def __build_steiner_forest(self):
forest = SuitabilityGraph()
dist = cost = occupancy = 0
for t, dest in self.__confirmed.iteritems():
if dest is None:
dest = self.__medoids[t]
self.__confirmed[t] = dest
for t_ in self.__term_tree[t]:
self.__detour[t_] += self.__graph.dist[tuple(
sorted([t, dest]))]
# # Print groups of terminals that travel together.
# if dest in self.__pois:
# print self.__term_tree[t]
# print t, dest
self.__graph.compute_dist_paths(origins=[t],
destinations=[dest],
recompute=True)
try:
forest.append_path(
self.__graph.paths[tuple(sorted([t, dest]))], self.__graph)
dist = self.__graph.dist[tuple(sorted([t, dest]))]
cost += dist
except KeyError:
# pdb.set_trace()
print 'key error:', t, dest
# Who is driver?
# if t in self.__terminals and (dist > max_wd or dest == self.__medoids[t]):
# print "Driver:", t, "Meeting point:", dest
occupancy += len(self.__term_tree[t]) * dist
# print "Num. terms:", len(self.__term_tree[t]), "Shared distance:", dist
return forest, cost, occupancy / cost
def steiner_tree(self):
subtrees = []
_, paths = dijkstra(self.__graph, self.__poi, self.__terminals)
for t in self.__terminals:
path = paths[t]
subtree = SuitabilityGraph()
# j = 0
# for i in range(len(path_to_poi)):
# if path_to_poi[i] in self.__graph.contracted_regions:
# region_id = path_to_poi[i]
# subtree.append_from_path(path_to_poi[j:i], self.__original_graph)
# closest_node = self.__find_closest_node_to_poi_within_region(region_id)
# path_endpoint_1 = self.__dist_paths_node_within_region_node[closest_node][1][path_to_poi[i - 1]]
# subtree.append_from_path(path_endpoint_1, self.__original_graph)
# path_endpoint_2 = self.__dist_paths_node_within_region_node[closest_node][1][path_to_poi[i + 1]]
# subtree.append_from_path(path_endpoint_2, self.__original_graph)
# j = i + 1
# subtree.append_from_path(path_to_poi[j:], self.__original_graph)
subtree.append_path(path, self.__graph)
subtrees.append(subtree)
steiner_tree = self.__merge_subtrees(subtrees)
self.__prune_steiner_tree(steiner_tree)
if self.__contract_graph:
self.__decontract_steiner_tree(steiner_tree)
return steiner_tree
class Baltz:
def __init__(self, graph):
# Init some instance variables.
self.__graph = SuitabilityGraph()
self.__graph.append_graph(graph)
self.__edges = graph.get_edges()
self.__congestion = {}
l_G = float(0)
for e, le in self.__edges.iteritems():
self.__congestion[e] = 0
l_G += le
self.__costs = {e: le / l_G for e, le in self.__edges.iteritems()}
def steiner_forest(self, requests, k=4):
lambda_ = 1
MSTs = {}
n = len(self.__graph)
for i, r in enumerate(requests):
self.__graph.update_edge_weights({
e: self.__costs[e] * n**(self.__congestion[e] / lambda_ - 1)
for e in self.__edges
})
mz = VST_RS(self.__graph)
Ti, l, _, _, _, _, _ = mz.steiner_forest(r[1:], [r[0]], k,
sys.maxint)
MSTs[i] = (Ti, l)
for e in Ti.get_edges():
self.__congestion[e] += 1
lambda_ = max(lambda_, self.__congestion[e])
iteration = 1
while iteration <= 100:
for i, r in enumerate(requests):
cmax = max(self.__congestion.values())
E_Ti = MSTs[i][0].get_edges()
A = len(E_Ti)
self.__graph.update_edge_weights({
e: self.__costs[e] * A**(self.__congestion[e] - cmax)
for e in self.__edges
})
self.__graph.update_edge_weights(
{e: le / A
for e, le in E_Ti.iteritems()})
mz = VST_RS(self.__graph)
Ti_, l, _, _, _, _, _ = mz.steiner_forest(
r[1:], [r[0]], k, sys.maxint)
self.__graph.update_edge_weights(
{e: le * A
for e, le in E_Ti.iteritems()})
if MSTs[i][1] > l:
for e in MSTs[i][0].get_edges():
self.__congestion[e] -= 1
for e in Ti_.get_edges():
self.__congestion[e] += 1
MSTs[i] = (Ti_, l)
iteration += 1
return MSTs
def enclosing_region(self):
enclosing = SuitabilityGraph()
paths = []
paths.append(self.__paths[tuple(sorted([323287670, 2392803740]))])
paths.append(self.__paths[tuple(sorted([2392803740, 127578100]))])
paths.append(self.__paths[tuple(sorted([127578100, 3109398450]))])
paths.append(self.__paths[tuple(sorted([3109398450, 342909685]))])
paths.append(self.__paths[tuple(sorted([342909685, 323287670]))])
for path in paths:
enclosing.append_path(path, self.__original_graph)
return enclosing
def __init__(self, graph):
# Init some instance variables.
self.__graph = SuitabilityGraph()
self.__graph.append_graph(graph)
self.__edges = graph.get_edges()
self.__congestion = {}
l_G = float(0)
for e, le in self.__edges.iteritems():
self.__congestion[e] = 0
l_G += le
self.__costs = {e: le / l_G for e, le in self.__edges.iteritems()}
def __init__(self, graph, terminals, contract_graph=True, contracted_graph=None, within_convex_hull=False,
dist_paths=None, nodes=None, use_medoid=False):
# Check whether graph is node-weighted.
if not graph.is_node_weighted():
raise (ValueError, "Dreyfus with IMRs algorithm only works with node-weighted graphs.")
# Extract POI from the terminals list.
if len(terminals) > 0:
self.__poi = terminals[0]
else:
return
# Set object variables.
generator = SuitableNodeWeightGenerator()
self.__original_graph = graph
self.__terminals = terminals
self.__contract_graph = contract_graph
self.__use_medoid = use_medoid
# Contracted graph...
if contract_graph:
if contracted_graph is not None:
self.__graph = contracted_graph.copy()
else:
self.__graph = SuitabilityGraph()
self.__graph.append_graph(graph)
self.__graph.contract_suitable_regions(generator, excluded_nodes=terminals,
get_centroid_medoid=use_medoid)
else:
self.__graph = SuitabilityGraph()
self.__graph.append_graph(graph)
#
if nodes is not None:
self.__nodes = list(nodes)
else:
if within_convex_hull:
pass
# self.__nodes = self.__graph.get_suitable_nodes_within_convex_set(terminals, generator, dist_paths)
else:
self.__nodes = self.__graph.get_suitable_nodes(generator, excluded_nodes=terminals)
#
for t in terminals:
self.__nodes.append(t)
# print(self.__nodes)
#
self.__dist = {}
self.__paths = {}
if dist_paths is not None:
self.__dist = dict(dist_paths[0])
self.__paths = dict(dist_paths[1])
else:
self.__dist, self.__paths = self.__graph.get_dist_paths(origins=self.__nodes, destinations=self.__nodes)
#
self.__s_d = {}
def __init__(self, graph, terminals):
# Check whether graph is node-weighted.
if not graph.is_node_weighted():
raise (
ValueError,
"Cluster-based algorithm only works with node-weighted graphs."
