def stat_worker(inputlist, outqueue, print_queue):
for el in inputlist:
alg = el[0]
A = el[1]
B = el[2]
if alg == "clust_ratio":
a = nm.average(nx.clustering(A).values())
b = nm.average(nx.clustering(B).values())
v = a / b
if alg == "degree_corr":
v = gu.correlation(sorted(nx.degree(A)), sorted(nx.degree(B)))
if alg == "dist_dist":
v = gu.correlate_dist_dict(gu.node_distance_dist(A),
gu.node_distance_dist(B))
if alg == "deg_bet_corr":
v = gu.correlate_dist_dict(gu.get_degree_betweeness(A),
gu.get_degree_betweeness(B))
if alg == "kcore_corr":
v = gu.correlate_dist_dict(gu.get_kcoreness(A),
gu.get_kcoreness(B))
if alg == "comm_neigh_corr":
v = gu.correlate_dist_dict(gu.get_common_neigh_dist(A),
gu.get_common_neigh_dist(B))
if alg == "mod_ratio":
v = community.modularity(community.best_partition(A), A) /\
community.modularity(community.best_partition(B), B)
if alg == "avg_neigh_deg_corr":
v = gu.correlate_dist_dict(gu.get_avg_neighbour_degree(A),
gu.get_avg_neighbour_degree(B))
eg.info("Computed " + alg)
outqueue.put({alg: v})
def graph_worker_oneshot(inputlist, queue, print_queue):
for duty in inputlist:
name = duty[0]
G = duty[1]
algo = duty[2]
param = duty[3]
A = nx.to_numpy_matrix(G)
x, l = eg.eigen_centrality(A)
eg.info("Setup completed")
start_time = time.time()
if algo == "spectre":
m = float(param)
n = nm.shape(A)[0]
B = eg.sample_simm_matrix2(A, int(round(n * m)))
H = None
algo += str(m)
elif algo == "traj":
H = traj.random_graph(G)
elif algo == "sah":
dists = param.split(",")
H = sah.random_graph(G, dists[0], dists[1])
elif algo == "modularity":
m = float(param)
n = nm.shape(A)[0]
B = eg.modularity_clone_matrix(A, int(round(n * m)))
H = gu.simm_matrix_2_graph(B)
while nx.is_isomorphic(G, H):
B = eg.modularity_clone_matrix(A, int(round(n * m)))
H = gu.simm_matrix_2_graph(B)
algo += str(m)
elif algo == "25k":
H = twok.gen_25k_graph(G)
elif algo == "sbm":
H = sbm.generate(G)
eg.info("Graph Generated")
gc.collect()
stat = get_statistics1(G, H, time.time() - start_time, fraction=0.1)
s = algo + "," + name + "," + str(len(G.nodes()))
for el in stat:
s += "," + str(el)
print_queue.put(s)
print_queue.put("\n")
gc.collect()
def graph_worker_oneshot(inputlist, queue, print_queue):
for duty in inputlist:
name = duty[0]
G = duty[1]
algo = duty[2]
param = duty[3]
names = G.nodes()
names.sort(key=int)
A = nx.to_numpy_matrix(G, nodelist=names)
eg.info("Setup completed")
start_time = time.time()
if algo == "modularity":
m = float(param)
n = nm.shape(A)[0]
B = eg.modularity_clone_matrix(A, int(round(n*m)))
H = gu.simm_matrix_2_graph(B, names)
while nx.is_isomorphic(G, H):
B = eg.modularity_clone_matrix(A, int(round(n*m)))
H = gu.simm_matrix_2_graph(B, names)
algo += str(m)
elif algo == "traj":
H = traj.random_graph(G)
trnames = H.nodes()
mapping = {trnames[i]: names[i] for i in range(len(names))}
H = nx.relabel_nodes(H, mapping)
elif algo == "25k":
H = twok.gen_25k_graph(G, names)
elif algo == "sbm":
H = sbm.generate(G, names)
eg.info("Graph Generated " + name)
gc.collect()
stat = get_statistics(G, H, time.time()-start_time, fraction=0.1)
s = name + "," + str(len(G.nodes())) + "," + algo
for el in stat:
s += "," + str(el)
print_queue.put(s)
print_queue.put("\n")
gc.collect()
def write_statistics(A, B, label, net, x, l, output=True):
eg.info("Computing new centrality..")
