-
Notifications
You must be signed in to change notification settings - Fork 1
Expand file tree
/
Copy pathmodels.py
More file actions
270 lines (216 loc) · 12.3 KB
/
Copy pathmodels.py
File metadata and controls
270 lines (216 loc) · 12.3 KB
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
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
import torch
import torch.nn as nn
from torch_geometric.nn import MessagePassing
from torch_geometric.nn import global_mean_pool
import torch.nn.functional as F
from conv_syn import GINConv, GNNSynEncoder, GraphSynMasker
# from conv_mol import GINMolHeadEncoder, GraphMolMasker, GNNMolTailEncoder, vGINMolHeadEncoder
from conv_mol_new import GINMolHeadEncoder, GraphMolMasker, GNNMolTailEncoder
import pdb
class CausalAdvGNNSyn(torch.nn.Module):
def __init__(self, num_class,
in_dim,
emb_dim=300,
fro_layer=2,
bac_layer=2,
cau_layer=2,
att_layer=2,
dropout_rate=0.5,
cau_gamma=0.4,
adv_gamma_node=1.0,
adv_gamma_edge=1.0):
super(CausalAdvGNNSyn, self).__init__()
self.cau_gamma = cau_gamma
self.adv_gamma_node = adv_gamma_node
self.adv_gamma_edge = adv_gamma_edge
self.dropout_rate = dropout_rate
self.emb_dim = emb_dim
self.num_class = num_class
self.wasserstein_distance = nn.MSELoss()
self.graph_front = GNNSynEncoder(fro_layer, in_dim, emb_dim, dropout_rate)
self.graph_backs = GNNSynEncoder(bac_layer, emb_dim, emb_dim, dropout_rate)
self.causaler = GraphSynMasker(cau_layer, in_dim, emb_dim, dropout_rate)
self.attacker = GraphSynMasker(att_layer, in_dim, emb_dim, dropout_rate)
self.pool = global_mean_pool
self.predictor = torch.nn.Linear(emb_dim, num_class)
def forward_causal(self, data, vis=False):
x, edge_index, batch = data.x, data.edge_index, data.batch
x_encode = self.graph_front(x, edge_index)
causaler_output = self.causaler(data)
node_cau, edge_cau = causaler_output["node_key"], causaler_output["edge_key"]
if vis:
return node_cau, edge_cau
h_node_cau = self.graph_backs(x_encode, edge_index, node_cau, edge_cau)
h_graph_cau = self.pool(h_node_cau, batch)
pred_cau = self.predictor(h_graph_cau)
return pred_cau
def forward_combined_inference(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
x_encode = self.graph_front(x, edge_index)
attacker_output = self.attacker(data)
causaler_output = self.causaler(data)
node_adv, edge_adv = attacker_output["node_key"], attacker_output["edge_key"]
node_cau, edge_cau = causaler_output["node_key"], causaler_output["edge_key"]
node_com = (1 - node_cau) * node_adv + node_cau
edge_com = (1 - edge_cau) * edge_adv + edge_cau
h_node_com = self.graph_backs(x_encode, edge_index, node_com, edge_com)
h_graph_com = self.pool(h_node_com, batch)
pred_com = self.predictor(h_graph_com)
return pred_com
def forward_attack(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
x_encode = self.graph_front(x, edge_index)
attacker_output = self.attacker(data)
node_adv, edge_adv = attacker_output["node_key"], attacker_output["edge_key"]
node_adv_num, node_env_num = attacker_output["node_key_num"], attacker_output["node_env_num"]
edge_adv_num, edge_env_num = attacker_output["edge_key_num"], attacker_output["edge_env_num"]
h_node_adv = self.graph_backs(x_encode, edge_index, node_adv, edge_adv)
h_node_ori = self.graph_backs(x_encode, edge_index)
h_graph_adv = self.pool(h_node_adv, batch)
h_graph_ori = self.pool(h_node_ori, batch)
loss_dis = self.wasserstein_distance(h_graph_adv, h_graph_ori)
pred_adv = self.predictor(h_graph_adv)
adv_node_reg = self.reg_mask_loss(node_adv_num, node_env_num, self.adv_gamma_node, self.attacker.non_zero_node_ratio)
