def rwrw_size_estimate(graph, n_samples=-1, walk_length=200, thinning=40):
# determine the number of samples
if n_samples == -1: n_samples = graph.size() * 2
samples = RWRW(graph, n_samples, length=walk_length, thinning=thinning)
node_degrees = [graph.degree(node) for node in samples]
sum_of_degrees = sum(node_degrees)
sum_of_inverse_degrees = sum(graphsize.inverse_seq(node_degrees))
collisions = graphsize.collision_count(samples)
return graphsize.estimate_size(sum_of_degrees, sum_of_inverse_degrees, collisions)
def wis_wr_size_estimate(graph, n_samples=-1):
# determine the number of samples
if n_samples == -1:
n_samples = graph.size() * 2
samples = graphsize.WIS_WR(graphsize.degree_weighted_nodes_for(graph), n_samples)
node_degrees = [graph.degree(node) for node in samples]
sum_of_degrees = sum(node_degrees)
sum_of_inverse_degrees = sum(graphsize.inverse_seq(node_degrees))
collisions = graphsize.collision_count(samples)
return graphsize.estimate_size(sum_of_degrees, sum_of_inverse_degrees, collisions)
def estimate_size_with_mhrw(graph, n_samples=-1, thinning=1, random_walk_length=20):
# determine the number of samples
if n_samples == -1: n_samples = graph.size() * 4
#sample the graph and process the results
node_samples = MHRW(graph, graph.nodes(), n_samples, length=random_walk_length, thinning=thinning)
degrees = [graph.degree(node) for node in node_samples]
sum_of_degrees = sum(degrees)
sum_of_inverse_degrees = sum(graphsize.inverse_seq(degrees))
collisions = graphsize.collision_count(node_samples)
print 'Sum of degrees: ', sum_of_degrees
print 'Sum of inverse degrees: ', sum_of_inverse_degrees
print 'Repeated samples: ', collisions
return graphsize.estimate_size(sum_of_degrees, sum_of_inverse_degrees, collisions)
def reweighted_sample(samples, graph):
degrees = [graph.degree(node) for node in samples]
sum_of_inverse_degrees = sum(graphsize.inverse_seq(degrees))
num_samples = len(samples)
binned_samples = graphsize.bin_samples(samples)
# print "original sample bins", binned_samples
# reuse the WIS_WR probability distribution sampling method
# create a tuple list of nodes and weights for each bin element
# i.e. [(1, 0.2), (3, 0.4), (2, 0.5) ...}
hh_weighted_nodes = [
(node,
graphsize.hh_node_weight(graph.degree(node), occurrences, sum_of_inverse_degrees)
)
for [node, occurrences] in binned_samples
]
new_samples = graphsize.WIS_WR(hh_weighted_nodes, num_samples)
# print "reweighted sample bins", graphsize.bin_samples(new_samples)
return new_samples
def test_inverse_seq(self):
self.assertEqual([1, 2, 4, 1], graphsize.inverse_seq([1, 0.5, 0.25, 1]))