G = nx.from_numpy_matrix(A)
H = nx.from_numpy_matrix(B)
nx.write_weighted_edgelist(
H,
net + "_" + label + "_" + str(random.randint(0, 99999999)) + ".edges")
y, m = eg.eigen_centrality(B, maxiter=100000)
eg.info("Printing out results..")
eigen_err = eg.vec_dist(x, y)
clust_err = nm.average(nx.clustering(G).values()) /\
nm.average(nx.clustering(H).values())
lambda_err = abs(l / m)
degree_corr = gu.correlation(sorted(nx.degree(G)), sorted(nx.degree(H)))
if nx.is_connected(H):
conn = 1
else:
conn = 0
out = (
str(label) + "," + str(net.split(".")[0].split("/")[-1]) + "," +
str(eigen_err) + "," + str(degree_corr) + "," + str(clust_err) + "," +
str(lambda_err) + "," + str(-1) + "," + str(-1) + "," +
# str(gu.correlate_dist_dict(gu.node_distance_dist(A),
# gu.node_distance_dist(B))) + "," +
# str(gu.correlate_dist_dict(gu.get_degree_betweeness(A),
# gu.get_degree_betweeness(B))) + "," +
str(gu.correlate_dist_dict(gu.get_kcoreness(A), gu.get_kcoreness(B))) +
"," + str(
gu.correlate_dist_dict(gu.get_common_neigh_dist(A),
gu.get_common_neigh_dist(B))) + "," +
str(-1) + "," +
# str(community.modularity(
# community.best_partition(G), G) /
# community.modularity(
# community.best_partition(H), H)) + "," +
str(
gu.correlate_dist_dict(gu.get_avg_neighbour_degree(A),
gu.get_avg_neighbour_degree(B))) + "," +
str(conn))
if output:
print(out, file=sys.stderr)
return out
def csv_test_unparallel():
eg.__eigen_info = True
print("Strategy," + "Graph," + str("EigErr") + "," + str("DegCorr") + "," +
str("ClustRatio") + "," + str("EigVErr") + "," +
str("NodeDistCorr") + "," + str("DegBetCorr") + "," +
str("KCoreCorr") + "," + str("CommNeighDist") + "," +
str("PartRatio") + "," + str("AvgNeighDegCorr") + "," +
str("Connected"),
file=sys.stderr)
for filo in os.listdir("/home/baldo/tmp/graph_generator/PL200/"):
with open("/home/baldo/tmp/graph_generator/PL200/" + filo, "r") as net:
eg.info(filo)
eg.info("Loading graph..")
G = nx.read_weighted_edgelist(net) # , delimiter=":")
# G = read_graphml(net)
A = nx.to_numpy_matrix(G)
n = nm.shape(A)[0]
joint_degrees = nx.algorithms.mixing.degree_mixing_dict(G)
eg.info("Computing centrality..")
x, l = eg.eigen_centrality(A)
for i in range(10):
eg.info("Run: " + str(i))
eg.info("Building JDM graph..")
H = joint_degree_graph(joint_degrees)
B = nx.to_numpy_matrix(H)
write_statistics(A, B, "2k", filo, x, l)
eg.info("Building degree sequence graph..")
H = nx.random_degree_sequence_graph((nx.degree(G).values()))
B = nx.to_numpy_matrix(H)
write_statistics(A, B, "1k", filo, x, l)
precision = 0.01
eg.info("Building eigen " + str(precision) + " graph..")
B = eg.build_matrix(x, l, precision)
write_statistics(A, B, "eig" + str(precision), filo, x, l)
precision = 0.001
eg.info("Building eigen " + str(precision) + " graph..")
B = eg.build_matrix(x, l, precision)
write_statistics(A, B, "eig" + str(precision), filo, x, l)
precision = 0.0001
eg.info("Building eigen " + str(precision) + " graph..")
B = eg.build_matrix(x, l, precision)
write_statistics(A, B, "eig" + str(precision), filo, x, l)
m = 0.25
eg.info("Building spectral " + str(m) + " graph..")
