-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
313 lines (282 loc) · 12.3 KB
/
utils.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
import logging
import numpy as np
from pathlib import Path
import torch
import torch.nn as nn
import torch.nn.functional as F
def choose_predict2(predict_d1,domain_id):
predict_d1_cse, predict_d2_cse = [], []
for i in range(domain_id.shape[0]):
if domain_id[i][0] == 0:
predict_d1_cse.append(predict_d1[i,:])
else:
predict_d2_cse.append(predict_d1[i,:])
if len(predict_d1_cse)!=0:
predict_d1_cse = np.array(predict_d1_cse)
if len(predict_d2_cse)!=0:
predict_d2_cse = np.array(predict_d2_cse)
return predict_d1_cse, predict_d2_cse
def choose_predict(predict_d1,predict_d2,domain_id):
predict_d1_cse, predict_d2_cse = [], []
for i in range(domain_id.shape[0]):
if domain_id[i][0] == 0:
predict_d1_cse.append(predict_d1[i,:])
else:
predict_d2_cse.append(predict_d2[i,:])
if len(predict_d1_cse)!=0:
predict_d1_cse = np.array(predict_d1_cse)
if len(predict_d2_cse)!=0:
predict_d2_cse = np.array(predict_d2_cse)
return predict_d1_cse, predict_d2_cse
def choose_predict_SDoverlap(predict_d1,overlap_label):
predict_d1_cse_over, predict_d1_cse_nono = [], []
for i in range(predict_d1.shape[0]):
if overlap_label[i][0]==0:
predict_d1_cse_nono.append(predict_d1[i,:])
else:
predict_d1_cse_over.append(predict_d1[i,:])
if len(predict_d1_cse_over)!=0:
predict_d1_cse_over = np.array(predict_d1_cse_over)
if len(predict_d1_cse_nono)!=0:
predict_d1_cse_nono = np.array(predict_d1_cse_nono)
return predict_d1_cse_over, predict_d1_cse_nono
def choose_predict_overlap(predict_d1,predict_d2,domain_id,overlap_label):
predict_d1_cse_over, predict_d1_cse_nono, predict_d2_cse_over, predict_d2_cse_nono = [], [], [], []
for i in range(domain_id.shape[0]):
if domain_id[i][0] == 0:
if overlap_label[i][0]==0:
predict_d1_cse_nono.append(predict_d1[i,:])
else:
predict_d1_cse_over.append(predict_d1[i,:])
else:
if overlap_label[i][0]==0:
predict_d2_cse_nono.append(predict_d2[i,:])
else:
predict_d2_cse_over.append(predict_d2[i,:])
if len(predict_d1_cse_over)!=0:
predict_d1_cse_over = np.array(predict_d1_cse_over)
if len(predict_d1_cse_nono)!=0:
predict_d1_cse_nono = np.array(predict_d1_cse_nono)
if len(predict_d2_cse_over)!=0:
predict_d2_cse_over = np.array(predict_d2_cse_over)
if len(predict_d2_cse_nono)!=0:
predict_d2_cse_nono = np.array(predict_d2_cse_nono)
return predict_d1_cse_over, predict_d1_cse_nono, predict_d2_cse_over, predict_d2_cse_nono
def cal_loss_cl_all(u_feat_m1_d1, u_feat_m1_d2, u_feat_m2_d1,u_feat_m2_d2, u_feat_m3_d1,u_feat_m3_d2, u_feat_m4_d1, u_feat_m4_d2):
user_m1_query, user_m1_key = [], []
user_m1_query.append(u_feat_m1_d1)
user_m1_key.append(u_feat_m1_d2)
user_m1_key.append(u_feat_m2_d1)
user_m1_key.append(u_feat_m2_d2)
