_to_junction[next_node]]
player_instance.reward += self.reward_list[self.rowcol_to_junction[next_node]]
else: #cant collect due to capacity
if player_instance.capacity < self.min_reward: #check is player capacity less than min reward on map
shortest_path = True
player_instance.path = self.env_map.find_best_route(current_node, player_instance.destination)
player_instance.node_path = [self.env_map.edges[x]._to for x in player_instance.path.edges]
player_instance.node_index = 0 #reset path info
#need to reset player path setting index =0
#player_instance = self.add_stm(player_instance, next_node)
return next_node, player_instance, shortest_path
def start_sim(self, algo, replay=False, load=False):
shortest_path = False
#print('len reward is ', len(self.reward_list))
#simulation start
if self.gui:
self.default_mode()
if not replay: self.cap = DataCapture(len(self.env_map.junctions), self.rowcol_to_junction) #reset if its not replay
#if no predefined players, randomly spawn players
if not self.player_list:
i = 0
while i < self.setting.car_numbers:
row_col = choice(list(self.env_map.junctions.keys()))
row, column = self.rowcol_to_junction[row_col].split('_')[0], self.rowcol_to_junction[row_col].split('_')[1]
suc = self.add_player(row, column)
if suc:
i +=1
print(f'player added to {row}, {column}')
else:
print(f'failed to add player at {row}, {column}')
arrived_locations = [] #destinations of all vehicles to reset after
while self.player_list:
player_left = 0
if self.gui:
self.update()
time.sleep(self.setting.simulation_delay)
temp_dict = {}
for location, players in self.player_list.items():
for i, player in enumerate(players):
player_left+=1
#junction value in sumo
#insert logic for game theory,
if algo=='gta' and not load: #if its loading from file then just play players
next_node, player = self.GTA_next_node(location, player, self.rowcol_to_junction[location])
elif algo=='greedy' and not load:
#run greedy algo, only go towards highest rewards. check capacity and reduce capacity based on reward value else result in infinite loop
if not shortest_path:
next_node, player, shortest_path = self.greedy_next_node(location, player, self.rowcol_to_junction[location])
if shortest_path:
print('player taking shortest path in greedy')
next_node = player.get_next()
elif algo=='random' and not load:
if not shortest_path:
next_node, player, shortest_path = self.random_next_node(location, player, self.rowcol_to_junction[location])
if shortest_path:
print('player taking shortest path in random')
next_node = player.get_next()
elif algo == 'base' and not load:
next_node = player.get_next()
if self.rowcol_to_junction[next_node] in self.reward_list:
if self.reward_list[self.rowcol_to_junction[next_node]] <= player.capacity:
player.collected_sp_list.append(self.rowcol_to_junction[next_node])
player.capacity -= self.reward_list[self.rowcol_to_junction[next_node]]
player.reward += self.reward_list[self.rowcol_to_junction[next_node]]
player.node_hit.append(next_node) # for post processing
player.reward_hit.append(player.capacity) # for post processing
button_name = self.rowcol_to_junction[next_node]
button_row, button_column = button_name.split('_')
if next_node == player.destination:
print(f'player has arrived to {self.rowcol_to_junction[player.destination]} from {self.rowcol_to_junction[player.start]} nodes traveled:{len(player.node_hit)}')
arrived_locations.append(player.destination)
self.cap.player_list.append(player) #add player to the post processing list
else:
#if final destination is not reached add it to the temp dict
if button_name in temp_dict:
temp_dict[button_name].append(player)
else:
temp_dict[button_name] = [player]
self.env_map.junctions[self.rowcol_to_junction[button_name]].number_players += 1
self.env_map.junctions[self.rowcol_to_junction[location]].number_players -= 1
if self.gui:
self.grid_list[int(button_row)][int(button_column)].configure(bg='black')
self.grid_list[int(button_row)][int(button_column)].configure(text=self.env_map.junctions[self.rowcol_to_junction[button_name]].number_players)
#every time a player move away check if the edge contains more players
player_number = self.env_map.junctions[self.rowcol_to_junction[location]].number_players
prev_button_row, prev_button_column = location.split('_')
if self.gui:
if player_number == 0:
self.grid_list[int(prev_button_row)][int(prev_button_column)].configure(bg='white')
else:
self.grid_list[int(prev_button_row)][int(prev_button_column)].configure(text=player_number)
self.player_list = temp_dict
print(f'{player_left} remaining')
#if capacity too low make random jumps towards destination
if algo=='gta': #reduce capacity based on sensing plan
for location, players in self.player_list.items():
print('location is ', location)
location_cost = self.env_map.junctions[self.rowcol_to_junction[location]].cost
number_players = self.adjusting_sensing_plan(players, location, location_cost) #will return key error if there is no reward at location
#print(f"{number_players} _ number players_ {}")
for i, player in enumerate(players):
try:
self.player_list[location][i].reward += (self.reward_list[location]/len(self.player_list[location])) #question mark.. if location no reward dont calculate sensing plan
player_sensing_plan = self.compute_sensing_plan(number_players, self.reward_list[location], location_cost)
#whats the default expected sensing plan when there is only 1 player??
if self.reward_list[location] != 0:
# if you arrived and your cost is more than your player capacity might as well take what ever your capacity can handle
if player_sensing_plan <= player.capacity:
self.player_list[location][i].collected_sp_list.append(location)
self.player_list[location][i].capacity -= player_sensing_plan
if self.player_list[location][i].capacity < 0:
self.player_list[location][i].capacity = 0
else:
print(f'{player_sensing_plan} and cap is {player.capacity}')
except KeyError as e:
#print(f'no rewards at {location}')
continue
except TypeError as t:
#print(f'no player matching adjust at {location}, {number_players}')
#type error occurs when ajusting sensing plan returns None
continue
print('simulation completed')
self.reset_junction_players(arrived_locations)
return True
def compute_sensing_plan(self, player_amount, reward, cost):
if player_amount ==1:
return 1
return ((player_amount-1)*reward)/((player_amount**2)*cost)
def adjusting_sensing_plan(self, players, location, location_cost):
#return key error because the location has no reward?
try:
for i in range(len(players), 0, -1):
esp = self.compute_sensing_plan(i, self.reward_list[location], location_cost)
counter = 0 #this to count how many fits the capacity
for player in players:
if esp<=player.capacity:
counter+=1
print(f'esp:{esp} cap:{player.capacity} len:{len(players)}')
if counter == i:
return counter
print(f'i find no one matching sensing plan')
return
except KeyError:
print('i sense no rewards')
return
def clear(self):
if self.mode == 'default':
self.player_list = {}
self.default_mode()
else:
self.mode='default'
self.reward_list = {}
self.spawn_reward_mode()
def reset_junction_players(self, arrived_locations):
for value in set(arrived_locations):
self.env_map.junctions[value].number_players = 0
def on_closing(self):
self.destroy()
traci.close()