def aggregate_measurements(number_of_trials,
adjacency_matrix,
profile_individual_stat_folder,
profile_aggregate_stat_folder):
adjacency_matrix = adjacency_matrix.tocsr()
number_of_nodes = adjacency_matrix.shape[0]
# Calculate random walk transition probability matrix.
rw_transition, out_degree, in_degree = get_natural_random_walk_matrix(adjacency_matrix)
iterate_nodes = np.where(out_degree != 0)[0]
fpr_stats = pstats.Stats()
lpr_stats = pstats.Stats()
fcprd_stats = pstats.Stats()
for trial in range(number_of_trials):
for n in range(iterate_nodes.size):
node = iterate_nodes[n]
# Fast PR filename.
filename = profile_individual_stat_folder + "/fpr_vertex_" + str(node) + "_trial_" + str(trial) + ".stats"
fpr_stats.add(filename)
# os.remove(filename)
# Lazy PR filename.
filename = profile_individual_stat_folder + "/lpr_vertex_" + str(node) + "_trial_" + str(trial) + ".stats"
lpr_stats.add(filename)
# os.remove(filename)
# Fast CPRD filename.
filename = profile_individual_stat_folder + "/fcprd_vertex_" + str(node) + "_trial_" + str(trial) + ".stats"
fcprd_stats.add(filename)
# os.remove(filename)
fpr_stats.dump_stats(profile_aggregate_stat_folder + "/fpr_vertex.stats")
lpr_stats.dump_stats(profile_aggregate_stat_folder + "/lpr_vertex.stats")
fcprd_stats.dump_stats(profile_aggregate_stat_folder + "/fcprd_vertex.stats")
def similarity_slice_benchmark(number_of_trials,
adjacency_matrix,
rho_effective,
epsilon,
laziness_factor,
profile_individual_stat_folder):
"""
Compares the efficiency of approaches calculating similarity matrix slices.
"""
adjacency_matrix = adjacency_matrix.tocsr()
number_of_nodes = adjacency_matrix.shape[0]
# Calculate random walk transition probability matrix.
rw_transition, out_degree, in_degree = get_natural_random_walk_matrix(adjacency_matrix)
# Calculate base communities(ego excluded) and out-degrees.
adjacent_nodes = np.ndarray(number_of_nodes, dtype=np.ndarray)
base_transitions = np.ndarray(number_of_nodes, dtype=np.ndarray)
for n in np.arange(number_of_nodes):
rw_transition_row = rw_transition.getrow(n)
adjacent_nodes[n] = rw_transition_row.indices.astype(np.int64)
base_transitions[n] = rw_transition_row.data
# Calculate restart probability in the case of lazy PageRank.
rho = (rho_effective * (1 - laziness_factor))/(1 - laziness_factor * rho_effective)
s = np.zeros(number_of_nodes, dtype=np.float64)
r = np.zeros(number_of_nodes, dtype=np.float64)
iterate_nodes = np.where(out_degree != 0)[0]
for trial in range(number_of_trials):
gc.collect()
print("Trial #", trial+1)
for n, node in enumerate(iterate_nodes):
if n % 500 == 0:
gc.collect()
epsilon_effective = calculate_epsilon_effective(None,
epsilon,
out_degree[node],
out_degree[adjacent_nodes[node]],
None)
# Perform fast PageRank.
s[:] = 0.0
r[:] = 0.0
filename = profile_individual_stat_folder + "/fpr_vertex_" + str(node) + "_trial_" + str(trial) + ".stats"
profile.runctx("""similarity.fast_approximate_personalized_pagerank(s,
r,
base_transitions,
adjacent_nodes,
out_degree,
in_degree,
node,
rho_effective,
epsilon)""",
globals(),
{"s": s,
"r": r,
"base_transitions": base_transitions,
"adjacent_nodes": adjacent_nodes,
"out_degree": out_degree,
"in_degree": in_degree,
"node": node,
"rho_effective": rho_effective,
"epsilon": epsilon_effective},
filename=filename)
# Perform lazy PageRank
s[:] = 0.0
r[:] = 0.0
filename = profile_individual_stat_folder + "/lpr_vertex_" + str(node) + "_trial_" + str(trial) + ".stats"
profile.runctx("""similarity.lazy_approximate_personalized_pagerank(s,
r,
base_transitions,
adjacent_nodes,
out_degree,
in_degree,
node,
rho,
epsilon)""",
globals(),
{"s": s,
"r": r,
"base_transitions": base_transitions,
"adjacent_nodes": adjacent_nodes,
"out_degree": out_degree,
"in_degree": in_degree,
"node": node,
"rho": rho,
"epsilon": epsilon_effective},
filename=filename)
# Perform Regularized Commute-Time
s[:] = 0.0
r[:] = 0.0
filename = profile_individual_stat_folder + "/fcprd_vertex_" + str(node) + "_trial_" + str(trial) + ".stats"
profile.runctx("""similarity.fast_approximate_regularized_commute(s,
r,
base_transitions,
adjacent_nodes,
out_degree,
in_degree,
node,
rho_effective,
epsilon)""",
globals(),
{"s": s,
"r": r,
"base_transitions": base_transitions,
"adjacent_nodes": adjacent_nodes,
"out_degree": out_degree,
"in_degree": in_degree,
"node": node,
"rho_effective": rho_effective,
"epsilon": epsilon_effective},
filename=filename)