user_m1_key.append(u_feat_m3_d1)
user_m1_key.append(u_feat_m3_d2)
user_m1_key.append(u_feat_m4_d1)
user_m1_key.append(u_feat_m4_d2)
user_m1_key = torch.stack(user_m1_key,dim=-1).squeeze()
user_m1_query = torch.stack(user_m1_query,dim=-1).squeeze()
user_m1_query = torch.transpose(user_m1_query.unsqueeze(-1),1,2)
logits_m1 = torch.matmul(user_m1_query,user_m1_key).squeeze()
user_m2_query, user_m2_key = [], []
user_m2_query.append(u_feat_m2_d1)
user_m2_key.append(u_feat_m2_d2)
user_m2_key.append(u_feat_m1_d1)
user_m2_key.append(u_feat_m1_d2)
user_m2_key.append(u_feat_m3_d1)
user_m2_key.append(u_feat_m3_d2)
user_m2_key.append(u_feat_m4_d1)
user_m2_key.append(u_feat_m4_d2)
user_m2_key = torch.stack(user_m2_key,dim=-1).squeeze()
user_m2_query = torch.stack(user_m2_query,dim=-1).squeeze()
user_m2_query = torch.transpose(user_m2_query.unsqueeze(-1),1,2)
logits_m2 = torch.matmul(user_m2_query,user_m2_key).squeeze()
user_m3_query, user_m3_key = [], []
user_m3_query.append(u_feat_m3_d1)
user_m3_key.append(u_feat_m3_d2)
user_m3_key.append(u_feat_m1_d1)
user_m3_key.append(u_feat_m1_d2)
user_m3_key.append(u_feat_m2_d1)
user_m3_key.append(u_feat_m2_d2)
user_m3_key.append(u_feat_m4_d1)
user_m3_key.append(u_feat_m4_d2)
user_m3_key = torch.stack(user_m3_key,dim=-1).squeeze()
user_m3_query = torch.stack(user_m3_query,dim=-1).squeeze()
user_m3_query = torch.transpose(user_m3_query.unsqueeze(-1),1,2)
logits_m3 = torch.matmul(user_m3_query,user_m3_key).squeeze()
user_m4_query, user_m4_key = [], []
user_m4_query.append(u_feat_m4_d1)
user_m4_key.append(u_feat_m4_d2)
user_m4_key.append(u_feat_m1_d1)
user_m4_key.append(u_feat_m1_d2)
user_m4_key.append(u_feat_m2_d1)
user_m4_key.append(u_feat_m2_d2)
user_m4_key.append(u_feat_m3_d1)
user_m4_key.append(u_feat_m3_d2)
user_m4_key = torch.stack(user_m4_key,dim=-1).squeeze()
user_m4_query = torch.stack(user_m4_query,dim=-1).squeeze()
user_m4_query = torch.transpose(user_m4_query.unsqueeze(-1),1,2)
logits_m4 = torch.matmul(user_m4_query,user_m4_key).squeeze()
# print(user_spf3_query.shape,user_spf3_key.shape) #torch.Size([1024, 1, 128]) torch.Size([1024, 128, 7])
# u_feat_enhance_m1_d1 = u_feat_enhance_m1_d1.unsqueeze(-1)
# u_feat_enhance_m1_d2 = u_feat_enhance_m1_d2.unsqueeze(-1).repeat(1,1,u_feat_enhance_m1_d2.shape[0])
# u_feat_enhance_m1_d2 = torch.transpose(u_feat_enhance_m1_d2,1,2)
# logit_cl_m1 = torch.matmul(u_feat_enhance_m1_d2,u_feat_enhance_m1_d1).squeeze()
labels = torch.LongTensor(torch.zeros(logits_m1.shape[0]).long()).cuda()
# for i in range(logit_cl_m1.shape[0]):
# labels[i] = i
loss_cl = nn.CrossEntropyLoss()(logits_m1,labels)+nn.CrossEntropyLoss()(logits_m2,labels)+nn.CrossEntropyLoss()(logits_m3,labels)+nn.CrossEntropyLoss()(logits_m4,labels)
return loss_cl