B = eg.sample_simm_matrix(A, int(round(n * m)))
write_statistics(A, B, "spectre" + str(m), filo, x, l)
m = 0.5
eg.info("Building spectral " + str(m) + " graph..")
B = eg.sample_simm_matrix(A, int(round(n * m)))
write_statistics(A, B, "spectre" + str(m), filo, x, l)
m = 0.75
eg.info("Building spectral " + str(m) + " graph..")
B = eg.sample_simm_matrix(A, int(round(n * m)))
write_statistics(A, B, "spectre" + str(m), filo, x, l)
m = 0.9
eg.info("Building spectral " + str(m) + " graph..")
B = eg.sample_simm_matrix(A, int(round(n * m)))
write_statistics(A, B, "spectre" + str(m), filo, x, l)
m = 0.95
eg.info("Building spectral " + str(m) + " graph..")
B = eg.sample_simm_matrix(A, int(round(n * m)))
write_statistics(A, B, "spectre" + str(m), filo, x, l)
eg.info("Building D2.5 graph..")
test25 = Estimation()
gen25 = Generation()
test25.load_graph("", graph=G)
test25.calcfull_CCK()
test25.calcfull_JDD()
gen25.set_JDD(test25.get_JDD('full'))
gen25.set_KTRI(test25.get_KTRI('full'))
gen25.construct_triangles_2K()
gen25.mcmc_improved_2_5_K(error_threshold=0.05)
H = gen25.G
B = nx.to_numpy_matrix(H)
write_statistics(A, B, "25k", filo, x, l)
def graph_worker_oneshot(inputlist, queue, print_queue):
for duty in inputlist:
name = duty[0]
G = duty[1]
algo = duty[2]
param = duty[3]
A = nx.to_numpy_matrix(G)
# x, l = eg.eigen_centrality(A)
eg.info("Setup completed")
start_time = time.time()
if algo == "1k":
H = nx.random_degree_sequence_graph((nx.degree(G).values()))
# B = nx.to_numpy_matrix(H)
elif algo == "2k":
joint_degrees = nx.algorithms.mixing.degree_mixing_dict(G)
H = joint_degree_graph(joint_degrees)
# B = nx.to_numpy_matrix(H)
elif algo == "eig":
precision = float(param)
# B = eg.build_matrix(x, l, precision)
B = eg.generate_matrix(x, l * x, precision, gu.get_degrees(A))
H = None
algo += str(precision)
elif algo == "modeig":
precision = float(param)
B = eg.synthetic_modularity_matrix(A, precision)
H = None
algo += str(precision)
elif algo == "spectre":
m = float(param)
n = nm.shape(A)[0]
B = eg.sample_simm_matrix2(A, int(round(n * m)))
H = gu.simm_matrix_2_graph(B)
while nx.is_isomorphic(G, H):
B = eg.sample_simm_matrix2(A, int(round(n * m)))
H = gu.simm_matrix_2_graph(B)
algo += str(m)
elif algo == "laplacian":
m = float(param)
n = nm.shape(A)[0]
B = eg.laplacian_clone_matrix(A, int(round(n * m)))
H = gu.simm_matrix_2_graph(B)
while nx.is_isomorphic(G, H):
B = eg.sample_simm_matrix2(A, int(round(n * m)))
H = gu.simm_matrix_2_graph(B)
algo += str(m)
elif algo == "modspec":
m = float(param)
n = nm.shape(A)[0]
B = eg.modspec_clone_matrix(A, int(round(n * m)))
H = None
algo += str(m)
elif algo == "franky":
m = float(param)
n = nm.shape(A)[0]
B = eg.franky_clone_matrix(A, int(round(n * m)))
H = None
algo += str(m)
elif algo == "modularity":
m = float(param)
n = nm.shape(A)[0]
B = eg.modularity_clone_matrix(A, int(round(n * m)))
H = gu.simm_matrix_2_graph(B)
while nx.is_isomorphic(G, H):
B = eg.modularity_clone_matrix(A, int(round(n * m)))
H = gu.simm_matrix_2_graph(B)
algo += str(m)
elif algo == "25k":
test25 = Estimation()
gen25 = Generation()
test25.load_graph("", graph=G)
test25.calcfull_CCK()
test25.calcfull_JDD()
gen25.set_JDD(test25.get_JDD('full'))
gen25.set_KTRI(test25.get_KTRI('full'))
gen25.construct_triangles_2K()
gen25.mcmc_improved_2_5_K(error_threshold=0.05)
H = gen25.G
# B = nx.to_numpy_matrix(H)
eg.info("Graph Generated")
stat = get_statistics1(G, H, time.time() - start_time)
s = algo + "," + name + "," + str(len(G.nodes()))
for el in stat:
s += "," + str(el)
print_queue.put(s)
print_queue.put("\n")
gc.collect()
def get_statistics(G, H, A, B, x, l, duration):
eg.info("Computing statistics...")
