def main():
parser = argparse.ArgumentParser()
# parser.add_argument('--num-walks', default=10, type=int)
# parser.add_argument('--walk-length', default=40, type=int)
# parser.add_argument('--dimension', type=int, default=128, help='Embeddings dimension')
# parser.add_argument('--iter', default=1, type=int, help='Number of epochs in SGD')
# parser.add_argument('--model', default='word2vec', help='Type of model to apply on walks (word2vec/skipgram)')
# # parser.add_argument("--emb", required=True)
# parser.add_argument('--emb', default='.deepwalk.embeddings', help='the embedding file')
# # parser.add_argument("--net", required=True)
# # parser.add_argument("--labels", required=True)
# parser.add_argument('--dic-network-name', default='network')
# parser.add_argument('--dic-label-name', default='label')
parser.add_argument('--walk-length', type=int, default=40)
parser.add_argument('--num-walks', type=int, default=10)
parser.add_argument('--dimension',
type=int,
default=128,
help='Embeddings dimension')
parser.add_argument('--iter',
default=1,
type=int,
help='Number of epochs in SGD')
parser.add_argument('--p',
type=float,
default=1,
help='Return hyperparameter')
parser.add_argument('--q',
type=float,
default=1,
help='Input hyperparameter')
parser.add_argument('--dataset', default='cora', help='dataset')
args = parser.parse_args()
#load dataset
if args.dataset == "polblogs":
graph, _, labels, idx_train, idx_test = load_polblogs_data()
else:
graph, _, labels, idx_train, idx_val, idx_test = load_data(
args.dataset)
## Load Embeddings
# embeddings_file = op.join(args.dataset, args.dataset + args.emb)
# model = KeyedVectors.load_word2vec_format(embeddings_file, binary=False)
model = node2vec(graph, args)
# Map nodes to their features (note: assumes nodes are labeled as integers 1:N)
print('model', model)
print(len(graph))
features_matrix = np.asarray(
[model[str(node)] for node in range(len(graph))])
labels_matrix = np.matrix(labels)
labels_matrix = sp.csr_matrix(labels_matrix)
labels = np.reshape(labels, (labels.shape[0], ))
X, y = features_matrix, labels
# y = MultiLabelBinarizer().fit_transform(y)
y_train = y[idx_train]
y_test = y[idx_test]
X_train = X[idx_train]
X_test = X[idx_test]
# y_train = format_csr(y_train_)
# y_test = format_csr(y_test_)
## Logistic Regression
# Train on data
logisticRegr = LogisticRegression()
logisticRegr.fit(X_train, y_train)
# Measure accuracy
score = logisticRegr.score(X_test, y_test)
clean_pred = logisticRegr.predict(X_test)
# Output results
print('---------------------------------')
print('Accuracy Score : ', score)
print('the prediction of y: ', clean_pred)
print('---------------------------------')
total_asr = []
for i in range(10):
# for i in [0,7,8,9]:
print('The perturbation idx', i)
perturb = np.array([
float(line.rstrip('\n')) for line in open(
'../GUA/perturbation_results/{1}_xi4_epoch100/perturbation_{1}_{0}.txt'
.format(i, args.dataset))
])
perturb = np.where(perturb > 0.5, 1, 0)
pt = np.where(perturb > 0)[0].tolist()
print('The perturbations are', pt)
asr_coll = []
for j in range(len(idx_test)):
neigh = list(graph.neighbors(idx_test[j]))
print('the neighrbors of node {} is'.format(j), neigh)
tmp_G = graph.copy()
# print ('the neighrbors of node {} is'.format(j), neigh)
# print ('The edges in clean graph', tmp_G.number_of_edges())
for k in pt:
if k in neigh:
print('the node is', idx_test[j])
# print ('the neighor is', k)
# print ('the edges between', tmp_G.has_edge(idx_test[j],k))
# print ('the edges between', tmp_G.has_edge(k, idx_test[j]))
# print ('the edges of 12', tmp_G.has_edge(k,j))
tmp_G.remove_edge(idx_test[j], k)
# print ('the edges of 12', tmp_G.has_edge(k,j))
# tmp_G.remove_edge(k, j)
else:
tmp_G.add_edge(idx_test[j], k, weight=1)
# tmp_G.add_edge(k, j)
print('The edges in attacked graph', tmp_G.number_of_edges())
model = node2vec(tmp_G, args) #the attacked graph embedding
features_matrix = np.asarray(
[model[str(node)] for node in range(len(graph))])
X = features_matrix
X_test = X[idx_test]
# idx_pred = j - idx_test[0]
att_pred = logisticRegr.predict(X_test)
if clean_pred[j] == att_pred[j]:
asr_coll.append(0)
print('attack fail', j)
else:
asr_coll.append(1)
print('attack success', j)
print('the success number', sum(asr_coll))
print('the total number', len(asr_coll))
avg_asr = float(sum(asr_coll) / len(asr_coll))
print('the asr is', avg_asr)
total_asr.append(avg_asr)
print('the total asr over 10 experiments is:', total_asr)
type=str,
default="cora",
help='The name of the network dataset.')
parser.add_argument('--radius',
type=int,
default=4,
help='The radius of l2 norm projection')
args = parser.parse_args()
np.random.seed(args.seed)
torch.manual_seed(args.seed)
if args.cuda:
torch.cuda.manual_seed(args.seed)
if args.dataset == "polblogs":
tmp_adj, features, labels, idx_train, idx_test = load_polblogs_data()
else:
_, features, labels, idx_train, idx_val, idx_test, tmp_adj = load_data(
args.dataset)
num_classes = labels.max().item() + 1
# tmp_adj = tmp_adj.toarray()
adj = tmp_adj
adj = np.eye(tmp_adj.shape[0]) + adj
adj, _ = normalize(adj)
adj = torch.from_numpy(adj.astype(np.float32))
# print (sum(features))
# print (labels.shape)
# print (idx_train.shape)
type=int,
default=10,
help='The learning step of updating the connection entries')
parser.add_argument('--sample_percent',
type=int,
default=40,
help='The sampling ratio of train set')
args = parser.parse_args()
np.random.seed(args.seed)
torch.manual_seed(args.seed)
if args.cuda:
torch.cuda.manual_seed(args.seed)
if args.dataset == "polblogs":
tmp_adj, tmp_feat, labels, train_idx, val_idx, test_idx = load_polblogs_data(
)
else:
_, _, labels, train_idx, val_idx, test_idx, tmp_adj, tmp_feat = load_data(
args.dataset)
num_classes = labels.max().item() + 1
# tmp_adj = tmp_adj.toarray()
adj = tmp_adj
adj = np.eye(tmp_adj.shape[0]) + adj
adj, _ = normalize(adj)
adj = torch.from_numpy(adj.astype(np.float32))
feat, _ = normalize(tmp_feat)
feat = torch.FloatTensor(np.array(feat.todense()))
tmp_feat = tmp_feat.todense()
num_fake = int(tmp_adj.shape[0] * args.fake_rate)