def cal_loss_cl_refine(u_feat_enhance_m3_d1,u_feat_enhance_m4_d1):
u_feat_enhance_m3_d1 = F.normalize(u_feat_enhance_m3_d1, dim=-1)
u_feat_enhance_m4_d1 = F.normalize(u_feat_enhance_m4_d1, dim=-1)
logit_cl = torch.matmul(u_feat_enhance_m3_d1,u_feat_enhance_m4_d1.T)
# norm_num = 10 ** (len(str(int(logit_cl[0][0].item()))))
# logit_cl = logit_cl / norm_num
# print("logit_cl init:{}".format(logit_cl))
# logit_cl = F.normalize(logit_cl, p=1, dim=1)
logit_cl = torch.exp(logit_cl/0.07)#/0.07
# print("logit_cl:{}".format(logit_cl))
pos_logit = torch.diag(logit_cl)
neg_logit = logit_cl.sum(dim=1)
# print(neg_logit)
loss_cl = -torch.log(pos_logit/neg_logit)
return loss_cl.mean()
def cal_loss_cl(u_feat_enhance_m1_d1,u_feat_enhance_m1_d2):
user_m1_query = []
user_m1_query.append(u_feat_enhance_m1_d1)
# user_m1_key.append(u_feat_enhance_m1_d2)
u_feat_enhance_m1_d2_neg = u_feat_enhance_m1_d2.unsqueeze(-1).repeat(1,1,u_feat_enhance_m1_d2.shape[0])
for i in range(u_feat_enhance_m1_d2.shape[0]):
u_feat_enhance_m1_d2_neg[i,:,i] = 0
# user_m1_key.append(u_feat_enhance_m1_d2_neg)
# user_m1_key = torch.stack(user_m1_key,dim=-1).squeeze()
u_feat_enhance_m1_d2 = u_feat_enhance_m1_d2.unsqueeze(-1)
user_m1_key = torch.cat((u_feat_enhance_m1_d2,u_feat_enhance_m1_d2_neg),-1)
user_m1_query = torch.stack(user_m1_query,dim=-1).squeeze()
user_m1_query = torch.transpose(user_m1_query.unsqueeze(-1),1,2)
logits_m1 = torch.matmul(user_m1_query,user_m1_key).squeeze()
loss_cl = nn.CrossEntropyLoss()(logits_m1,labels)
return loss_cl
def sce_loss(x, y, alpha=3):
x = F.normalize(x, p=2, dim=-1)
y = F.normalize(y, p=2, dim=-1)
# loss = - (x * y).sum(dim=-1)
# loss = (x_h - y_h).norm(dim=1).pow(alpha)
loss = (1 - (x * y).sum(dim=-1)).pow_(alpha)
loss = loss.mean()
return loss
def concat_emb_listsV1(dspf_embs, dsad_embs):
pos_embs, neg_embs = list(),list()
for i in range(len(dspf_embs)):
for j in range(len(dspf_embs[i])):
pos_embs.append(dspf_embs[i][j].squeeze())
neg_embs.append(dsad_embs[i][0].squeeze())
for k in range(len(dspf_embs[i])):
if k != j:
pos_embs.append(dspf_embs[i][j].squeeze())
neg_embs.append(dspf_embs[i][k].squeeze())
pos_embs = torch.cat(pos_embs,0)
neg_embs = torch.cat(neg_embs,0)
label = torch.Tensor(torch.zeros(pos_embs.shape[0])).cuda()
return pos_embs, neg_embs, label
def concat_emb_listsV0(mf_embs_compare,ml_embs_compare):
mf_embs,ml_embs = list(),list()
for tmp in mf_embs_compare:
tmp = torch.cat(tmp,dim=1).squeeze()
mf_embs.append(tmp)
for tmp in ml_embs_compare:
tmp = torch.cat(tmp,dim=1).squeeze()
ml_embs.append(tmp)
mf_embs = torch.cat(mf_embs,dim=0)
ml_embs = torch.cat(ml_embs,dim=0)
label = torch.Tensor(torch.zeros(mf_embs.shape[0])).cuda()
return mf_embs, ml_embs, label
class ContrastiveLoss(torch.nn.Module):
"""
Contrastive loss function.
