def naively_expand_query_nodes(graph_db, query_nodes, a = 1): api_calls = 0 budget = a + 1 # initialize graph graph = nx.Graph() if len(query_nodes) == 0: return graph # we add query nodes for node in query_nodes: graph.add_node(node) opened_nodes = set([]) frontier = set(query_nodes[:]) step = 0 while step < budget and frontier: for f in frontier: neighbors = graph_db.get_neighbors(f) if neighbors: graph.add_edges_from([(f,n) for n in neighbors]) api_calls = api_calls + len(frontier) opened_nodes |= frontier frontier = set(graph.nodes()) - opened_nodes step = step + 1 return (graph, api_calls)
def expand_query_nodes(graph_db, query_nodes, a = 1): api_calls = 0 budget = a + 1 # initialize graph graph = nx.Graph() if len(query_nodes) == 0: return graph # we add query nodes for node in query_nodes: graph.add_node(node) # graph.add_nodes_from(query_nodes) opened_nodes = set([]) frontier = _get_new_frontier(graph, opened_nodes, query_nodes, budget) while frontier: for f in frontier: neighbors = graph_db.get_neighbors(f) if neighbors: graph.add_edges_from([(f,n) for n in neighbors]) api_calls = api_calls + len(frontier) opened_nodes |= set(frontier) frontier = _get_new_frontier(graph, opened_nodes, query_nodes, budget) # print(api_calls) # if api_calls > 10000: # sys.exit() #print(frontier) return (graph, api_calls)
def smart_expand(graph_db, query_nodes, a = 1, approx_threshold = 1.00, samples = 1):
api_calls = 0
num_of_edges = 0
schedule = 64
schedule_multiplier = 2
# initialize graph
graph = nx.Graph()
if len(query_nodes) == 0:
return (graph, 0)
# we add query nodes
for node in query_nodes:
graph.add_node(node)
# maintain a list of nodes on the frontier
nodes_seen = set(query_nodes)
nodeComponent = {}
components = []
for q in query_nodes:
comp = Component( set([q]) )
components.append( comp )
nodeComponent[q] = comp
current_best = 1.0
upper_bound = 1.0 * len(query_nodes)
approx_ratio = current_best / upper_bound
# print(approx_ratio, current_best, upper_bound)
while would_you_expand(components) and len(components) > 1 and (approx_ratio < approx_threshold):
# select the next node to expand
next_components = []
for c in components:
if not c.is_active() or len(c.nodes_to_expand) == 0:
continue
lucky_node = heappop( c.nodes_to_expand )[1]
comp_of_lucky_node = nodeComponent.get(lucky_node, None)
assert comp_of_lucky_node != None
neighbors = graph_db.get_neighbors(lucky_node)
if neighbors == None:
neighbors = []
api_calls += 1
for nbor in neighbors:
if not nbor in nodes_seen:
# nodes_to_expand.add(nbor)
heappush( c.nodes_to_expand, (random.random(), nbor) )
nodes_seen.add(nbor)
nodeComponent[nbor] = comp_of_lucky_node
comp_of_lucky_node.add(nbor)
comp_of_lucky_node.add_edge( lucky_node, nbor )
else:
comp_of_nbor = nodeComponent.get(nbor,None)
assert comp_of_nbor != None
if comp_of_lucky_node != comp_of_nbor:
union = Component( comp_of_lucky_node, comp_of_nbor )
union.add_edge( lucky_node, nbor )
comp_of_lucky_node.set_inactive()
comp_of_nbor.set_inactive()
next_components.append(union)
for u in union.S:
nodeComponent[u] = union
comp_of_lucky_node = union
graph.add_edge(lucky_node, nbor)
num_of_edges += 1
components = [c for c in components + next_components if c.is_active()]
if api_calls >= schedule:
if schedule >= 8192:
schedule += 8192
else:
schedule *= schedule_multiplier
best_with = [1]
best_ever = [1]
# for each connected component, pick one node at the frontier
for comp in components:
partial_solution = comp.get_partial_solution( graph, query_nodes, a, samples)
if partial_solution != None:
best_with.append(partial_solution[0][0])
best_ever.append(partial_solution[1][0])
current_best = max(current_best, max(best_ever))
upper_bound = sum([sc for sc in best_with if sc > 0])
approx_ratio = 1.0 * current_best / upper_bound
sys.stderr.write(str(api_calls) + '\t' + str(len(components)) + '\t'
+ str(num_of_edges) + '\t'
+ str(approx_ratio) + '\n')
comp_sizes = [len(c.S) for c in components]
comp_sizes.sort(reverse = True)
sys.stderr.write(str(comp_sizes) + '\n')
return (graph, api_calls)