)
# Extract POI from the terminals list.
if len(terminals) > 0:
self.__poi = terminals[0]
else:
return
#
generator = SuitableNodeWeightGenerator()
# Set object variables.
self.__graph = SuitabilityGraph()
self.__graph.append_graph(graph)
self.__terminals = terminals
#
# self.__regions = self.__graph.get_suitable_regions(generator, excluded_nodes=terminals,
# get_border_internal_nodes=True, get_centroid_medoid=True,
# get_characteristic_nodes=True)
self.__regions = self.__graph.get_suitable_regions(
generator,
excluded_nodes=terminals,
get_border_internal_nodes=True,
get_centroid_medoid=True)
#
self.__nodes = []
for id_r in self.__regions:
#
# characteristic_nodes = self.__regions[id_r][6]
# self.__nodes.extend(characteristic_nodes)
border_nodes = self.__regions[id_r][1]
self.__nodes.extend(border_nodes)
#
# if len(characteristic_nodes) == 0:
# medoid = self.__regions[id_r][4]
# # print(medoid)
# self.__nodes.append(medoid)
for t in terminals:
self.__nodes.append(t)
#
self.__dist, self.__paths = self.__graph.get_dist_paths(
origins=self.__nodes, destinations=self.__nodes)
def get_suitability_graph_from_session(request):
graph = request.session['graph']
suitability_graph = SuitabilityGraph()
for node_id, (node_weight, adj_dict, data) in graph.iteritems():
new_adj_dict = {int(neighbour): edge_cost for neighbour, edge_cost in adj_dict.iteritems()}
suitability_graph[int(node_id)] = (node_weight, new_adj_dict, data)
dist = {}
for k, v in request.session['dist'].iteritems():
k_ = k.split(",")
dist[(long(k_[0]), long(k_[1]))] = v
suitability_graph.dist = dist
pairs = set()
for k in request.session['pairs_dist_paths']:
k_ = k.split(",")
pairs.add((long(k_[0]), long(k_[1])))
suitability_graph.pairs_dist_paths = pairs
return suitability_graph
def get_suitability_graph_from_session(request):
graph = request.session['graph']
suitability_graph = SuitabilityGraph()
for node_id, (node_weight, adj_dict, data) in graph.iteritems():
new_adj_dict = {
int(neighbour): edge_cost
for neighbour, edge_cost in adj_dict.iteritems()
}
suitability_graph[int(node_id)] = (node_weight, new_adj_dict, data)
return suitability_graph
def __build_steiner_tree(self, node, subset):
steiner_tree = SuitabilityGraph()
next_node = self.__s_d[node][subset][0][3]
print(node, self.__s_d[node][subset])
# pdb.set_trace()
if next_node is not None:
try:
steiner_tree.append_path(
self.__paths[tuple(sorted([node, next_node]))],
self.__graph)
except KeyError:
_, paths = dijkstra(self.__graph, node, [next_node])
steiner_tree.append_path(paths[next_node], self.__graph)
(set_e, set_f) = self.__s_d[node][subset][0][4]
steiner_branch_e = SuitabilityGraph()
if set_e is not None and set_e != [next_node]:
steiner_branch_e = self.__build_steiner_tree(next_node, set_e)
steiner_branch_f = SuitabilityGraph()
if set_f is not None and set_f != [next_node] and len(set_f) > 0:
steiner_branch_f = self.__build_steiner_tree(next_node, set_f)
steiner_tree.append_graph(steiner_branch_e)
steiner_tree.append_graph(steiner_branch_f)
return steiner_tree
def __build_steiner_tree_bactracking(self, node, subset):
steiner_tree = SuitabilityGraph()
next_node = self.__steiner_distances[node][tuple(subset)][3]
# pdb.set_trace()
if next_node is not None:
try:
steiner_tree.append_path(
self.__paths[tuple(sorted([node, next_node]))],
self.__graph)
except KeyError:
_, paths = dijkstra(self.__graph, node, [next_node])
steiner_tree.append_path(paths[next_node], self.__graph)
(best_e, best_f) = self.__steiner_distances[node][tuple(subset)][4]
steiner_branch_e = SuitabilityGraph()
if best_e is not None and best_e != [next_node]:
steiner_branch_e = self.__build_steiner_tree_bactracking(
next_node, best_e)
steiner_branch_f = SuitabilityGraph()
if best_f is not None and best_f != [next_node] and len(best_f) > 0:
steiner_branch_f = self.__build_steiner_tree_bactracking(
next_node, best_f)
steiner_tree.append_graph(steiner_branch_e)
steiner_tree.append_graph(steiner_branch_f)
return steiner_tree
def __build_steiner_tree_bactracking(self, node, subset):
steiner_tree = SuitabilityGraph()
next_node = self.__s_d[node][tuple(subset)][1]
if self.__contract_graph:
print(node, self.__s_d[node][tuple(subset)])
# pdb.set_trace()
if next_node is not None:
steiner_tree.append_path(self.__paths[tuple(sorted([node, next_node]))], self.__graph)
(best_e, best_f) = self.__s_d[node][tuple(subset)][2]
# pdb.set_trace()
steiner_branch_e = SuitabilityGraph()
if best_e is not None and best_e != [next_node]:
steiner_branch_e = self.__build_steiner_tree_bactracking(next_node, best_e)
steiner_branch_f = SuitabilityGraph()
if best_f is not None and best_f != [next_node] and len(best_f) > 0:
steiner_branch_f = self.__build_steiner_tree_bactracking(next_node, best_f)
steiner_tree.append_graph(steiner_branch_e)
steiner_tree.append_graph(steiner_branch_f)
return steiner_tree
# node_weights[401] = generator.weights['VERY_SUITABLE'][0]
node_weights[405] = generator.weights['VERY_SUITABLE'][0]
gh = GridDigraphGenerator()
node_weighted = gh.generate(
m,
n,
edge_weighted=False,
node_weighted=True,
# node_weight_generator=generator,
node_weights=node_weights,
seed=seed)
terminals = [200, 760, 763, 766, 499]
suitability_graph = SuitabilityGraph()
suitability_graph.append_graph(node_weighted)
dr = DreyfusIMR(suitability_graph,
terminals,
contract_graph=False,
within_convex_hull=False)
start_time = time.clock()
steiner_tree = dr.steiner_tree(consider_terminals=False)
elapsed_time = time.clock() - start_time
cost, node_cost = steiner_tree.compute_total_weights(terminals)
ngh = NetworkXGraphHelper(suitability_graph)
ngh.draw_graph(nodes_1=terminals,
subgraphs_2=[steiner_tree],
m = n = 40
gh = GridDigraphGenerator()
graph = gh.generate(
m,
n,
edge_weighted=False,
node_weighted=True,
node_weight_generator=generator,
# node_weights=node_weights,
seed=seed)
terminals = [470, 388, 750, 1185, 1222, 739, 487, 850, 1299, 333]
poi = 899
suitability_graph = SuitabilityGraph()
suitability_graph.append_graph(graph)
suitability_graph.extend_suitable_regions(seed, generator)
suitability_graph.extend_suitable_regions(seed, generator)
regions = suitability_graph.get_suitable_regions(generator)
# suitability_graph.contract_suitable_regions(generator)
# paths = []
# for t in terminals:
# _, paths_t = dijkstra(suitability_graph, t, [poi], consider_node_weights=False)
# paths.append(paths_t[poi])
# _, paths_poi = suitability_graph.compute_shortest(poi, terminals)
terminals = [23, 17, 65]
# terminals = [703, 858, 668, 171, 628, 886, 240, 383, 268, 686]
# terminals = [3381, 2580, 2655, 3622, 2161, 5247, 5073, 871, 4946, 1017]
# terminals = [23, 45, 56, 289, 365]
# terminals = [331, 356, 297]
# poi = 294
# terminals = [197, 221]
# poi = 74
# terminals = [123, 230, 310, 588, 625, 700]
# poi = 464
suitability_graph = SuitabilityGraph()