Based on:
"""
def __init__(self, margin=1.0):
super(ContrastiveLoss, self).__init__()
self.margin = margin
def check_type_forward(self, in_types):
assert len(in_types) == 3
x0_type, x1_type, y_type = in_types
assert x0_type.size() == x1_type.shape
assert x1_type.size()[0] == y_type.shape[0]
assert x1_type.size()[0] > 0
assert x0_type.dim() == 2
assert x1_type.dim() == 2
assert y_type.dim() == 1
def forward(self, x0, x1, y):
self.check_type_forward((x0, x1, y))
# euclidian distance
diff = x0 - x1
dist_sq = torch.sum(torch.pow(diff, 2), 1)
print(dist_sq)
dist = torch.sqrt(dist_sq)
mdist = self.margin - dist
dist = torch.clamp(mdist, min=0.0)
loss = y * dist_sq + (1 - y) * torch.pow(dist, 2)
loss = torch.sum(loss) / 2.0 / x0.size()[0]
return loss
def split_domain(predict,labels,domain_ids):
p_d1, l_d1, p_d2, l_d2, p_d3, l_d3 = list(), list(), list(), list(), list(), list()
for i in range(predict.shape[0]):
if domain_ids[i] == 0:
p_d1.append(predict[i].item())
l_d1.append(labels[i].item())
elif domain_ids[i] == 1:
p_d2.append(predict[i].item())
l_d2.append(labels[i].item())
elif domain_ids[i] == 2:
p_d3.append(predict[i].item())
l_d3.append(labels[i].item())
else:
print("error in domain id\n")
return p_d1, l_d1, p_d2, l_d2, p_d3, l_d3#np.array(p_d1), np.array(l_d1), np.array(p_d2), np.array(l_d2), np.array(p_d3), np.array(l_d3)
class AverageMeter(object):
def __init__(self, *keys: str):
self.totals = {key: 0.0 for key in keys}
self.counts = {key: 0 for key in keys}
def update(self, **kwargs: float) -> None:
for key, value in kwargs.items():
self._check_attr(key)
self.totals[key] += value
self.counts[key] += 1
def __getattr__(self, attr: str) -> float:
self._check_attr(attr)
total = self.totals[attr]
count = self.counts[attr]
return total / count if count else 0.0
def _check_attr(self, attr: str) -> None:
assert attr in self.totals and attr in self.counts
def init_logger(log_dir: str, log_file: str) -> None:
logger = logging.getLogger()
format_str = r'[%(asctime)s] %(message)s'
logging.basicConfig(
level=logging.INFO,
datefmt=r'%Y/%m/%d %H:%M:%S',
format=format_str
)
log_dir = Path(log_dir)
log_dir.mkdir(parents=True, exist_ok=True)
fh = logging.FileHandler(str(log_dir / log_file))
fh.setFormatter(logging.Formatter(format_str))
logger.addHandler(fh)
def get_sample_scores(pred_list):
pred_list = (-pred_list).argsort().argsort()[:, 0]
HIT_1, NDCG_1, MRR = get_metric(pred_list, 1)
HIT_5, NDCG_5, MRR = get_metric(pred_list, 5)
HIT_10, NDCG_10, MRR = get_metric(pred_list, 10)
return HIT_1, NDCG_1, HIT_5, NDCG_5, HIT_10, NDCG_10, MRR
def get_metric(pred_list, topk=10):
NDCG = 0.0
HIT = 0.0
MRR = 0.0
# [batch] the answer's rank
for rank in pred_list:
MRR += 1.0 / (rank + 1.0)
if rank < topk:
NDCG += 1.0 / np.log2(rank + 2.0)
HIT += 1.0
return HIT /len(pred_list), NDCG /len(pred_list), MRR /len(pred_list)