-
Notifications
You must be signed in to change notification settings - Fork 9
/
Copy pathmodel.py
322 lines (257 loc) · 12.7 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
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
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
import os
import math
import pickle
import numpy as np
import torch
import torch.autograd as autograd
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from LSTMLinear import LSTMModel
from sklearn.metrics.pairwise import pairwise_distances, cosine_similarity
USE_CUDA = torch.cuda.is_available()
if USE_CUDA:
longTensor = torch.cuda.LongTensor
floatTensor = torch.cuda.FloatTensor
else:
longTensor = torch.LongTensor
floatTensor = torch.FloatTensor
class TTransEModel(nn.Module):
def __init__(self, config):
super(TTransEModel, self).__init__()
self.learning_rate = config.learning_rate
self.early_stopping_round = config.early_stopping_round
self.L1_flag = config.L1_flag
self.filter = config.filter
self.embedding_size = config.embedding_size
self.entity_total = config.entity_total
self.relation_total = config.relation_total
self.tem_total = config.tem_total
self.batch_size = config.batch_size
ent_weight = floatTensor(self.entity_total, self.embedding_size)
rel_weight = floatTensor(self.relation_total, self.embedding_size)
tem_weight = floatTensor(self.tem_total, self.embedding_size)
# Use xavier initialization method to initialize embeddings of entities and relations
nn.init.xavier_uniform(ent_weight)
nn.init.xavier_uniform(rel_weight)
nn.init.xavier_uniform(tem_weight)
self.ent_embeddings = nn.Embedding(self.entity_total, self.embedding_size)
self.rel_embeddings = nn.Embedding(self.relation_total, self.embedding_size)
self.tem_embeddings = nn.Embedding(self.tem_total, self.embedding_size)
self.ent_embeddings.weight = nn.Parameter(ent_weight)
self.rel_embeddings.weight = nn.Parameter(rel_weight)
self.tem_embeddings.weight = nn.Parameter(tem_weight)
normalize_entity_emb = F.normalize(self.ent_embeddings.weight.data, p=2, dim=1)
normalize_relation_emb = F.normalize(self.rel_embeddings.weight.data, p=2, dim=1)
normalize_temporal_emb = F.normalize(self.tem_embeddings.weight.data, p=2, dim=1)
self.ent_embeddings.weight.data = normalize_entity_emb
self.rel_embeddings.weight.data = normalize_relation_emb
self.tem_embeddings.weight.data = normalize_temporal_emb
def forward(self, pos_h, pos_t, pos_r, pos_tem, neg_h, neg_t, neg_r, neg_tem):
pos_h_e = self.ent_embeddings(pos_h)
pos_t_e = self.ent_embeddings(pos_t)
pos_r_e = self.rel_embeddings(pos_r)
pos_tem_e = self.tem_embeddings(pos_tem)
neg_h_e = self.ent_embeddings(neg_h)
neg_t_e = self.ent_embeddings(neg_t)
neg_r_e = self.rel_embeddings(neg_r)
neg_tem_e = self.tem_embeddings(neg_tem)
if self.L1_flag:
pos = torch.sum(torch.abs(pos_h_e + pos_r_e + pos_tem_e - pos_t_e), 1)
neg = torch.sum(torch.abs(neg_h_e + neg_r_e + neg_tem_e - neg_t_e), 1)
else:
pos = torch.sum((pos_h_e + pos_r_e + pos_tem_e - pos_t_e) ** 2, 1)
neg = torch.sum((neg_h_e + neg_r_e + neg_tem_e - neg_t_e) ** 2, 1)
return pos, neg
class TADistmultModel(nn.Module):
def __init__(self, config):
super(TADistmultModel, self).__init__()
self.learning_rate = config.learning_rate
self.early_stopping_round = config.early_stopping_round
self.L1_flag = config.L1_flag
self.filter = config.filter
self.embedding_size = config.embedding_size
self.entity_total = config.entity_total
self.relation_total = config.relation_total
self.tem_total = config.tem_total # 32
self.batch_size = config.batch_size
self.criterion = nn.Softplus()
torch.nn.BCELoss()
self.dropout = nn.Dropout(config.dropout)
self.lstm = LSTMModel(self.embedding_size, n_layer=1)
