-
Notifications
You must be signed in to change notification settings - Fork 4
/
model.py
242 lines (220 loc) · 10.6 KB
/
model.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
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
from pyHGT.conv import *
class Classifier(nn.Module):
def __init__(self, n_hid, n_out):
super(Classifier, self).__init__()
self.n_hid = n_hid
self.n_out = n_out
self.linear = nn.Linear(n_hid, n_out)
def forward(self, x):
tx = self.linear(x)
return torch.log_softmax(tx.squeeze(), dim=-1)
def __repr__(self):
return '{}(n_hid={}, n_out={})'.format(
self.__class__.__name__, self.n_hid, self.n_out)
class Matcher(nn.Module):
'''
Matching between a pair of nodes to conduct link prediction.
Use multi-head attention as matching model.
'''
def __init__(self, n_hid):
super(Matcher, self).__init__()
self.left_linear = nn.Linear(n_hid, n_hid)
self.right_linear = nn.Linear(n_hid, n_hid)
self.sqrt_hd = math.sqrt(n_hid)
self.cache = None
def forward(self, x, y, infer = False, pair = False):
ty = self.right_linear(y)
if infer:
'''
During testing, we will consider millions or even billions of nodes as candidates (x).
It's not possible to calculate them again for different query (y)
Since the model is fixed, we propose to cache them, and dirrectly use the results.
'''
if self.cache != None:
tx = self.cache
else:
tx = self.left_linear(x)
self.cache = tx
else:
tx = self.left_linear(x)
if pair:
res = (tx * ty).sum(dim=-1)
else:
res = torch.matmul(tx, ty.transpose(0,1))
return res / self.sqrt_hd
def __repr__(self):
return '{}(n_hid={})'.format(
self.__class__.__name__, self.n_hid)
# class GNN(nn.Module):
# def __init__(self, in_dim, n_hid, num_types, num_relations, n_heads, n_layers, dropout = 0.2, conv_name = 'hgt', prev_norm = True, last_norm = True, use_RTE = True):
# super(GNN, self).__init__()
# self.gcs = nn.ModuleList()
# self.num_types = num_types
# self.in_dim = in_dim
# self.n_hid = n_hid
# self.adapt_ws = nn.ModuleList()
# self.drop = nn.Dropout(dropout)
# self.att =None
# self.emb =None
# self.conv_name = conv_name
# for t in range(num_types):
# self.adapt_ws.append(nn.Linear(in_dim, n_hid))
# for l in range(n_layers - 1):
# self.gcs.append(GeneralConv(conv_name, n_hid, n_hid, num_types, num_relations, n_heads, dropout, use_norm = prev_norm, use_RTE = use_RTE))
# self.gcs.append(GeneralConv(conv_name, n_hid, n_hid, num_types, num_relations, n_heads, dropout, use_norm = last_norm, use_RTE = use_RTE))
# def forward(self, node_feature, node_type, edge_time, edge_index, edge_type):
# res = torch.zeros(node_feature.size(0), self.n_hid).to(node_feature.device)
# for t_id in range(self.num_types):
# idx = (node_type == int(t_id))
# if idx.sum() == 0:
# continue
# res[idx] = torch.tanh(self.adapt_ws[t_id](node_feature[idx]))
# meta_xs = self.drop(res)
# del res
# self.att = {}
# i=0
# self.emb={}
# for gc in self.gcs:
# meta_xs = gc(meta_xs, node_type, edge_index, edge_type, edge_time)
# if (self.conv_name == 'hgt'):
# #self.att = gc.res_att
# self.att[i]=gc.res_att
# #self.emb[i]=gc.res
# i=i+1
# #print(gc.res_att)
# #for p in gc.parameters():
# # print(p)
# #self.att = gc.res_att
# self.att = self.att[0]
# return meta_xs
class GNN(nn.Module):
def __init__(self, in_dim, n_hid, num_types, num_relations, n_heads, n_layers, dropout = 0.2, conv_name = 'hgt', prev_norm = True, last_norm = True, use_RTE = True):
super(GNN, self).__init__()
self.gcs = nn.ModuleList()
self.num_types = num_types
self.in_dim = in_dim
self.n_hid = n_hid
self.adapt_ws = nn.ModuleList()
self.drop = nn.Dropout(dropout)
self.att =None
self.emb =None
self.conv_name = conv_name
for t in range(num_types):
self.adapt_ws.append(nn.Linear(in_dim, n_hid))
for l in range(n_layers - 1):
self.gcs.append(GeneralConv(conv_name, n_hid, n_hid, num_types, num_relations, n_heads, dropout, use_norm = prev_norm, use_RTE = use_RTE))
self.gcs.append(GeneralConv(conv_name, n_hid, n_hid, num_types, num_relations, n_heads, dropout, use_norm = last_norm, use_RTE = use_RTE))
def forward(self, node_feature, node_type, edge_time, edge_index, edge_type, cell_res):
res = torch.zeros(node_feature.size(0), self.n_hid).to(node_feature.device)
for t_id in range(self.num_types):
idx = (node_type == int(t_id))
if idx.sum() == 0:
continue
if int(t_id) == 0:
res[idx] = torch.tanh(self.adapt_ws[t_id](node_feature[idx]))