adv_edge_reg = self.reg_mask_loss(edge_adv_num, edge_env_num, self.adv_gamma_edge, self.attacker.non_zero_edge_ratio)
adv_loss_reg = adv_node_reg + adv_edge_reg
output = {'pred_adv': pred_adv, 'loss_dis': loss_dis,
'adv_loss_reg': adv_loss_reg,
'node_adv': node_adv.mean().item(),
'edge_adv': edge_adv.mean().item()}
return output
def forward_advcausal(self, data, vis=False):
x, edge_index, batch = data.x, data.edge_index, data.batch
x_encode = self.graph_front(x, edge_index)
attacker_output = self.attacker(data)
causaler_output = self.causaler(data)
node_adv, edge_adv = attacker_output["node_key"], attacker_output["edge_key"]
node_cau, edge_cau = causaler_output["node_key"], causaler_output["edge_key"]
node_cau_num, node_env_num = causaler_output["node_key_num"], causaler_output["node_env_num"]
edge_cau_num, edge_env_num = causaler_output["edge_key_num"], causaler_output["edge_env_num"]
node_com = (1 - node_cau) * node_adv + node_cau
edge_com = (1 - edge_cau) * edge_adv + edge_cau
h_node_cau = self.graph_backs(x_encode, edge_index, node_cau, edge_cau)
h_graph_cau = self.pool(h_node_cau, batch)
pred_cau = self.predictor(h_graph_cau)
h_node_com = self.graph_backs(x_encode, edge_index, node_com, edge_com)
h_graph_com = self.pool(h_node_com, batch)
pred_com = self.predictor(h_graph_com)
cau_node_reg = self.reg_mask_loss(node_cau_num, node_env_num, self.cau_gamma, self.causaler.non_zero_node_ratio)
cau_edge_reg = self.reg_mask_loss(edge_cau_num, edge_env_num, self.cau_gamma, self.causaler.non_zero_edge_ratio)
cau_loss_reg = cau_node_reg + cau_edge_reg
if vis:
# self.plot_state(node_cau, edge_cau, "cau")
# self.plot_state(node_adv, edge_adv, "adv")
# self.plot_state(node_com, edge_com, "com")
# return node_com.cpu(), edge_com.cpu()
return node_com, edge_com
pred_advcausal = {'pred_cau': pred_cau, 'pred_com': pred_com,
'cau_loss_reg':cau_loss_reg,
'node_cau': node_cau.mean().item(),
'edge_cau': edge_cau.mean().item(),
'node_adv': node_adv.mean().item(),
'edge_adv': edge_adv.mean().item(),
'node_com': node_com.mean().item(),
'edge_com': edge_com.mean().item()}
return pred_advcausal
def reg_mask_loss(self, key_mask, env_mask, gamma, non_zero_ratio):
loss_reg = torch.abs(key_mask / (key_mask + env_mask) - gamma * torch.ones_like(key_mask)).mean()
loss_reg += (non_zero_ratio - gamma * torch.ones_like(key_mask)).mean()
return loss_reg
class CausalAdvGNNMol(torch.nn.Module):
def __init__(self, num_class,
emb_dim=300,
fro_layer=2,
bac_layer=2,
cau_layer=2,
att_layer=2,
cau_gamma=0.4,
adv_gamma_node=1.0,
adv_gamma_edge=1.0,):
super(CausalAdvGNNMol, self).__init__()
self.cau_gamma = cau_gamma
self.adv_gamma_node = adv_gamma_node
self.adv_gamma_edge = adv_gamma_edge
self.dropout_rate = 0.5
self.emb_dim = emb_dim
self.num_class = num_class
self.wasserstein_distance = nn.MSELoss()
self.graph_front = GINMolHeadEncoder(fro_layer, emb_dim)
# self.graph_front = vGINMolHeadEncoder(fro_layer, emb_dim)
self.graph_backs = GNNMolTailEncoder(bac_layer, emb_dim)
self.causaler = GraphMolMasker(cau_layer, emb_dim)
self.attacker = GraphMolMasker(att_layer, emb_dim)
self.pool = global_mean_pool
self.predictor = torch.nn.Linear(emb_dim, num_class)
# self.predictor = nn.Sequential(
# nn.Linear(emb_dim, 2 * emb_dim),
# nn.BatchNorm1d(2 * emb_dim),
# nn.ReLU(),
# nn.Linear(2 * emb_dim, num_class))
def forward_causal(self, data, vis=False):
x, edge_index, edge_attr, batch = data.x, data.edge_index, data.edge_attr, data.batch
x_encode = self.graph_front(x, edge_index, edge_attr, batch)