suitability_graph.append_graph(graph)
# suitability_graph.extend_suitable_regions(seed, generator)
# suitability_graph.extend_suitable_regions(seed, generator)
# regions = suitability_graph.get_suitable_regions(generator)
contract_graph = False
within_convex_hull = False
consider_terminals = True
start_time = time.clock()
dr = DreyfusIMR(suitability_graph,
terminals,
contract_graph=contract_graph,
from grid_digraph_generator import GridDigraphGenerator
from suitability import SuitableNodeWeightGenerator, SuitabilityGraph
if __name__ == '__main__':
generator = SuitableNodeWeightGenerator()
seed = 0
m = n = 30
gh = GridDigraphGenerator()
node_weighted = gh.generate(m,
n,
edge_weighted=True,
node_weighted=True,
node_weight_generator=generator,
seed=seed)
suitability_graph = SuitabilityGraph()
suitability_graph.append_graph(node_weighted)
weights = {i: generator.weights["VERY_SUITABLE"][0] for i in range(m * n)}
suitability_graph.update_node_weights(weights)
if __name__ == '__main__':
generator = SuitableNodeWeightGenerator()
seed = 10
m = n = 50
gh = GridDigraphGenerator()
node_weighted = gh.generate(m,
n,
edge_weighted=True,
node_weighted=True,
node_weight_generator=generator,
seed=seed)
suitability_graph = SuitabilityGraph()
suitability_graph.append_graph(node_weighted)
# suitability_graph.extend_suitable_regions(seed, generator)
hotspots = suitability_graph.get_suitable_nodes(generator)
terminals = np.random.choice(a=m * n, size=30, replace=False)
while set(hotspots).intersection(terminals) != set():
terminals = np.random.choice(a=m * n, size=30, replace=False)
pois = terminals[:3]
terminals = terminals[3:]
regions = suitability_graph.get_suitable_regions(generator)
def __init__(self,
graph,
terminals,
poi,
max_level_attraction=2,
contract_graph=True,
contracted_graph=None,
within_convex_hull=False,
dist_paths_suitable_nodes=None):
if not graph.is_node_weighted():
raise (
ValueError,
"Gravitation algorithm only works with node-weighted graphs.")
# Store class variables for future references.
self.__original_graph = graph
self.__terminals = terminals
self.__poi = poi
self.__contract_graph = contract_graph
#
terminals_poi = list(terminals)
terminals_poi.append(poi)
generator = SuitableNodeWeightGenerator()
# Contracted graph...
if contract_graph:
if contracted_graph is not None:
self.__graph = contracted_graph.copy()
else:
self.__graph = SuitabilityGraph()
self.__graph.append_graph(graph)
self.__graph.contract_suitable_regions(
generator, excluded_nodes=terminals_poi)
else:
self.__graph = SuitabilityGraph()
self.__graph.append_graph(graph)
# #
# ngh = NetworkXGraphHelper(self.__original_graph)
# ngh.draw_graph(nodes_1=terminals,
# nodes_2=[poi],
# subgraphs_1=[r for _, (r, _, _) in self.__graph.contracted_regions.iteritems()],
# node_weight_generator=generator,
# node_size=25)
# #
# ngh = NetworkXGraphHelper(self.__graph)
# ngh.draw_graph(node_weight_generator=generator, node_size=25, node_labels=True)
# Copy distances and paths dictionary since it will be changed.
dist_paths = None
if dist_paths_suitable_nodes is not None:
dist_paths = dict(dist_paths_suitable_nodes)
for e in terminals_poi:
dist_paths[e] = dijkstra(self.__graph, e)
# Get the suitable nodes.
if within_convex_hull:
self.__suitable_nodes = self.__graph.get_suitable_nodes_within_convex_set(
terminals_poi, generator, dist_paths)
else:
self.__suitable_nodes = self.__graph.get_suitable_nodes(
generator, excluded_nodes=terminals_poi)
# print(self.__suitable_nodes)
#
self.__dist_paths_node_node = {}
if dist_paths is not None:
self.__dist_paths_node_node = {
n: dist_paths[n]
for n in self.__suitable_nodes
}
else:
self.__dist_paths_node_node = \
{n: dijkstra(self.__graph, n) for n in self.__suitable_nodes}
for e in terminals_poi:
if e not in self.__dist_paths_node_node:
self.__dist_paths_node_node[e] = dijkstra(self.__graph, e)
#
max_distances = [
max(self.__dist_paths_node_node[n][0].values())
for n in self.__suitable_nodes
if len(self.__dist_paths_node_node[n][0].values()) > 0
]
if len(max_distances) > 0:
self.__max_dist = max(max_distances)
else:
self.__max_dist = 0
# #
# max_level_attraction_poi = 0
# for t in terminals:
# max_level_attraction_poi = max(max_level_attraction_poi, len(self.__dist_paths_node_node[poi][1][t]))
# mass = self.__calculate_mass_suitable_node(poi)
# self.__attract_nodes_to(poi, mass, poi, max_level_attraction_poi, 0, [])
#
dist_to_poi = {}
for n in self.__suitable_nodes:
try:
dist_to_poi[n] = self.__dist_paths_node_node[n][0][poi]
except KeyError:
dist_to_poi[n] = sys.maxint
# dist_to_poi = {n: self.__dist_paths_node_node[n][0][poi] for n in self.__suitable_nodes}
# ord_suit_nodes = sorted(dist_to_poi.iteritems(), key=operator.itemgetter(1), reverse=True)
ord_suit_nodes = sorted(dist_to_poi.iteritems(),
key=operator.itemgetter(1))
for n, _ in ord_suit_nodes:
mass = self.__calculate_mass_suitable_node(n)
self.__attract_nodes_to(n, mass, n, max_level_attraction, 0, [])
class Gravitation:
def __init__(self,
graph,
terminals,
poi,
max_level_attraction=2,
contract_graph=True,
contracted_graph=None,
within_convex_hull=False,
dist_paths_suitable_nodes=None):
if not graph.is_node_weighted():
raise (
ValueError,
"Gravitation algorithm only works with node-weighted graphs.")
# Store class variables for future references.
self.__original_graph = graph
self.__terminals = terminals
self.__poi = poi
self.__contract_graph = contract_graph
#
terminals_poi = list(terminals)
terminals_poi.append(poi)
generator = SuitableNodeWeightGenerator()
# Contracted graph...
if contract_graph:
if contracted_graph is not None:
self.__graph = contracted_graph.copy()
else:
self.__graph = SuitabilityGraph()
self.__graph.append_graph(graph)
self.__graph.contract_suitable_regions(
generator, excluded_nodes=terminals_poi)
else:
self.__graph = SuitabilityGraph()
self.__graph.append_graph(graph)
# #
# ngh = NetworkXGraphHelper(self.__original_graph)
# ngh.draw_graph(nodes_1=terminals,
# nodes_2=[poi],
# subgraphs_1=[r for _, (r, _, _) in self.__graph.contracted_regions.iteritems()],
# node_weight_generator=generator,
# node_size=25)
# #
# ngh = NetworkXGraphHelper(self.__graph)
# ngh.draw_graph(node_weight_generator=generator, node_size=25, node_labels=True)
# Copy distances and paths dictionary since it will be changed.
dist_paths = None
if dist_paths_suitable_nodes is not None:
dist_paths = dict(dist_paths_suitable_nodes)
for e in terminals_poi:
dist_paths[e] = dijkstra(self.__graph, e)
