def calculate_auc_Node2Vec(train_set, poss_set, neg_set, d=10, directed=False):
fout = open('node2vec.in', 'w')
for edge in train_set:
if edge not in poss_set:
u, w = map(int, edge.split())
fout.write(str(u) + ' ' + str(w) + '\n')
fout.close()
if not directed:
print os.system(
'python ../node2vec/src/main.py --input node2vec.in --output node2vec.out '
+ '--dimensions %d --walk-length 80 --p 1 --iter 1' % d)
else:
print os.system(
'python ../node2vec/src/main.py --input node2vec.in --output node2vec.out '
+ '--dimensions %d --walk-length 80 --p 1 --iter 1 --directed' % d)
w2v = {}
with open('node2vec.out') as fin:
print(fin.readline())
for line in fin:
line = list(map(float, line.strip().split()))
w2v[int(line[0])] = np.array(line[1:])
features = tools.Node2VecFeatures(w2v)
X, Y = tools.make_dataset(poss_set, neg_set, [features.score])
return roc_auc_score(Y, X)
def calculate_auc(train_set, nodes, poss_set, neg_set, auc, gap, verbose,
directed, bipartite, node2vec, max_iter, p):
g = Graph(directed=False)
g.add_vertex(max(nodes) + 1)
for edge in train_set:
if edge not in poss_set:
u, w = map(int, edge.split())
g.add_edge(g.vertex(u), g.vertex(w))
add_auc(auc, "sfdp-default",
calculate_auc_default(g, max_iter, poss_set, neg_set, p, directed))
if not directed:
add_auc(auc, "PA", calculate_auc_PA(g, poss_set, neg_set))
add_auc(auc, "CN", calculate_auc_CN(g, poss_set, neg_set))
add_auc(auc, "Adamic-Adar",
calculate_auc_Adamic_Adar(g, poss_set, neg_set))
add_auc(
auc, "NMF-100",
calculate_auc_NMF(100, train_set, nodes, poss_set, neg_set, directed))
add_auc(
auc, "svds-100",
calculate_auc_SVDS(100, train_set, nodes, poss_set, neg_set, directed))
if bipartite:
is_bi, part = graph_tool.topology.is_bipartite(g, partition=True)
if is_bi:
groups = g.new_vertex_property("int")
for u in g.vertices():
groups[u] = int(part[u])
left = "repulse-aliens"
right = "repulse-aliens"
pos_bip = sfdp_layout(g,
groups=groups,
verbose=verbose,
bipartite=True,
bipartite_method=[left, right],
gap=gap)
features = tools.TopologicalFeatures(g, pos_bip, gap=gap)
X, Y = tools.make_dataset(poss_set, neg_set, [features.dist])
add_auc(auc, "sfdp-bipartite", roc_auc_score(Y, X))
if directed:
g_di = Graph(directed=False)
g_di.add_vertex(2 * max(nodes) + 2)
for edge in train_set:
if edge not in poss_set:
u, w = map(int, edge.split())
g_di.add_edge(g_di.vertex(2 * u + 1), g_di.vertex(2 * w))
add_auc(auc, "sfdp-directed",
calculate_auc_directed(g_di, verbose, gap, poss_set, neg_set))
if node2vec:
add_auc(
auc, 'Node2Vec-100-undir',
calculate_auc_Node2Vec(train_set, poss_set, neg_set, 100, False))
def calculate_auc_NMF(n_components, train_set, nodes, poss_set, neg_set,
directed):
from sklearn.decomposition import NMF
model = NMF(n_components=n_components, init='random', random_state=0)
matrix = tools.make_sparse_matrix(train_set,
nodes,
poss_set,
directed=directed)
features = tools.MFFeatures(model, matrix)
X, Y = tools.make_dataset(poss_set, neg_set, [features.score])
return roc_auc_score(Y, X)
def calculate_auc_PA(g, poss_set, neg_set):
features = tools.TopologicalFeatures(g)
edges = [e for e in g.edges()]
poss_set = set(poss_set)
neg_set = set(neg_set)
for e in edges:
e1 = str(e.source()) + ' ' + str(e.target())
e2 = str(e.target()) + ' ' + str(e.source())
X, Y = tools.make_dataset(poss_set, neg_set,
[features.preferential_attachment])
return roc_auc_score(Y, X)
def calculate_auc_directed(g, verbose, gap, poss_set, neg_set):
groups = g.new_vertex_property("int")
for u in g.vertices():
groups[u] = int(u) % 2
pos_directed = sfdp_layout(g,
groups=groups,
verbose=verbose,
bipartite=True,
gap=gap)
features = tools.TopologicalFeatures(g,
pos_directed,
directed=True,
gap=gap)
X, Y = tools.make_dataset(poss_set, neg_set, [features.dist])
return roc_auc_score(Y, X)
def calculate_auc_SVDS(n_components, train_set, nodes, poss_set, neg_set,
directed):
from scipy.sparse import linalg
import numpy
matrix = tools.make_sparse_matrix(train_set,
nodes,
poss_set,
directed=directed)
U, s, Vh = linalg.svds(matrix.asfptype(), k=n_components)
U = U * s
def score(u, w):
return numpy.dot(U[u], Vh.T[w])
X, Y = tools.make_dataset(poss_set, neg_set, [score])
return roc_auc_score(Y, X)
def calculate_auc_Adamic_Adar(g, poss_set, neg_set):
features = tools.TopologicalFeatures(g)
X, Y = tools.make_dataset(poss_set, neg_set,
[features.Adamic_Adar_coefficient])
return roc_auc_score(Y, X)
def calculate_auc_CN(g, poss_set, neg_set):
features = tools.TopologicalFeatures(g)
X, Y = tools.make_dataset(poss_set, neg_set, [features.common_neighbors])
return roc_auc_score(Y, X)
def calculate_auc_default(g, max_iter, poss_set, neg_set, p, directed):
pos_default = sfdp_layout(g, max_iter=max_iter, p=p)
features = tools.TopologicalFeatures(g, pos_default, directed=directed)
X, Y = tools.make_dataset(poss_set, neg_set, [features.dist])
return roc_auc_score(Y, X)