causaler_output = self.causaler(data)
node_cau, edge_cau = causaler_output["node_key"], causaler_output["edge_key"]
if vis:
return node_cau, edge_cau
h_node_cau = self.graph_backs(x_encode, edge_index, edge_attr, batch, node_cau, edge_cau)
h_graph_cau = self.pool(h_node_cau, batch)
pred_cau = self.predictor(h_graph_cau)
return pred_cau
def forward_attack(self, data):
x, edge_index, edge_attr, batch = data.x, data.edge_index, data.edge_attr, data.batch
x_encode = self.graph_front(x, edge_index, edge_attr, batch)
attacker_output = self.attacker(data)
node_adv, edge_adv = attacker_output["node_key"], attacker_output["edge_key"]
node_adv_num, node_env_num = attacker_output["node_key_num"], attacker_output["node_env_num"]
edge_adv_num, edge_env_num = attacker_output["edge_key_num"], attacker_output["edge_env_num"]
h_node_adv = self.graph_backs(x_encode, edge_index, edge_attr, batch, node_adv, edge_adv)
h_node_ori = self.graph_backs(x_encode, edge_index, edge_attr, batch)
h_graph_adv = self.pool(h_node_adv, batch)
h_graph_ori = self.pool(h_node_ori, batch)
loss_dis = self.wasserstein_distance(h_graph_adv, h_graph_ori)
pred_adv = self.predictor(h_graph_adv)
adv_node_reg = self.reg_mask_loss(node_adv_num, node_env_num, self.adv_gamma_node, self.attacker.non_zero_node_ratio)
adv_edge_reg = self.reg_mask_loss(edge_adv_num, edge_env_num, self.adv_gamma_edge, self.attacker.non_zero_edge_ratio)
adv_loss_reg = adv_node_reg + adv_edge_reg
output = {'pred_adv': pred_adv, 'loss_dis': loss_dis,
'adv_loss_reg': adv_loss_reg,
'node_adv': node_adv.mean().item(),
'edge_adv': edge_adv.mean().item()}
return output
def forward_advcausal(self, data, vis=False):
x, edge_index, edge_attr, batch = data.x, data.edge_index, data.edge_attr, data.batch
x_encode = self.graph_front(x, edge_index, edge_attr, batch)
attacker_output = self.attacker(data)
causaler_output = self.causaler(data)
node_adv, edge_adv = attacker_output["node_key"], attacker_output["edge_key"]
node_cau, edge_cau = causaler_output["node_key"], causaler_output["edge_key"]
node_cau_num, node_env_num = causaler_output["node_key_num"], causaler_output["node_env_num"]
edge_cau_num, edge_env_num = causaler_output["edge_key_num"], causaler_output["edge_env_num"]
node_com = (1 - node_cau) * node_adv + node_cau
edge_com = (1 - edge_cau) * edge_adv + edge_cau
if vis:
return node_com, edge_com
h_node_cau = self.graph_backs(x_encode, edge_index, edge_attr, batch, node_cau, edge_cau)
h_graph_cau = self.pool(h_node_cau, batch)
pred_cau = self.predictor(h_graph_cau)
h_node_com = self.graph_backs(x_encode, edge_index, edge_attr, batch, node_com, edge_com)
h_graph_com = self.pool(h_node_com, batch)
pred_com = self.predictor(h_graph_com)
cau_node_reg = self.reg_mask_loss(node_cau_num, node_env_num, self.cau_gamma, self.causaler.non_zero_node_ratio)
cau_edge_reg = self.reg_mask_loss(edge_cau_num, edge_env_num, self.cau_gamma, self.causaler.non_zero_edge_ratio)
cau_loss_reg = cau_node_reg + cau_edge_reg
pred_advcausal = {'pred_cau': pred_cau, 'pred_com': pred_com,
'cau_loss_reg':cau_loss_reg,
'node_cau': node_cau.mean().item(),
'edge_cau': edge_cau.mean().item(),
'node_adv': node_adv.mean().item(),
'edge_adv': edge_adv.mean().item(),
'node_com': node_com.mean().item(),
'edge_com': edge_com.mean().item()}
return pred_advcausal
def reg_mask_loss(self, key_mask, env_mask, gamma, non_zero_ratio):
loss_reg = torch.abs(key_mask / (key_mask + env_mask) - gamma * torch.ones_like(key_mask)).mean()
loss_reg += (non_zero_ratio - gamma * torch.ones_like(key_mask)).mean()
return loss_reg