# Get the suitable nodes.
if within_convex_hull:
self.__suitable_nodes = self.__graph.get_suitable_nodes_within_convex_set(
terminals_poi, generator, dist_paths)
else:
self.__suitable_nodes = self.__graph.get_suitable_nodes(
generator, excluded_nodes=terminals_poi)
# print(self.__suitable_nodes)
#
self.__dist_paths_node_node = {}
if dist_paths is not None:
self.__dist_paths_node_node = {
n: dist_paths[n]
for n in self.__suitable_nodes
}
else:
self.__dist_paths_node_node = \
{n: dijkstra(self.__graph, n) for n in self.__suitable_nodes}
for e in terminals_poi:
if e not in self.__dist_paths_node_node:
self.__dist_paths_node_node[e] = dijkstra(self.__graph, e)
#
max_distances = [
max(self.__dist_paths_node_node[n][0].values())
for n in self.__suitable_nodes
if len(self.__dist_paths_node_node[n][0].values()) > 0
]
if len(max_distances) > 0:
self.__max_dist = max(max_distances)
else:
self.__max_dist = 0
# #
# max_level_attraction_poi = 0
# for t in terminals:
# max_level_attraction_poi = max(max_level_attraction_poi, len(self.__dist_paths_node_node[poi][1][t]))
# mass = self.__calculate_mass_suitable_node(poi)
# self.__attract_nodes_to(poi, mass, poi, max_level_attraction_poi, 0, [])
#
dist_to_poi = {}
for n in self.__suitable_nodes:
try:
dist_to_poi[n] = self.__dist_paths_node_node[n][0][poi]
except KeyError:
dist_to_poi[n] = sys.maxint
# dist_to_poi = {n: self.__dist_paths_node_node[n][0][poi] for n in self.__suitable_nodes}
# ord_suit_nodes = sorted(dist_to_poi.iteritems(), key=operator.itemgetter(1), reverse=True)
ord_suit_nodes = sorted(dist_to_poi.iteritems(),
key=operator.itemgetter(1))
for n, _ in ord_suit_nodes:
mass = self.__calculate_mass_suitable_node(n)
self.__attract_nodes_to(n, mass, n, max_level_attraction, 0, [])
# #
# ngh = NetworkXGraphHelper(self.__graph)
# ngh.draw_graph(node_weight_generator=generator, node_size=25, node_labels=False)
'''
'''
def __calculate_mass_suitable_node(self, node):
hits = 0
for t in self.__terminals:
p = self.__dist_paths_node_node[self.__poi][1][t]
if node in p:
hits += 10
mass = 0
if node in self.__graph.contracted_regions:
region = self.__graph.contracted_regions[node][0]
for n in region:
mass += 1. / region[n][0] * 100
else:
mass = 1. / self.__graph[node][0] * 100
return mass + hits
'''
'''
def __calculate_adjusted_edge_cost(self, edge_cost, mass_attracting_node,
distance_from_attracting_node):
if self.__max_dist > 0:
scaled_dist = float(
distance_from_attracting_node) / self.__max_dist
else:
scaled_dist = 1
return edge_cost * scaled_dist / mass_attracting_node
'''
'''
def __attract_nodes_to(self, attracting_node, mass_attracting_node,
current_node, level_attraction, distance_so_far,
already_affected_edges):
if level_attraction < 1:
return
for neighbour in self.__graph[current_node][1]:
edge = tuple(sorted([current_node, neighbour]))
distance_from_attracting_node = self.__dist_paths_node_node[
attracting_node][0][neighbour]
edge_cost = self.__graph[current_node][1][neighbour]
if distance_so_far + edge_cost == distance_from_attracting_node and edge not in already_affected_edges:
adjusted_edge_cost = \
self.__calculate_adjusted_edge_cost(edge_cost, mass_attracting_node, distance_from_attracting_node)
self.__graph[current_node][1][neighbour] = adjusted_edge_cost
self.__graph[neighbour][1][current_node] = adjusted_edge_cost
already_affected_edges.append(edge)
self.__attract_nodes_to(attracting_node, mass_attracting_node,
neighbour, level_attraction - 1,
distance_so_far + edge_cost,
already_affected_edges)
def __find_closest_node_to_node_within_region(self, node, region_id):
region = self.__graph.contracted_regions[region_id][0]
min_dist = sys.maxint
closest_node = None
distances, paths = dijkstra(self.__original_graph, node, region.keys())
for n in region:
if distances[n] < min_dist:
closest_node = n
min_dist = distances[n]
return closest_node, paths[closest_node]
def __decontract_steiner_tree(self, steiner_tree):
regions = []
paths = []
trees = []
for r in steiner_tree:
if r in self.__graph.contracted_regions:
regions.append(r)
neighbors = steiner_tree[r][1].keys()
new_terminals = []
for n in neighbors:
closest_node_to_n, path = self.__find_closest_node_to_node_within_region(
n, r)
paths.append(path)
new_terminals.append(closest_node_to_n)
del steiner_tree[n][1][r]
if len(new_terminals) > 1:
region = self.__graph.contracted_regions[r][0]
g = Gravitation(region,
new_terminals[1:],
new_terminals[0],
contract_graph=False)
st = g.steiner_tree()
trees.append(st)
for r in regions:
del steiner_tree[r]
for p in paths:
steiner_tree.append_path(p, self.__original_graph)
for st in trees:
steiner_tree.append_graph(st)