ent_weight = floatTensor(self.entity_total, self.embedding_size)
rel_weight = floatTensor(self.relation_total, self.embedding_size)
tem_weight = floatTensor(self.tem_total, self.embedding_size)
# Use xavier initialization method to initialize embeddings of entities and relations
nn.init.xavier_uniform(ent_weight)
nn.init.xavier_uniform(rel_weight)
nn.init.xavier_uniform(tem_weight)
self.ent_embeddings = nn.Embedding(self.entity_total, self.embedding_size)
self.rel_embeddings = nn.Embedding(self.relation_total, self.embedding_size)
self.tem_embeddings = nn.Embedding(self.tem_total, self.embedding_size)
self.ent_embeddings.weight = nn.Parameter(ent_weight)
self.rel_embeddings.weight = nn.Parameter(rel_weight)
self.tem_embeddings.weight = nn.Parameter(tem_weight)
normalize_entity_emb = F.normalize(self.ent_embeddings.weight.data, p=2, dim=1)
normalize_relation_emb = F.normalize(self.rel_embeddings.weight.data, p=2, dim=1)
normalize_temporal_emb = F.normalize(self.tem_embeddings.weight.data, p=2, dim=1)
self.ent_embeddings.weight.data = normalize_entity_emb
self.rel_embeddings.weight.data = normalize_relation_emb
self.tem_embeddings.weight.data = normalize_temporal_emb
def scoring(self, h, t, r):
return torch.sum(h * t * r, 1, False)
def forward(self, pos_h, pos_t, pos_r, pos_tem, neg_h, neg_t, neg_r, neg_tem):
pos_h_e = self.ent_embeddings(pos_h)
pos_t_e = self.ent_embeddings(pos_t)
pos_rseq_e = self.get_rseq(pos_r, pos_tem)
neg_h_e = self.ent_embeddings(neg_h)
neg_t_e = self.ent_embeddings(neg_t)
neg_rseq_e = self.get_rseq(neg_r, neg_tem)
pos_h_e = self.dropout(pos_h_e)
pos_t_e = self.dropout(pos_t_e)
pos_rseq_e = self.dropout(pos_rseq_e)
neg_h_e = self.dropout(neg_h_e)
neg_t_e = self.dropout(neg_t_e)
neg_rseq_e = self.dropout(neg_rseq_e)
pos = self.scoring(pos_h_e, pos_t_e, pos_rseq_e)
neg = self.scoring(neg_h_e, neg_t_e, neg_rseq_e)
return pos, neg
def get_rseq(self, r, tem):
r_e = self.rel_embeddings(r)
r_e = r_e.unsqueeze(0).transpose(0, 1)
bs = tem.shape[0] # batch size
tem_len = tem.shape[1]
tem = tem.contiguous()
tem = tem.view(bs * tem_len)
token_e = self.tem_embeddings(tem)
token_e = token_e.view(bs, tem_len, self.embedding_size)
seq_e = torch.cat((r_e, token_e), 1)
hidden_tem = self.lstm(seq_e)
hidden_tem = hidden_tem[0, :, :]
rseq_e = hidden_tem
return rseq_e
class TATransEModel(nn.Module):
def __init__(self, config):
super(TATransEModel, self).__init__()
self.learning_rate = config.learning_rate
self.early_stopping_round = config.early_stopping_round
self.L1_flag = config.L1_flag
self.filter = config.filter
self.embedding_size = config.embedding_size
self.entity_total = config.entity_total
self.relation_total = config.relation_total
# print(self.relation_total)
# exit()
self.tem_total = 32
self.batch_size = config.batch_size
self.dropout = nn.Dropout(config.dropout)
self.lstm = LSTMModel(self.embedding_size, n_layer=1)
ent_weight = floatTensor(self.entity_total, self.embedding_size)
rel_weight = floatTensor(self.relation_total, self.embedding_size)
tem_weight = floatTensor(self.tem_total, self.embedding_size)
# Use xavier initialization method to initialize embeddings of entities and relations
nn.init.xavier_uniform(ent_weight)
nn.init.xavier_uniform(rel_weight)
nn.init.xavier_uniform(tem_weight)
self.ent_embeddings = nn.Embedding(self.entity_total, self.embedding_size)
self.rel_embeddings = nn.Embedding(self.relation_total, self.embedding_size)
self.tem_embeddings = nn.Embedding(self.tem_total, self.embedding_size)
self.ent_embeddings.weight = nn.Parameter(ent_weight)
self.rel_embeddings.weight = nn.Parameter(rel_weight)
self.tem_embeddings.weight = nn.Parameter(tem_weight)