if not H:
eg.info("Building networkx graph...")
H = gu.simm_matrix_2_graph(B)
eg.info("Computing centrality...")
y, m = (-1, -1) #eg.eigen_centrality(B, maxiter=100000)
# nx.write_weighted_edgelist(H, "pgp_spectre0.9_" +
# str(random.randint(0, 99999999)) + ".edges")
eg.info("Computing centrality distance...")
eigen_err = -1 #eg.vec_dist(x, y)
eg.info("Computing clustering ratio...")
clust_err = gu.average_clustering(A) / gu.average_clustering(B)
# clust_err = nm.average(nx.clustering(G).values()) /\
# nm.average(nx.clustering(H).values())
eg.info("Computing lambda ratio...")
lambda_err = -1 #abs(l / m)
eg.info("Computing degree correlation...")
degree_corr = gu.correlation(gu.get_degrees(A), gu.get_degrees(B))
eg.info("Check connectivity...")
if nx.is_connected(H):
conn = 1
else:
conn = 0
eg.info("Distance distribution correlation...")
distance_dist_corr = -1 #gu.correlate_dist_dict(gu.node_distance_dist(A),
# gu.node_distance_dist(B))
eg.info("Degree betweenness correlation...")
degree_bet_corr = -1 #gu.correlate_dist_dict(gu.get_degree_betweeness(A),
# gu.get_degree_betweeness(B))
eg.info("K-coreness correlation...")
kcore_corr = -1 #gu.correlate_dist_dict(gu.get_kcoreness(A),
# gu.get_kcoreness(B))
eg.info("Common neighbourhood correlation...")
common_neigh_corr = -1 #gu.correlate_dist_dict(gu.get_common_neigh_dist(A),
# gu.get_common_neigh_dist(B))
eg.info("Modularity ratio...")
Gm = gu.norm_modularity(G)
Hm = gu.norm_modularity(H)
modularity_ratio = Gm[0] / Hm[0]
partition_ratio = Gm[1] / float(Hm[1])
eg.info("Avg neighbourhood degree correlation...")
avg_neigh_deg_corr = -1 #gu.correlate_dist_dict(gu.get_avg_neighbour_degree(A),
# gu.get_avg_neighbour_degree(B))
eg.info("Done with stats")
return (eigen_err, degree_corr, clust_err, lambda_err, distance_dist_corr,
degree_bet_corr, kcore_corr, common_neigh_corr, modularity_ratio,
partition_ratio, avg_neigh_deg_corr, conn, duration)
def get_statistics1(G, H, duration):
sample_size = 100
gsample = random.sample(G.nodes(), sample_size)
hsample = random.sample(H.nodes(), sample_size)
eg.info("Computing statistics...")
eg.info("Computing centrality...")
x = nx.eigenvector_centrality_numpy(G)
y = nx.eigenvector_centrality_numpy(H)
eg.info("Computing centrality distance...")
eigen_err = gu.sorted_correlation(x.values(), y.values())
eg.info("Computing clustering ratio...")
clust1 = nx.clustering(G, gsample)
clust2 = nx.clustering(H, hsample)
clust_err = sum(clust1.values()) / sum(clust2.values())
lambda_err = -1
eg.info("Computing degree correlation...")
