-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathconstruct_graph.py
216 lines (188 loc) · 9.66 KB
/
construct_graph.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
import numpy as np
import torch
from torch_geometric.utils import mask_feature,add_random_edge,dropout_adj
import copy
from torch_geometric.utils import to_dense_adj, dense_to_sparse
from torch_geometric.data import Data, InMemoryDataset
def construct_noisy_graph(data,perturb_ratio,mode='raw'):
noisy_data = copy.deepcopy(data)
if(mode=='raw'):
noisy_edge_index = data.edge_index
noisy_edge_weights = data.edge_weight
noisy_x = data.x
elif(mode=='random_noise'):
# random noise: inject/remove edges
print("raw graph:",data.edge_index.shape)
noisy_edge_index,added_edges=add_random_edge(data.edge_index,force_undirected=True,p=perturb_ratio)
print("add edge:",noisy_edge_index.shape)
noisy_edge_index,removed_edges=dropout_adj(data.edge_index,data.edge_weight,force_undirected=True,p=perturb_ratio)
print("remove edge:",noisy_edge_index.shape)
noisy_edge_index = torch.cat([noisy_edge_index,added_edges],dim=1).long()
print("updated graph:",noisy_edge_index.shape)
noisy_data.edge_index = noisy_edge_index
return noisy_data
def drop_feature(x, drop_prob):
drop_mask = torch.empty(
(x.size(1), ),
dtype=torch.float32,
device=x.device).uniform_(0, 1) < drop_prob
x = x.clone()
x[:, drop_mask] = 0
return x
def drop_adj_1by1(args,edge_index, edge_weight, p,device):
# update edge_index according to edge_weight
if(edge_weight!=None):
edge_index = edge_index[:,edge_weight.nonzero().flatten().long()]
edge_weight = torch.ones([edge_index.shape[1],]).to(device)
# rs = np.random.RandomState(args.seed)
# remain_mask = rs.binomial(1,p,edge_index.shape[1])
remain_mask = np.random.binomial(1,p,edge_index.shape[1])
remain_index = remain_mask.nonzero()[0]
remain_edge_index = edge_index[:,remain_index]
remain_edge_weight = torch.ones([remain_edge_index.shape[1],]).to(device)
return remain_edge_index,remain_edge_weight
def construct_augmentation_1(args, x, edge_index, edge_weight=None):
#
# graph 1:
noisy_edge_index,added_edges=add_random_edge(edge_index,force_undirected=True,p=args.add_edge_rate_1)
noisy_edge_index,noisy_edge_weight=dropout_adj(edge_index,edge_weight,force_undirected=True,p=args.drop_edge_rate_1)
if(len(added_edges)>0):
aug_edge_index_1 = torch.cat([noisy_edge_index,added_edges],dim=1).long()
else:
aug_edge_index_1 = noisy_edge_index.long()
aug_x_1 = drop_feature(x,drop_prob=args.drop_feat_rate_1)
aug_edge_weight_1 = edge_weight
# graph 2:
noisy_edge_index,added_edges=add_random_edge(edge_index,force_undirected=True,p=args.add_edge_rate_2)
noisy_edge_index,noisy_edge_weight=dropout_adj(edge_index,edge_weight,force_undirected=True,p=args.drop_edge_rate_2)
if(len(added_edges)>0):
aug_edge_index_2 = torch.cat([noisy_edge_index,added_edges],dim=1).long()
else:
aug_edge_index_2 = noisy_edge_index.long()
aug_x_2 = drop_feature(x,drop_prob=args.drop_feat_rate_2)
aug_edge_weight_2 = edge_weight
return aug_edge_index_1,aug_x_1,aug_edge_weight_1,aug_edge_index_2,aug_x_2,aug_edge_weight_2
def construct_augmentation_1by1(args, x, edge_index, edge_weight=None):
#
# graph 1:
noisy_edge_index,added_edges=add_random_edge(edge_index,force_undirected=True,p=args.add_edge_rate_1)
noisy_edge_index,noisy_edge_weight=dropout_adj(edge_index,edge_weight,force_undirected=True,p=args.drop_edge_rate_1)
if(len(added_edges)>0):
aug_edge_index_1 = torch.cat([noisy_edge_index,added_edges],dim=1).long()
else:
aug_edge_index_1 = noisy_edge_index.long()
aug_x_1 = drop_feature(x,drop_prob=args.drop_feat_rate_1)
aug_edge_weight_1 = noisy_edge_weight
# graph 2:
noisy_edge_index,added_edges=add_random_edge(edge_index,force_undirected=True,p=args.add_edge_rate_2)
noisy_edge_index,noisy_edge_weight=dropout_adj(edge_index,edge_weight,force_undirected=True,p=args.drop_edge_rate_2)
if(len(added_edges)>0):
aug_edge_index_2 = torch.cat([noisy_edge_index,added_edges],dim=1).long()
else:
aug_edge_index_2 = noisy_edge_index.long()
aug_x_2 = drop_feature(x,drop_prob=args.drop_feat_rate_2)
aug_edge_weight_2 = noisy_edge_weight
return aug_edge_index_1,aug_x_1,aug_edge_weight_1,aug_edge_index_2,aug_x_2,aug_edge_weight_2
# aug_edge_index_1,aug_x_1,aug_edge_index_2,aug_x_2 = construct_augmentation(noisy_data)
def construct_augmentation_overall(args, x, edge_index, edge_weight=None, device=None):
# graph 1:
aug_edge_index_1,aug_edge_weight_1 = drop_adj_1by1(args,edge_index, edge_weight, 1-args.drop_edge_rate_1,device)
aug_edge_index_1 = aug_edge_index_1.long()
aug_x_1 = drop_feature(x,drop_prob=args.drop_feat_rate_1)
