def run_optimality_evaluation(G, model, randomness):
model.reset_states()
try:
start_node_id = np.random.choice(G.nodes)
goal_node_id = np.random.choice(G.nodes)
try:
truth_route = nx.shortest_path(G,
start_node_id,
goal_node_id,
weight='best_travel_time')
except nx.NetworkXNoPath:
print("[ERROR] No path found")
return None, None
curr_node_id = start_node_id
ml_route = []
visited = collections.deque(maxlen=7)
# Start node
ml_route.append(curr_node_id)
use_random_actions = False
gt_duration = osmnx_utils.get_route_duration(truth_route, G)
history = collections.deque(maxlen=5)
while curr_node_id != goal_node_id:
curr_node_id = algorithm_commons.predict_next_node_id(
G, curr_node_id, goal_node_id, model, randomness=randomness)
if curr_node_id:
ml_route.append(curr_node_id)
if osmnx_utils.get_route_duration(ml_route, G) > gt_duration * 1.5:
print("x2 length, try again...")
return None, None
ml_duration = osmnx_utils.get_route_duration(ml_route, G)
# print("------ METRICS ------")
# print("GT duration: ", gt_duration, ". ML duration: ", ml_duration, ". Abs Diff: ", ml_duration - gt_duration, ". Ratio: ", gt_duration / ml_duration)
# print("---------------------")
# ##### PLOT GT AND ML PATHS ######
# ox.plot_graph_route(G, truth_route)
# ox.plot_graph_route(G, ml_route)
if gt_duration == 0 or ml_duration == 0:
return None, None
else:
return ml_duration, gt_duration
except nx.NetworkXError:
return None, None
def run_optimality_evaluation(G):
start_node_id = np.random.choice(G.nodes)
goal_node_id = np.random.choice(G.nodes)
truth_route = nx.shortest_path(G,
start_node_id,
goal_node_id,
weight='best_travel_time')
curr_node_id = start_node_id
ml_route = []
visited = collections.deque(maxlen=7)
# Start node
ml_route.append(curr_node_id)
use_random_actions = False
history = collections.deque(maxlen=5)
while curr_node_id != goal_node_id:
curr_node_id = predict_next_node_id(G,
curr_node_id,
goal_node_id,
model,
randomness=0.1)
ml_route.append(curr_node_id)
gt_duration = get_route_duration(truth_route, G)
ml_duration = get_route_duration(ml_route, G)
print("------ METRICS ------")
print("GT duration: ", gt_duration, ". ML duration: ", ml_duration,
". Abs Diff: ", ml_duration - gt_duration, ". Ratio: ",
gt_duration / ml_duration)
print("---------------------")
###### PLOT GT AND ML PATHS ######
ox.plot_graph_route(G, truth_route)
ox.plot_graph_route(G, ml_route)
def run_generalization_evaluation(G, model):
experiment_count = 50
experiment_node_pairs = [(np.random.choice(G.nodes),
np.random.choice(G.nodes))
for i in range(1, experiment_count)]
# Eval optimality and arrival rate to later check how
# dynamic network changes affect these
# optimalities = []
# for start_goal_pair in experiment_node_pairs:
# ml_route = []
# optimal_route = []
# curr_node_id = start_goal_pair[0]
# goal_node_id = start_goal_pair[1]
# before_modification_node_id = curr_node_id
# while curr_node_id != start_goal_pair[1]:
# for i in range(0, 5):
# curr_node_id = predict_next_node_id(G, curr_node_id, goal_node_id, model)
# ml_route.append(curr_node_id)
# optimal_route = utils.split_list_at_value(nx.shortest_path(G, before_modification_node_id,
# goal_node_id, weight='best_travel_time'), curr_node_id)
# optimalities.append(get_route_duration(ml_route, G) / get_route_duration(optimal_route, G))
# ml_route = []
# optimal_route = []
# before_modification_node_id = curr_node_id
# # Modify graph
# print("Optimality rate (dynamic): ", reduce(lambda a,b: a + b, optimalities) / len(optimalities))
optimalities_static = []
for start_goal_pair in experiment_node_pairs:
ml_route = []
optimal_route = []
curr_node_id = start_goal_pair[0]
goal_node_id = start_goal_pair[1]
before_modification_node_id = curr_node_id
while curr_node_id != start_goal_pair[1]:
curr_node_id = predict_next_node_id(G, curr_node_id, goal_node_id,
model)
ml_route.append(curr_node_id)
optimal_route = nx.shortest_path(G,
start_goal_pair[0],
goal_node_id,
weight='best_travel_time')
optimalities_static.append(
get_route_duration(ml_route, G) /
get_route_duration(optimal_route, G))
random_network_modification(G)
print(
"Optimality rate (dynamic): ",
reduce(lambda a, b: a + b, optimalities_static) /
len(optimalities_static))
optimalities_static = []
for start_goal_pair in experiment_node_pairs:
ml_route = []
optimal_route = []
curr_node_id = start_goal_pair[0]
goal_node_id = start_goal_pair[1]
before_modification_node_id = curr_node_id
while curr_node_id != start_goal_pair[1]:
curr_node_id = predict_next_node_id(G, curr_node_id, goal_node_id,
model)
ml_route.append(curr_node_id)
optimal_route = nx.shortest_path(G,
start_goal_pair[0],
goal_node_id,
weight='best_travel_time')
optimalities_static.append(
get_route_duration(ml_route, G) /
get_route_duration(optimal_route, G))
print(
"Optimality rate (static): ",
reduce(lambda a, b: a + b, optimalities_static) /
len(optimalities_static))
def run_generalization_evaluation(G, model):
experiment_count = 50
experiment_node_pairs = [(np.random.choice(G.nodes),
np.random.choice(G.nodes))
for i in range(1, experiment_count)]
optimalities_static = []
for start_goal_pair in experiment_node_pairs:
ml_route = []
optimal_route = []
curr_node_id = start_goal_pair[0]
goal_node_id = start_goal_pair[1]
before_modification_node_id = curr_node_id
while curr_node_id != start_goal_pair[1]:
curr_node_id = predict_next_node_id(G, curr_node_id, goal_node_id,
model)
ml_route.append(curr_node_id)
optimal_route = nx.shortest_path(G,
start_goal_pair[0],
goal_node_id,
weight='best_travel_time')
optimalities_static.append(
get_route_duration(ml_route, G) /
get_route_duration(optimal_route, G))
random_network_modification(G)
print(
"Optimality rate (dynamic): ",
reduce(lambda a, b: a + b, optimalities_static) /
len(optimalities_static))
optimalities_static = []
for start_goal_pair in experiment_node_pairs:
ml_route = []
optimal_route = []
curr_node_id = start_goal_pair[0]
goal_node_id = start_goal_pair[1]
before_modification_node_id = curr_node_id
while curr_node_id != start_goal_pair[1]:
curr_node_id = predict_next_node_id(G, curr_node_id, goal_node_id,
model)
ml_route.append(curr_node_id)
optimal_route = nx.shortest_path(G,
start_goal_pair[0],
goal_node_id,
weight='best_travel_time')
optimalities_static.append(
get_route_duration(ml_route, G) /
get_route_duration(optimal_route, G))
print(
"Optimality rate (static): ",
reduce(lambda a, b: a + b, optimalities_static) /
len(optimalities_static))