# Fix the edge costs.
for v in steiner_tree:
for w in steiner_tree[v][1]:
steiner_tree[v][1][w] = self.__original_graph[v][1][w]
@staticmethod
def __merge_subtrees(subtrees):
result = SuitabilityGraph()
for subtree in subtrees:
result.append_graph(subtree)
return result
def __prune_steiner_tree(self, steiner_tree):
while True:
keys_to_prune = []
for n in steiner_tree:
if n not in self.__terminals and n != self.__poi:
neighbours = steiner_tree[n][1].keys()
if len(neighbours) == 1:
neighbour = neighbours[0]
del steiner_tree[neighbour][1][n]
keys_to_prune.append(n)
if len(keys_to_prune) == 0:
break
for n in keys_to_prune:
del steiner_tree[n]
def steiner_tree(self):
subtrees = []
_, paths = dijkstra(self.__graph, self.__poi, self.__terminals)
for t in self.__terminals:
path = paths[t]
subtree = SuitabilityGraph()
# j = 0
# for i in range(len(path_to_poi)):
# if path_to_poi[i] in self.__graph.contracted_regions:
# region_id = path_to_poi[i]
# subtree.append_from_path(path_to_poi[j:i], self.__original_graph)
# closest_node = self.__find_closest_node_to_poi_within_region(region_id)
# path_endpoint_1 = self.__dist_paths_node_within_region_node[closest_node][1][path_to_poi[i - 1]]
# subtree.append_from_path(path_endpoint_1, self.__original_graph)
# path_endpoint_2 = self.__dist_paths_node_within_region_node[closest_node][1][path_to_poi[i + 1]]
# subtree.append_from_path(path_endpoint_2, self.__original_graph)
# j = i + 1
# subtree.append_from_path(path_to_poi[j:], self.__original_graph)
subtree.append_path(path, self.__graph)
subtrees.append(subtree)
steiner_tree = self.__merge_subtrees(subtrees)
self.__prune_steiner_tree(steiner_tree)
if self.__contract_graph:
self.__decontract_steiner_tree(steiner_tree)
return steiner_tree
def __merge_subtrees(subtrees):
result = SuitabilityGraph()
for subtree in subtrees:
result.append_graph(subtree)
return result
seed = 1
m = n = 50
gh = GridDigraphGenerator()
node_weighted = gh.generate(m,
n,
edge_weighted=True,
node_weighted=True,
node_weight_generator=generator,
seed=seed)
# pois = [359, 834, 520, 378, 755, 616, 1, 435]
# pois = [359, 834]
# terminals = [123, 456, 463, 897, 506, 639, 343, 232, 564, 766, 138, 469, 800]
suitability_graph = SuitabilityGraph()
suitability_graph.append_graph(node_weighted)
suitability_graph.extend_suitable_regions(seed, generator)
suitability_graph.extend_suitable_regions(seed, generator)
suitability_graph.extend_suitable_regions(seed, generator)
suitability_graph.extend_suitable_regions(seed, generator)
hotspots = suitability_graph.get_suitable_nodes(generator)
terminals = np.random.choice(a=m * n, size=60, replace=False)
while set(suitability_graph.keys()).intersection(terminals) != set(
terminals) or set(hotspots).intersection(terminals) != set():
terminals = np.random.choice(a=m * n, size=60, replace=False)
pois = terminals[:15]
seed = 5
m = n = 30
# m = n = 10
gh = GridDigraphGenerator()
node_weighted = gh.generate(m,
n,
edge_weighted=True,
node_weighted=True,
node_weight_generator=generator,
seed=seed)
terminals = [288, 315, 231, 312, 111, 609, 645, 434, 654, 469, 186]
# terminals = [36, 78, 28, 11]
suitability_graph = SuitabilityGraph()
suitability_graph.append_graph(node_weighted)
suitability_graph.extend_suitable_regions(seed, generator)
suitability_graph.extend_suitable_regions(seed, generator)
regions = suitability_graph.get_suitable_regions(generator)
start_time = time.clock()
dr = DreyfusIMRV2(suitability_graph,
terminals,
contract_graph=True,
within_convex_hull=False)
steiner_tree = dr.steiner_tree(consider_terminals=False)
elapsed_time = time.clock() - start_time
# node_weights[1221] = generator.weights['VERY_SUITABLE'][0]
gh = GridDigraphGenerator()
node_weighted = gh.generate(
m,
n,
edge_weighted=True,
node_weighted=True,
node_weight_generator=generator,
# node_weights=node_weights,
seed=seed)
terminals = [742, 870, 776, 578]
suitability_graph = SuitabilityGraph()
suitability_graph.append_graph(node_weighted)
suitability_graph.extend_suitable_regions(seed, generator)
suitability_graph.extend_suitable_regions(seed, generator)
suitability_graph.extend_suitable_regions(seed, generator)
suitability_graph.extend_suitable_regions(seed, generator)
regions = suitability_graph.get_suitable_regions(generator)
suitable_nodes = suitability_graph.get_suitable_nodes(
generator, excluded_nodes=terminals)
dist, _ = dijkstra(suitability_graph, terminals[0], terminals[1:])
upper_bound = sum([dist[t] for t in terminals[1:]])
results = []
generator = SuitableNodeWeightGenerator()
# try:
for seed in seeds:
print("seed:", seed)
for msns in range(len(ms)):
print("nodes:", ms[msns] * ns[msns])
graph = GridDigraphGenerator().generate(
ms[msns],
ns[msns],
node_weighted=True,
node_weight_generator=generator,
seed=seed)
suitability_graph = SuitabilityGraph()
suitability_graph.append_graph(graph)
hotspots = suitability_graph.get_suitable_nodes(generator)
start_time = time.clock()
suitability_graph.compute_dist_paths(origins=hotspots,
destinations=hotspots,
compute_paths=False)
print "compute", time.clock() - start_time, "# hotspots:", len(
hotspots)
for num_seats in capacity:
# suitability_graph.extend_suitable_regions(seed, generator)
# hotspots = suitability_graph.get_suitable_nodes(generator)
# print i, "# hotspots:", len(hotspots)
for num_terminals in nums_terminals:
def Ai(self, k, r, X, i):
T = SuitabilityGraph()
while k > 0:
TBEST = SuitabilityGraph()
for kprime in range(1, k + 1):
for v in self.__nodes:
if i > 1:
Tprime = self.Ai(kprime, v, X, i - 1)
p = self.__paths[tuple(sorted([r, v]))]
Tprime.append_path(p, self.__graph)
else:
dists = {}
for t in self.__terminals:
dists[t] = self.__dist[tuple(sorted([v, t]))]
ord_term = sorted(dists.iteritems(),
key=operator.itemgetter(1))
Tprime = SuitabilityGraph()
for j in range(kprime):
p = self.__paths[tuple(sorted([v,
ord_term[j][0]]))]
Tprime.append_path(p, self.__graph)
if self.d(TBEST) > self.d(Tprime):
TBEST = Tprime
T.append_graph(TBEST)
k -= len(set(TBEST.keys()).intersection(X))
X = set(X).difference(TBEST.keys())
return T
class ClusterBased:
def __init__(self, graph, terminals):
# Check whether graph is node-weighted.
if not graph.is_node_weighted():
raise (
ValueError,
"Cluster-based algorithm only works with node-weighted graphs."
)
# Extract POI from the terminals list.
if len(terminals) > 0:
self.__poi = terminals[0]
else:
return
#
generator = SuitableNodeWeightGenerator()
# Set object variables.
self.__graph = SuitabilityGraph()
self.__graph.append_graph(graph)
self.__terminals = terminals
#
# self.__regions = self.__graph.get_suitable_regions(generator, excluded_nodes=terminals,
# get_border_internal_nodes=True, get_centroid_medoid=True,
# get_characteristic_nodes=True)
self.__regions = self.__graph.get_suitable_regions(
generator,
excluded_nodes=terminals,
get_border_internal_nodes=True,
get_centroid_medoid=True)
#
self.__nodes = []
for id_r in self.__regions:
#
# characteristic_nodes = self.__regions[id_r][6]
# self.__nodes.extend(characteristic_nodes)
border_nodes = self.__regions[id_r][1]
self.__nodes.extend(border_nodes)
#
# if len(characteristic_nodes) == 0:
# medoid = self.__regions[id_r][4]
# # print(medoid)
# self.__nodes.append(medoid)
for t in terminals:
self.__nodes.append(t)
#
self.__dist, self.__paths = self.__graph.get_dist_paths(
origins=self.__nodes, destinations=self.__nodes)
'''
'''
def steiner_tree(self, consider_terminals=False):
dr = DreyfusIMR(graph=self.__graph,
terminals=self.__terminals,
contract_graph=False,
nodes=self.__nodes,
dist_paths=(self.__dist, self.__paths))
if consider_terminals:
steiner_tree = dr.steiner_tree(consider_terminals=True)
else:
steiner_tree = dr.steiner_tree(consider_terminals=False)
# self.__refine_steiner_tree(steiner_tree)
return steiner_tree
'''
'''
def __refine_steiner_tree(self, steiner_tree):
clusters = []
for id_r in self.__regions:
region = self.__regions[id_r][0]
for n in steiner_tree:
if n in region:
clusters.append(id_r)
break
print clusters
def generate_graph(results, generator, cost_type="distance", capacitated=False):
graph = SuitabilityGraph(capacitated=capacitated)
#
prev_way_id = None
prev_node_id = None
hotspots = set()
pois = set()
for r in results:
way_id = r[0]
node_id = r[1]
type_ = r[3]
stype = r[4]
poi_name = r[5]
lat = float(r[6])
lon = float(r[7])
sa1_code = r[8]
sa2_code = r[9]
hw_type = r[10]
if node_id not in graph:
if type_ == "hotspot":
graph[node_id] = (generator.weights["VERY_SUITABLE"][0], {}, {'lat': lat, 'lon': lon, 'sa1': sa1_code,
'sa2': sa2_code, 'subtype': stype})
hotspots.add(node_id)
else:
if type_ == "poi":
pois.add(node_id)
graph[node_id] = (generator.weights["WARNING"][0], {}, {'lat': lat, 'lon': lon, 'sa1': sa1_code,
'sa2': sa2_code, 'subtype': stype,
'name': poi_name})
if prev_way_id == way_id:
prev_lat = graph[prev_node_id][2]['lat']
prev_lon = graph[prev_node_id][2]['lon']
# Cost estimation
cost = 0
distance = haversine(lat, lon, prev_lat, prev_lon)
if cost_type == "distance":
cost = distance
elif cost_type == "travel_time":
cost = osm_avg(distance, hw_type)
#
graph[node_id][1][prev_node_id] = cost
graph[prev_node_id][1][node_id] = cost
prev_way_id = way_id
prev_node_id = node_id
#
# pdb.set_trace()
isolated = []