elif int(t_id) == 1:
res[idx] = cell_res
meta_xs = self.drop(res)
del res
self.att = {}
i=0
self.emb={}
for gc in self.gcs:
meta_xs = gc(meta_xs, node_type, edge_index, edge_type, edge_time)
if (self.conv_name == 'hgt'):
#self.att = gc.res_att
self.att[i]=gc.res_att
#self.emb[i]=gc.res
i=i+1
#print(gc.res_att)
#for p in gc.parameters():
# print(p)
#self.att = gc.res_att
self.att = self.att[0]
return meta_xs
class GNN_from_raw(nn.Module):
def __init__(self, in_dim, n_hid, num_types, num_relations, n_heads, n_layers, \
dropout = 0.2, conv_name = 'hgt', \
prev_norm = True, last_norm = True, \
use_RTE = True,\
AEtype=0\
):
super(GNN_from_raw, self).__init__()
self.gcs = nn.ModuleList()
self.num_types = num_types
self.in_dim = in_dim
self.n_hid = n_hid
self.adapt_ws = nn.ModuleList()
self.drop = nn.Dropout(dropout)
self.embedding1 = nn.ModuleList()
self.embedding2 = nn.ModuleList()
self.decode1 = nn.ModuleList()
self.decode2 = nn.ModuleList()
self.AEtype = AEtype
self.att =None
self.conv_name = conv_name
for ti in range(num_types):
#self.embedding.append(F.relu(nn.Linear(512,256)(F.relu(nn.Linear(in_dim[ti],512)))))
self.embedding1.append(nn.Linear(in_dim[ti],512)) #embedding1[0] [2713 x 512] embedding1[1] [24022 x 512]
self.embedding2.append(nn.Linear(512,256)) #embedding2[0] [512, 256] embedding2[1] [512,256]
if AEtype==1: #embedding autoencoder
for ti in range(num_types):
#self.embedding.append(F.relu(nn.Linear(512,256)(F.relu(nn.Linear(in_dim[ti],512)))))
self.decode1.append(nn.Linear(256,512)) #embedding1[0] [2713 x 512] embedding1[1] [24022 x 512]
self.decode2.append(nn.Linear(512,in_dim[ti])) #embedding2[0] [512, 256] embedding2[1] [512,256]
elif AEtype==2:
for ti in range(num_types):
#self.embedding.append(F.relu(nn.Linear(512,256)(F.relu(nn.Linear(in_dim[ti],512)))))
self.decode1.append(nn.Linear(n_hid,512)) #embedding1[0] [2713 x 512] embedding1[1] [24022 x 512]
self.decode2.append(nn.Linear(512,in_dim[ti])) #embedding2[0] [512, 256] embedding2[1] [512,256]
for t in range(num_types):
self.adapt_ws.append(nn.Linear(256, n_hid)) #256 could be one additional hyperparameter!!!
for l in range(n_layers - 1):
self.gcs.append(GeneralConv(conv_name, n_hid, n_hid, num_types, num_relations, n_heads, dropout, use_norm = prev_norm, use_RTE = use_RTE))
self.gcs.append(GeneralConv(conv_name, n_hid, n_hid, num_types, num_relations, n_heads, dropout, use_norm = last_norm, use_RTE = use_RTE))
def encode(self, x,t_id):
h1 = F.relu(self.embedding1[t_id](x))
return F.relu(self.embedding2[t_id](h1))
def decode(self, z,t_id):
h3 = F.relu(self.decode1[t_id](z))
return torch.relu(self.decode2[t_id](h3))
#return torch.relu(self.fc4(z))
def forward(self, node_feature, node_type, edge_time, edge_index, edge_type):
node_embedding=[] #len = 2
for t_id in range(self.num_types):
node_embedding += list(self.encode(node_feature[t_id],t_id))
node_embedding_stack = torch.stack(node_embedding)
#print("shape of node_embedding="+str(node_embedding_stack.shape)+"\n")
res = torch.zeros(node_embedding_stack.size(0), self.n_hid).to(node_feature[0].device)
for t_id in range(self.num_types):
idx = (node_type == int(t_id)) #0, 1
if idx.sum() == 0:
continue
#res[idx] = torch.tanh(self.adapt_ws[t_id](self.embedding[t_id](node_feature[idx])))
#print(idx)
res[idx] = torch.tanh(self.adapt_ws[t_id](node_embedding_stack[idx]))
#res[idx] = torch.tanh(self.adapt_ws[t_id](self.encode(node_feature[t_id],t_id)))
meta_xs = self.drop(res)
del res
for gc in self.gcs:
meta_xs = gc(meta_xs, node_type, edge_index, edge_type, edge_time)
if (self.conv_name == 'hgt'):
self.att = gc.res_att
if self.AEtype!=0:
if self.AEtype==1:#embedding auto-encoder
decode_embedding=[]
for t_id in range(self.num_types):
decode_embedding.append(self.decode(node_embedding_stack[node_type==t_id],t_id)) #0 genematrix 1 cellmatrix
#print(decode_embedding[t_id].shape)
#print(meta_xs[node_type==t_id].shape)
return meta_xs,decode_embedding
elif self.AEtype==2: #HGT embedding auto-encotder
decode_embedding = []
for t_id in range(self.num_types):
decode_embedding.append(self.decode(meta_xs[node_type==t_id],t_id)) #0 genematrix 1 cellmatrix
#print("in model decode_embedding shape tid="+str(t_id))
#print(decode_embedding[t_id].shape)
#print(meta_xs[node_type==t_id].shape)
return meta_xs,decode_embedding
else:
return meta_xs