def get_train_val_test(self):
"""Get training, validation, test splits according to self.setting (either 'nettack' or 'gcn').
"""
if self.setting == 'nettack':
return get_train_val_test(nnodes=self.adj.shape[0], val_size=0.1, test_size=0.8, stratify=self.labels, seed=self.seed)
if self.setting == 'gcn':
return get_train_val_test_gcn(self.labels, seed=self.seed)
def get_train_val_test(self):
if self.setting == 'nettack':
return get_train_val_test(nnodes=self.adj.shape[0],
val_size=0.1,
test_size=0.8,
stratify=self.labels,
seed=self.seed)
if self.setting == 'gcn':
return get_train_val_test_gcn(self.labels, seed=self.seed)
print(args)
np.random.seed(
15
) # Here the random seed is to split the train/val/test data, we need to set the random seed to be the same as that when you generate the perturbed graph
data = Dataset(root='/tmp/', name=args.dataset, setting='nettack')
adj, features, labels = data.adj, data.features, data.labels
idx_train, idx_val, idx_test = data.idx_train, data.idx_val, data.idx_test
if args.dataset == 'pubmed':
# just for matching the results in the paper; seed details in https://github.com/ChandlerBang/Pro-GNN/issues/2
idx_train, idx_val, idx_test = get_train_val_test(
adj.shape[0],
val_size=0.1,
test_size=0.8,
stratify=encode_onehot(labels),
seed=15)
if args.attack == 'no':
perturbed_adj = adj
if args.attack == 'random':
from deeprobust.graph.global_attack import Random
# to fix the seed of generated random attack, you need to fix both np.random and random
# you can uncomment the following code
# import random; random.seed(args.seed)
# np.random.seed(args.seed)
attacker = Random()
n_perturbations = int(args.ptb_rate * (adj.sum() // 2))
attacker.attack(adj, n_perturbations, type='add')