# Both dictionaries will INCLUDE HOT SPOTS AND POIs.
nodes_by_sa1_code = {}
nodes_by_sa2_code = {}
#
for node_id, info in graph.iteritems():
if len(info[1]) == 0 or (len(info[1]) == 1 and info[1].keys()[0] == node_id):
isolated.append(node_id)
else:
sa1_code = info[2]['sa1']
sa2_code = info[2]['sa2']
if sa1_code in nodes_by_sa1_code:
nodes_by_sa1_code[sa1_code].append(node_id)
else:
nodes_by_sa1_code[sa1_code] = [node_id]
if sa2_code in nodes_by_sa2_code:
nodes_by_sa2_code[sa2_code].append(node_id)
else:
nodes_by_sa2_code[sa2_code] = [node_id]
for node_id in isolated:
del graph[node_id]
if node_id in hotspots:
hotspots.remove(node_id)
if node_id in pois:
pois.remove(node_id)
#
print "h:", len(hotspots), "p:", len(pois)
return graph, list(hotspots), list(pois), nodes_by_sa1_code, nodes_by_sa2_code
class LazySteinerTree:
def __init__(self,
graph,
terminals,
hot_spots=None,
generator=None,
distances=None):
# Check whether graph is node-weighted.
if not graph.is_node_weighted():
raise (ValueError,
"Lazy Steiner Tree only works with node-weighted graphs.")
# Extract POI from the terminals list.
if len(terminals) > 0:
self.__poi = terminals[0]
else:
return
# Set object variables.
self.__graph = SuitabilityGraph()
self.__graph.append_graph(graph)
self.__terminals = terminals
self.__hot_spots = None
self.__nodes = None
self.__s_d = {}
self.__paths = {}
self.__refs = {}
# Set hot spots.
if hot_spots is None:
if generator is None:
generator = SuitableNodeWeightGenerator()
self.__hot_spots = self.__graph.get_suitable_nodes(
generator, excluded_nodes=terminals)
else:
self.__hot_spots = list(hot_spots)
# Set nodes = hot spots + terminals.
self.__nodes = list(self.__hot_spots)
for t in terminals:
self.__nodes.append(t)
# Set distances.
if distances is None:
len_hot_spots = len(self.__hot_spots)
self.__distances = {}
for t in self.__terminals:
dist, paths = dijkstra(self.__graph, t, self.__nodes)
for n in self.__nodes:
try:
self.__distances[tuple(sorted([t,
n]))] = (dist[n], 'N')
self.__paths[tuple(sorted([t, n]))] = paths[n]
except KeyError:
self.__distances[tuple(sorted([t, n]))] = (sys.maxint,
'N')
self.__paths[tuple(sorted([t, n]))] = []
for h1 in self.__hot_spots:
for i in range(self.__hot_spots.index(h1), len_hot_spots):
h2 = self.__hot_spots[i]
distance = 0
d_type = 'E'
if h1 == h2:
d_type = 'N'
else:
distance = haversine(self.__graph[h1][2]['lat'],
self.__graph[h1][2]['lon'],
self.__graph[h2][2]['lat'],
self.__graph[h2][2]['lon'])
self.__distances[tuple(sorted([h1,
h2]))] = (distance, d_type)
else:
self.__distances = dict(distances)
'''
'''
@staticmethod
def __create_subsets_e(set_):
sets_e = [tuple([set_[0]])]
l_set = len(set_)
for x in range(1, l_set - 1):
for y in comb(set_[1:], x):
t = [set_[0]]
t.extend(y)
sets_e.append(tuple(t))
return sets_e
def steiner_tree(self,
h_l_sd=20,
h_l_hot_spots=3,
consider_terminals=False):
#
set_c = tuple(sorted(self.__terminals[1:]))
t_tuples = [tuple([t]) for t in set_c]
#
for j in self.__nodes:
self.__s_d[j] = {}
for t in t_tuples:
dist, d_t = self.__distances[tuple(sorted([j, t[0]]))]
self.__s_d[j][t] = [[
dist, 0, dist, t[0], (None, None), d_t, d_t, d_t, d_t, 0
]]
#
for m in range(2, len(set_c)):
#
sets_d = [tuple(set_d) for set_d in comb(set_c, m)]
for set_d in sets_d:
# target_hot_spots = CandidatesList(h_l_hot_spots)
for i in self.__nodes:
self.__s_d[i][set_d] = CandidatesList(h_l_sd)
#
sets_e = self.__create_subsets_e(set_d)
#
for j in self.__nodes:
u = sys.maxint
sets_e_f = None
d_ts = None
for set_e in sets_e:
set_f = tuple(sorted(list(set(set_d) - set(set_e))))
if len(set_f) > 0:
s = self.__s_d[j][set_e][0][0] + self.__s_d[j][
set_f][0][0]
else:
s = self.__s_d[j][set_e][0][0]
if s < u:
u = s
sets_e_f = (set_e, set_f)
d_ts = (self.__s_d[j][set_e][0][5],
self.__s_d[j][set_f][0][5])
for i in self.__nodes:
try:
dist, d_t = self.__distances[tuple(sorted([i, j]))]
except KeyError:
dist = sys.maxint
d_t = 'N'
if consider_terminals:
cost = dist + u
# if cost < self.__steiner_distances[i][set_d][0][0]:
if cost < sys.maxint:
dd_t = 'E'
if d_t == 'N' and d_ts[0] == 'N' and d_ts[
1] == 'N':
dd_t = 'N'
self.__s_d[i][set_d].append([
cost, u, dist, j, sets_e_f, dd_t, d_t,
d_ts[0], d_ts[1], 0
])
# dist_to_poi = self.__distances[tuple(sorted([self.__poi, i]))][0]
# target_hot_spots.append([dist_to_poi + cost, i])
else:
cost = dist + u
# if cost < self.__steiner_distances[i][set_d][0][0] and j not in self.__terminals:
if j not in self.__terminals and cost < sys.maxint:
dd_t = 'E'
if d_t == 'N' and d_ts[0] == 'N' and d_ts[
1] == 'N':
dd_t = 'N'
self.__s_d[i][set_d].append([
cost, u, dist, j, sets_e_f, dd_t, d_t,
d_ts[0], d_ts[1], 0
])
# dist_to_poi = self.__distances[tuple(sorted([self.__poi, i]))][0]
# target_hot_spots.append([dist_to_poi + cost, i])
target_hot_spots = CandidatesList(h_l_hot_spots)
for i in self.__nodes:
cost = self.__s_d[i][set_d][0][0]
j = self.__s_d[i][set_d][0][3]
set_e, set_f = self.__s_d[i][set_d][0][4]
dist_to_poi = self.__distances[tuple(
sorted([self.__poi, i]))][0]
target_hot_spots.append([dist_to_poi + cost, i])
if j in self.__refs:
if set_e in self.__refs[j]:
self.__refs[j][set_e].add((i, set_d))
else:
self.__refs[j][set_e] = {(i, set_d)}
if set_f in self.__refs[j]:
self.__refs[j][set_f].add((i, set_d))
else:
self.__refs[j][set_f] = {(i, set_d)}
else:
self.__refs[j] = {
set_e: {(i, set_d)},
set_f: {(i, set_d)}
}