# graph 2:
aug_edge_index_2,aug_edge_weight_2 = drop_adj_1by1(args,edge_index, edge_weight, 1-args.drop_edge_rate_2,device)
aug_edge_index_2 = aug_edge_index_2.long()
aug_x_2 = drop_feature(x,drop_prob=args.drop_feat_rate_2)
return aug_edge_index_1,aug_x_1,aug_edge_weight_1,aug_edge_index_2,aug_x_2,aug_edge_weight_2
def single_add_random_edges(idx_target, idx_add_nodes,device):
edge_list = []
for idx_add in idx_add_nodes:
edge_list.append([idx_target,idx_add])
edge_index = torch.tensor(edge_list).to(device).transpose(1,0)
row = torch.cat([edge_index[0], edge_index[1]])
col = torch.cat([edge_index[1],edge_index[0]])
edge_index = torch.stack([row,col])
return edge_index
def generate_node_noisy(args,data,idx_target,perturbation_size,device):
noisy_data = copy.deepcopy(data)
idx_overall = torch.tensor(range(data.num_nodes)).to(device)
# find connected nodes
idx_edge_index = (data.edge_index[0] == idx_target).nonzero().flatten()
idx_connected_nodes = data.edge_index[1][idx_edge_index]
idx_nonconnected_nodes = torch.tensor(list(set(np.array(idx_overall.cpu())) - set(np.array(idx_connected_nodes.cpu())))).to(device)
# permute the non-connected nodes
rs = np.random.RandomState(args.seed)
perm = rs.permutation(idx_nonconnected_nodes.shape[0])
idx_add_nodes = perm[:perturbation_size]
add_edge_index = single_add_random_edges(idx_target,idx_add_nodes,device)
update_edge_index = torch.cat([data.edge_index,add_edge_index],dim=1)
noisy_data.edge_index = update_edge_index
return noisy_data
def generate_node_noisy_global(args,data,perturbation_ratio,device):
rs = np.random.RandomState(args.seed)
N = data.x.shape[0]
noisy_data = copy.deepcopy(data)
perturbation_size = int(data.edge_index.shape[1] * perturbation_ratio)
edge_index_to_add = rs.randint(0, N, (2, perturbation_size))
edge_index_to_add = torch.tensor(edge_index_to_add)
# to undirect
row = torch.cat([edge_index_to_add[0], edge_index_to_add[1]])
col = torch.cat([edge_index_to_add[1],edge_index_to_add[0]])
edge_index_to_add = torch.stack([row,col]).to(device)
updated_edge_index = torch.cat([data.edge_index,edge_index_to_add],dim=1)
# updated_edge_index = torch.cat([data.edge_index,edge_index_to_add],dim=1)
noisy_data.edge_index = updated_edge_index
return noisy_data
def generate_graph_noisy(args,data,perturbation_size,device,to_undirected=True):
rs = np.random.RandomState(args.seed)
if(args.dataset == 'COLLAB'):
N = data.num_nodes
else:
N = data.x.shape[0]
noisy_data = copy.deepcopy(data)
edge_index_to_add = rs.randint(0, N, (2, perturbation_size))
edge_index_to_add = torch.tensor(edge_index_to_add)
if(to_undirected):
row = torch.cat([edge_index_to_add[0], edge_index_to_add[1]])
col = torch.cat([edge_index_to_add[1],edge_index_to_add[0]])
edge_index_to_add = torch.stack([row,col])
updated_edge_index = torch.cat([data.edge_index,edge_index_to_add],dim=1)
# updated_edge_index = torch.cat([data.edge_index,edge_index_to_add],dim=1)
noisy_data.edge_index = updated_edge_index
return noisy_data
def generate_graph_noisy(args,data,perturbation_size,device,to_undirected=True):
rs = np.random.RandomState(args.seed)
if(args.dataset == 'COLLAB'):
N = data.num_nodes
else:
N = data.x.shape[0]
noisy_data = copy.deepcopy(data)
edge_index_to_add = rs.randint(0, N, (2, perturbation_size))
edge_index_to_add = torch.tensor(edge_index_to_add)
if(to_undirected):
row = torch.cat([edge_index_to_add[0], edge_index_to_add[1]])
col = torch.cat([edge_index_to_add[1],edge_index_to_add[0]])
edge_index_to_add = torch.stack([row,col])
updated_edge_index = torch.cat([data.edge_index,edge_index_to_add],dim=1)
# updated_edge_index = torch.cat([data.edge_index,edge_index_to_add],dim=1)
noisy_data.edge_index = updated_edge_index
return noisy_data
def generate_graph_noisy_global(args,data,perturbation_ratio,device,to_undirected=True):
rs = np.random.RandomState(args.seed)
if(args.dataset == 'COLLAB'):
N = data.num_nodes
else:
N = data.x.shape[0]
noisy_data = copy.deepcopy(data)
perturbation_size = int(data.edge_index.shape[1] * perturbation_ratio)
edge_index_to_add = rs.randint(0, N, (2, perturbation_size))
edge_index_to_add = torch.tensor(edge_index_to_add)
if(to_undirected):
row = torch.cat([edge_index_to_add[0], edge_index_to_add[1]])
col = torch.cat([edge_index_to_add[1],edge_index_to_add[0]])
edge_index_to_add = torch.stack([row,col])
edge_index_to_add = edge_index_to_add.to(device)
updated_edge_index = torch.cat([data.edge_index,edge_index_to_add],dim=1)
# updated_edge_index = torch.cat([data.edge_index,edge_index_to_add],dim=1)
noisy_data.edge_index = updated_edge_index
return noisy_data