normalize_entity_emb = F.normalize(self.ent_embeddings.weight.data, p=2, dim=1)
normalize_relation_emb = F.normalize(self.rel_embeddings.weight.data, p=2, dim=1)
normalize_temporal_emb = F.normalize(self.tem_embeddings.weight.data, p=2, dim=1)
self.ent_embeddings.weight.data = normalize_entity_emb
self.rel_embeddings.weight.data = normalize_relation_emb
self.tem_embeddings.weight.data = normalize_temporal_emb
def forward(self,loss_type,entity_total,pos_h, pos_t, pos_r, pos_tem, neg_h, neg_t, neg_r, neg_tem):
# print(loss_type)
# exit()
pos_h_e = self.ent_embeddings(pos_h)
pos_t_e = self.ent_embeddings(pos_t)
pos_rseq_e = self.get_rseq(pos_r, pos_tem)
neg_h_e = self.ent_embeddings(neg_h)
neg_t_e = self.ent_embeddings(neg_t)
neg_rseq_e = self.get_rseq(neg_r, neg_tem)
pos_h_e = self.dropout(pos_h_e)
pos_t_e = self.dropout(pos_t_e)
pos_rseq_e = self.dropout(pos_rseq_e)
neg_h_e = self.dropout(neg_h_e)
neg_t_e = self.dropout(neg_t_e)
neg_rseq_e = self.dropout(neg_rseq_e)
# ent_embeddings = self.ent_embeddings
# print((pos_h[:20]))
# exit()
if loss_type == 0:
if self.L1_flag:
pos = torch.sum(torch.abs(pos_h_e + pos_rseq_e - pos_t_e), 1)
neg = torch.sum(torch.abs(neg_h_e + neg_rseq_e - neg_t_e), 1)
else:
pos = torch.sum((pos_h_e + pos_rseq_e - pos_t_e) ** 2, 1)
neg = torch.sum((neg_h_e + neg_rseq_e - neg_t_e) ** 2, 1)
return pos, neg
else:
mylist = list(range(entity_total))
my_list_tensor = torch.tensor(np.array(mylist)).cuda()
pred_pos_t = pos_h_e + pos_rseq_e
pred_neg_t = neg_h_e + neg_rseq_e
pred_pos_h = pos_t_e - pos_rseq_e
pred_neg_h = neg_t_e - neg_rseq_e
ent_embeddings = self.ent_embeddings(my_list_tensor).cuda()
n = pred_pos_t.size(0)
# print(n)
m = ent_embeddings.size(0)
# print(m)
d = pred_pos_t.size(1)
# print(d)
pred_pos_t = pred_pos_t.unsqueeze(1).expand(n, m, d).cuda()
my_list_tensor = ent_embeddings.unsqueeze(0).expand(n, m, d).cuda()
# print(my_list_tensor.shape)
# exit() torch.pow(x, 2) # torch.pow(pred_pos_t - my_list_tensor, 2)
z1 = (1/(torch.sum(torch.pow(pred_pos_t - my_list_tensor, 2), dim = 2)+0.0001)).cuda()
# print(z[0][:40])
# print("dada")
# print(z1.shape)
# exit()
pred1 = F.softmax(z1, dim=0)
# print(pred1.shape)
# exit()
n = pred_pos_h.size(0)
m = ent_embeddings.size(0)
d = pred_pos_h.size(1)
pred_pos_h = pred_pos_h.unsqueeze(1).expand(n, m, d).cuda()
# my_list_tensor = ent_embeddings.unsqueeze(0).expand(n, m, d).cuda()
z2 = (1/(torch.sum(torch.pow(pred_pos_h - my_list_tensor ,2 ), dim = 2)+0.0001)).cuda()
pred2 = F.softmax(z2, dim=0)
# n = pred_neg_t.size(0)
# m = ent_embeddings.size(0)
# d = pred_neg_t.size(1)
# pred_neg_t = pred_neg_t.unsqueeze(1).expand(n, m, d).cuda()
# # my_list_tensor = ent_embeddings.unsqueeze(0).expand(n, m, d).cuda()
# z3 = (1/(torch.sum(torch.abs(pred_neg_t - my_list_tensor), dim = 2)+0.0001)*100).cuda()
# pred3 = F.softmax(z3, dim=0)
# n = pred_neg_h.size(0)
# m = ent_embeddings.size(0)
# d = pred_neg_h.size(1)
# pred_neg_h = pred_neg_h.unsqueeze(1).expand(n, m, d).cuda()
# # my_list_tensor = ent_embeddings.unsqueeze(0).expand(n, m, d).cuda()
# z4 = (1/(torch.sum(torch.abs(pred_neg_h - my_list_tensor), dim = 2)+0.0001)*100).cuda()
# pred4 = F.softmax(z3, dim=0)
pred = torch.cat((pred1, pred2), 0)
target = torch.cat((pos_t, pos_h), 0)
return pred,target
def get_rseq(self, r, tem):
r_e = self.rel_embeddings(r)
r_e = r_e.unsqueeze(0).transpose(0, 1)
bs = tem.shape[0] # batch size
tem_len = tem.shape[1]
tem = tem.contiguous()
tem = tem.view(bs * tem_len)
token_e = self.tem_embeddings(tem)
token_e = token_e.view(bs, tem_len, self.embedding_size)
seq_e = torch.cat((r_e, token_e), 1)
hidden_tem = self.lstm(seq_e)
hidden_tem = hidden_tem[0, :, :]
rseq_e = hidden_tem
return rseq_e