# which is the best node for steiner tree between terminals in D and POI
# print('-------------------------------------------------------')
# print(set_d)
# print('-------------------------------------------------------')
# self.__print_target_hot_spots(target_hot_spots, set_d)
# print('-------------------------------------------------------')
# pdb.set_trace()
self.__correct_i_s(target_hot_spots, set_d)
self.__print_target_hot_spots(target_hot_spots, set_d)
# print('-------------------------------------------------------')
#
sets_e = self.__create_subsets_e(set_c)
#
# cost = sys.maxint
target_hot_spots = CandidatesList(h_l_hot_spots)
self.__s_d[self.__poi][set_c] = CandidatesList(h_l_sd)
for j in self.__nodes:
u = sys.maxint
sets_e_f = None
d_ts = None
for set_e in sets_e:
set_f = tuple(sorted(list(set(set_c) - set(set_e))))
if len(set_f) > 0:
s = self.__s_d[j][set_e][0][0] + self.__s_d[j][set_f][0][0]
else:
s = self.__s_d[j][set_e][0][0]
if s < u:
u = s
sets_e_f = (set_e, set_f)
d_ts = (self.__s_d[j][set_e][0][5],
self.__s_d[j][set_f][0][5])
try:
dist, d_t = self.__distances[tuple(sorted([self.__poi, j]))]
except KeyError:
dist = sys.maxint
d_t = 'N'
if consider_terminals:
# if dist + u < cost:
dd_t = 'E'
if d_t == 'N' and d_ts[0] == 'N' and d_ts[1] == 'N':
dd_t = 'N'
cost = dist + u
if cost < sys.maxint:
self.__s_d[self.__poi][set_c].append([
cost, u, dist, j, sets_e_f, dd_t, d_t, d_ts[0],
d_ts[1], 0
])
target_hot_spots.append([cost, self.__poi])
else:
cost = dist + u
if j not in self.__terminals and cost < sys.maxint:
dd_t = 'E'
if d_t == 'N' and d_ts[0] == 'N' and d_ts[1] == 'N':
dd_t = 'N'
self.__s_d[self.__poi][set_c].append([
cost, u, dist, j, sets_e_f, dd_t, d_t, d_ts[0],
d_ts[1], 0
])
target_hot_spots.append([cost, self.__poi])
j = self.__s_d[self.__poi][set_c][0][3]
set_e, set_f = self.__s_d[self.__poi][set_c][0][4]
if j in self.__refs:
if set_e in self.__refs[j]:
self.__refs[j][set_e].add((self.__poi, set_c))
else:
self.__refs[j][set_e] = {(self.__poi, set_c)}
if set_f in self.__refs[j]:
self.__refs[j][set_f].add((self.__poi, set_c))
else:
self.__refs[j][set_f] = {(self.__poi, set_c)}
else:
self.__refs[j] = {
set_e: {(self.__poi, set_c)},
set_f: {(self.__poi, set_c)}
}
# print('-------------------------------------------------------')
# print(set_c)
# print('-------------------------------------------------------')
# self.__print_target_hotspots(target_hot_spots, set_c)
# print('-------------------------------------------------------')
# pdb.set_trace()
# self.__print_target_hot_spots(target_hot_spots, set_c)
# pdb.set_trace()
self.__correct_i_s(target_hot_spots, set_c)
# print('-------------------------------------------------------')
# #
# while True:
# delta_cost = self.__steinerify(self.__poi, set_c)
# if delta_cost == 0:
# break
#
# Reconstruct the Steiner by backtracking
steiner_tree = self.__build_steiner_tree(self.__poi, set_c)
#
return steiner_tree
# def __correct_i_s(self, target_hot_spots, subset):
# while len(target_hot_spots) > 0:
# # pdb.set_trace()
# i = target_hot_spots[0][1]
# dd_t = self.__s_d[i][subset][0][5]
# if dd_t == 'E':
# self.__correct_j_s(i, subset)
# target_hot_spots.pop(0)
def __correct_i_s(self, target_hot_spots, subset):
if len(target_hot_spots) > 0:
#
target_hot_spots.sort()
i = target_hot_spots[0][1]
dd_t = self.__s_d[i][subset][0][5]
#
while dd_t == 'E':
#
delta = self.__correct_j_s(i, subset)
target_hot_spots[0][0] += delta
#
target_hot_spots.sort()
i = target_hot_spots[0][1]
dd_t = self.__s_d[i][subset][0][5]
def __propagate(self, delta, j, subset):
try:
for i, subset_i in self.__refs[j][subset]:
# if i == 552618963:
# print(subset_i, self.__s_d[i][subset_i][0][0], delta, j, subset)
self.__s_d[i][subset_i][0][0] += delta
self.__s_d[i][subset_i][0][1] += delta
self.__propagate(delta, i, subset_i)
except KeyError:
pass
def __correct_j_s(self, i, subset):
#
# self.__s_d[i][subset].sort()
j = self.__s_d[i][subset][0][3]
old_j = j
old_set_e, old_set_f = self.__s_d[i][subset][0][4]
dd_t = self.__s_d[i][subset][0][5]
count = 0
#
while dd_t == 'E':
# ----------------------------------------------------------------------------------------------------------
delta_dist = 0
d_t_1 = self.__s_d[i][subset][0][6]
if d_t_1 == 'E':
#
i_j_tuple = tuple(sorted([i, j]))
if self.__distances[i_j_tuple][1] == 'N':
dist = self.__distances[i_j_tuple][0]
else:
try:
distances, _ = dijkstra(self.__graph, i, [j])
dist = distances[j]
self.__distances[i_j_tuple] = (dist, 'N')
except KeyError:
dist = sys.maxint
old_dist = self.__s_d[i][subset][0][2]
delta_dist = dist - old_dist
#
self.__s_d[i][subset][0][0] += delta_dist
self.__s_d[i][subset][0][2] = dist
self.__s_d[i][subset][0][6] = 'N'
# ----------------------------------------------------------------------------------------------------------
delta_u = 0
d_t_2 = self.__s_d[i][subset][0][7]
d_t_3 = self.__s_d[i][subset][0][8]
if d_t_2 == 'E' or d_t_3 == 'E':
# if i == 3073194802L and subset == (30287961, 313278858, 1011956802, 1655220587):
# pdb.set_trace()
u, set_e, set_f, delta_e, delta_f = self.__correct_e_f(
j, subset)
self.__s_d[i][subset][0][4] = (set_e, set_f)
delta_u = delta_e + delta_f
#
if u != self.__s_d[i][subset][0][1]:
delta_u = u - self.__s_d[i][subset][0][1]
self.__s_d[i][subset][0][0] += delta_u
self.__s_d[i][subset][0][1] += delta_u
self.__s_d[i][subset][0][7] = 'N'
self.__s_d[i][subset][0][8] = 'N'
# ----------------------------------------------------------------------------------------------------------
d_t_1 = self.__s_d[i][subset][0][6]
d_t_2 = self.__s_d[i][subset][0][7]
d_t_3 = self.__s_d[i][subset][0][8]
if d_t_1 == 'N' and d_t_2 == 'N' and d_t_3 == 'N':
self.__s_d[i][subset][0][5] = 'N'