degdist1 = {}
degdist2 = {}
degree1 = nx.degree(G, nbunch=gsample)
degree2 = nx.degree(H, nbunch=hsample)
for d in degree1.values():
degdist1[d] = degdist1.get(d, 0) + 1
for d in degree2.values():
degdist2[d] = degdist2.get(d, 0) + 1
degree_corr = gu.correlate_dist_dict(degdist1, degdist2)
eg.info("Check connectivity...")
if nx.is_connected(H):
conn = 1
else:
conn = 0
eg.info("Distance distribution correlation...")
dist1 = {}
dist2 = {}
for i in range(sample_size):
for j in range(i + 1, sample_size):
try:
d = nx.algorithms.bidirectional_dijkstra(
G, gsample[i], gsample[j])[0]
dist1[d] = dist1.get(d, 0) + 1
except:
pass
try:
d = nx.algorithms.bidirectional_dijkstra(
H, hsample[i], hsample[j])[0]
dist2[d] = dist2.get(d, 0) + 1
except:
pass
distance_dist_corr = gu.correlate_dist_dict(dist1, dist2)
eg.info("Betweenness correlation...")
b1 = nx.betweenness_centrality(G, sample_size).values()
b2 = nx.betweenness_centrality(H, sample_size).values()
bet_corr = gu.sorted_correlation(b1, b2)
# gu.get_degree_betweeness(B))
eg.info("K-coreness correlation...")
kcore_corr = -1 #gu.correlate_dist_dict(gu.get_kcoreness(A),
# gu.get_kcoreness(B))
eg.info("Common neighbourhood correlation...")
comm1 = {}
comm2 = {}
for i in range(sample_size):
for j in range(i + 1, sample_size):
d = len(list(nx.common_neighbors(G, gsample[i], gsample[j])))
comm1[d] = comm1.get(d, 0) + 1
d = len(list(nx.common_neighbors(H, hsample[i], hsample[j])))
comm2[d] = comm2.get(d, 0) + 1
common_neigh_corr = gu.correlate_dist_dict(comm1, comm2)
eg.info("Modularity ratio...")
Gm = gu.norm_modularity(G)
Hm = gu.norm_modularity(H)
modularity_ratio = Gm[0] / Hm[0]
partition_ratio = Gm[1] / float(Hm[1])
eg.info("Avg neighbourhood degree correlation...")
avg1 = nx.average_neighbor_degree(G, nodes=gsample)
avg2 = nx.average_neighbor_degree(H, nodes=hsample)
avg_neigh_deg_corr = gu.sorted_correlation(avg1.values(), avg2.values())
eg.info("Done with stats")
return (eigen_err, degree_corr, clust_err, lambda_err, distance_dist_corr,
bet_corr, kcore_corr, common_neigh_corr, modularity_ratio,
partition_ratio, avg_neigh_deg_corr, conn, duration)
def get_statistics2(G, H, A, B, x, l):
eg.info("Computing statistics...")
if not H:
eg.info("Building networkx graph...")
H = gu.simm_matrix_2_graph(B)
gu.connect_components(H)
eg.info("Computing centrality...")
y, m = eg.eigen_centrality(B, maxiter=100000)
eg.info("Computing lambda ratio...")
lambda_err = abs(l / m)
eg.info("Computing centrality distance...")
eigen_err = eg.vec_dist(x, y)
inputs = [("avg_neigh_deg_corr", A, B), ("mod_ratio", G, H),
("comm_neigh_corr", A, B), ("kcore_corr", A, B),
("deg_bet_corr", A, B), ("dist_dist", A, B),
("degree_corr", G, H), ("clust_ratio", G, H)]
mets = parallelism.launch_workers(inputs,
stat_worker,
inputs_per_worker=1,
parallelism=4)
res = {}
for el in mets:
res.update(el)
eg.info("Check connectivity...")
if nx.is_connected(H):
conn = 1
else:
conn = 0
eg.info("Done with stats")
return (eigen_err, res['degree_corr'], res['clust_ratio'], lambda_err,
res['dist_dist'], res['deg_bet_corr'], res['kcore_corr'],
res['comm_neigh_corr'], res['mod_ratio'],
res['avg_neigh_deg_corr'], conn)