#
delta = delta_dist + delta_u
self.__s_d[i][subset][0][9] = delta
if delta > 0 and count == 0:
self.__propagate(delta, i, subset)
if self.__s_d[i][subset][0][5] == 'N':
try:
del self.__refs[i][subset]
except KeyError:
pass
count += 1
#
self.__s_d[i][subset].sort()
j = self.__s_d[i][subset][0][3]
dd_t = self.__s_d[i][subset][0][5]
# Update refs
j = self.__s_d[i][subset][0][3]
set_e, set_f = self.__s_d[i][subset][0][4]
if j != old_j or set_e != old_set_e or set_f != old_set_f:
#
try:
self.__refs[old_j][old_set_e].remove((i, subset))
self.__refs[old_j][old_set_f].remove((i, subset))
except KeyError:
pass
#
if j in self.__refs:
if set_e in self.__refs[j]:
self.__refs[j][set_e].add((i, subset))
else:
self.__refs[j][set_e] = {(i, subset)}
if set_f in self.__refs[j]:
self.__refs[j][set_f].add((i, subset))
else:
self.__refs[j][set_f] = {(i, subset)}
else:
self.__refs[j] = {set_e: {(i, subset)}, set_f: {(i, subset)}}
return self.__s_d[i][subset][0][9]
def __create_subsets_e_f(self, j, set_):
subsets_e_f = []
sets_e = self.__create_subsets_e(set_)
for set_e in sets_e:
set_f = tuple(sorted(list(set(set_) - set(set_e))))
if len(set_f) > 0:
s = self.__s_d[j][set_e][0][0] + self.__s_d[j][set_f][0][0]
else:
s = self.__s_d[j][set_e][0][0]
subsets_e_f.append([s, set_e, set_f, 0, 0])
return subsets_e_f
def __correct_e_f(self, j, set_):
#
subsets_e_f = self.__create_subsets_e_f(j, set_)
#
subsets_e_f.sort()
set_e = subsets_e_f[0][1]
set_f = subsets_e_f[0][2]
dd_t_e = self.__s_d[j][set_e][0][5]
dd_t_f = self.__s_d[j][set_f][0][5]
while dd_t_e == 'E' or dd_t_f == 'E':
#
delta_e = self.__correct_j_s(j, set_e)
delta_f = self.__correct_j_s(j, set_f)
subsets_e_f[0][0] += delta_e + delta_f
subsets_e_f[0][3] = delta_e
subsets_e_f[0][4] = delta_f
#
subsets_e_f.sort()
set_e = subsets_e_f[0][1]
set_f = subsets_e_f[0][2]
dd_t_e = self.__s_d[j][set_e][0][5]
dd_t_f = self.__s_d[j][set_f][0][5]
return subsets_e_f[0]
def __print_target_hot_spots(self, target_hot_spots, subset):
for th in target_hot_spots:
dist_to_poi = th[0]
i = th[1]
j = self.__s_d[i][subset][0][3]
dd_type = self.__s_d[i][subset][0][5]
d_type_1 = self.__s_d[i][subset][0][6]
d_type_2 = self.__s_d[i][subset][0][7]
d_type_3 = self.__s_d[i][subset][0][8]
print(dist_to_poi, i, j, dd_type, d_type_1, d_type_2, d_type_3)
def __build_steiner_tree(self, node, subset):
steiner_tree = SuitabilityGraph()
next_node = self.__s_d[node][subset][0][3]
print(node, self.__s_d[node][subset])
# pdb.set_trace()
if next_node is not None:
try:
steiner_tree.append_path(
self.__paths[tuple(sorted([node, next_node]))],
self.__graph)
except KeyError:
_, paths = dijkstra(self.__graph, node, [next_node])
steiner_tree.append_path(paths[next_node], self.__graph)
(set_e, set_f) = self.__s_d[node][subset][0][4]
steiner_branch_e = SuitabilityGraph()
if set_e is not None and set_e != [next_node]:
steiner_branch_e = self.__build_steiner_tree(next_node, set_e)
steiner_branch_f = SuitabilityGraph()
if set_f is not None and set_f != [next_node] and len(set_f) > 0:
steiner_branch_f = self.__build_steiner_tree(next_node, set_f)
steiner_tree.append_graph(steiner_branch_e)
steiner_tree.append_graph(steiner_branch_f)
return steiner_tree
generator = SuitableNodeWeightGenerator()
results = []
try:
for seed in range(num_seeds):
for size in sizes:
graph = GridDigraphGenerator().generate(
size,
size,
node_weighted=True,
node_weight_generator=generator,
seed=seed)
suitability_graph = SuitabilityGraph()
suitability_graph.append_graph(graph)
total_num_suitable_nodes = len(
suitability_graph.get_suitable_nodes(generator))
for num_terminals in nums_terminals:
for sample in range(num_samples):
line = [
seed, size * size, total_num_suitable_nodes,
num_terminals, sample + 1
]
terminals = np.random.choice(a=size * size,
def __init__(self,
graph,
terminals,
hot_spots=None,
generator=None,
distances=None):
# Check whether graph is node-weighted.
if not graph.is_node_weighted():
raise (ValueError,
"Lazy Steiner Tree only works with node-weighted graphs.")
# Extract POI from the terminals list.
if len(terminals) > 0:
self.__poi = terminals[0]
else:
return
# Set object variables.
self.__graph = SuitabilityGraph()
self.__graph.append_graph(graph)
self.__terminals = terminals
self.__hot_spots = None
self.__nodes = None
self.__s_d = {}
self.__paths = {}
self.__refs = {}
# Set hot spots.
if hot_spots is None:
if generator is None:
generator = SuitableNodeWeightGenerator()
self.__hot_spots = self.__graph.get_suitable_nodes(
generator, excluded_nodes=terminals)
else:
self.__hot_spots = list(hot_spots)
# Set nodes = hot spots + terminals.
self.__nodes = list(self.__hot_spots)
for t in terminals:
self.__nodes.append(t)
# Set distances.
if distances is None:
len_hot_spots = len(self.__hot_spots)
self.__distances = {}
for t in self.__terminals:
dist, paths = dijkstra(self.__graph, t, self.__nodes)
for n in self.__nodes:
try:
self.__distances[tuple(sorted([t,
n]))] = (dist[n], 'N')
self.__paths[tuple(sorted([t, n]))] = paths[n]
except KeyError:
self.__distances[tuple(sorted([t, n]))] = (sys.maxint,
'N')
self.__paths[tuple(sorted([t, n]))] = []
for h1 in self.__hot_spots:
for i in range(self.__hot_spots.index(h1), len_hot_spots):
h2 = self.__hot_spots[i]
distance = 0
d_type = 'E'
if h1 == h2:
d_type = 'N'
else:
distance = haversine(self.__graph[h1][2]['lat'],
self.__graph[h1][2]['lon'],
self.__graph[h2][2]['lat'],
self.__graph[h2][2]['lon'])
self.__distances[tuple(sorted([h1,
h2]))] = (distance, d_type)
else:
self.__distances = dict(distances)
