-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils.py
1097 lines (850 loc) · 39.4 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
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
import random
import torch
import os
import numpy as np
import re
import json
import scipy.sparse as ssp
import shutil
import pickle
import math
from collections import Counter
import pandas as pd
import rule_application as ra
def print_msg(msg):
msg = "## {} ##".format(msg)
length = len(msg)
msg = "\n{}\n".format(msg)
print(length * "#" + msg + length * "#")
def camel_to_normal(camel_string):
# 使用正则表达式将驼峰字符串转换为正常字符串
normal_string = re.sub(r'(?<!^)(?=[A-Z])', ' ', camel_string).lower()
return normal_string
def clean_symbol_in_rel(rel):
'''
clean symbol in relation
Args:
rel (str): relation name
'''
rel = rel.strip("_") # Remove heading
# Replace inv_ with inverse
# rel = rel.replace("inv_", "inverse ")
if "/" in rel:
if "inverse" in rel:
rel = rel.replace("inverse ", "")
rel = "inverse " + fb15k_rel_map[rel]
else:
rel = fb15k_rel_map[rel]
# WN-18RR
elif "_" in rel:
rel = rel.replace("_", " ") # Replace _ with space
# UMLS
elif "&" in rel:
rel = rel.replace("&", " ") # Replace & with space
# YAGO
else:
rel = camel_to_normal(rel)
return rel
def query(message, llm_model):
'''
Query ChatGPT API
:param message:·
:return:
'''
return llm_model.generate_sentence(message)
def unknown_check_prompt_length(prompt, condicate_list, return_rules, model):
'''Check whether the input prompt is too long. If it is too long, remove the first path and check again.'''
all_condicate = ";".join(condicate_list)
return_rules = return_rules.format(candidate_rels=all_condicate)
all_tokens = prompt + return_rules
maximun_token = model.maximum_token
if model.token_len(all_tokens) < maximun_token:
return all_condicate
else:
# Shuffle the paths
random.shuffle(condicate_list)
new_list_candcate = []
# check the length of the prompt
for p in condicate_list:
tmp_all_paths = ";".join(new_list_candcate + [p])
return_rules = return_rules.format(candidate_rels=tmp_all_paths)
tmp_all_tokens = prompt + return_rules
if model.token_len(tmp_all_tokens) > maximun_token:
return ";".join(new_list_candcate)
new_list_candcate.append(p)
def iteration_check_prompt_length(prompt, condicate_list, return_rules, model):
'''Check whether the input prompt is too long. If it is too long, remove the first path and check again.'''
all_condicate = ";".join(condicate_list)
return_rules = return_rules.format(candidate_rels=all_condicate)
all_tokens = prompt + return_rules
maximun_token = model.maximum_token
if model.token_len(all_tokens) < maximun_token:
return all_condicate
else:
# Shuffle the paths
random.shuffle(condicate_list)
new_list_candcate = []
# check the length of the prompt
for p in condicate_list:
tmp_all_paths = ";".join(new_list_candcate + [p])
return_rules = return_rules.format(candidate_rels=tmp_all_paths)
tmp_all_tokens = prompt + return_rules
if model.token_len(tmp_all_tokens) > maximun_token:
return ";".join(new_list_candcate)
new_list_candcate.append(p)
def check_prompt_length(prompt, list_of_paths, model):
'''Check whether the input prompt is too long. If it is too long, remove the first path and check again.'''
all_paths = "\n".join(list_of_paths)
all_tokens = prompt + all_paths
maximun_token = model.maximum_token
if model.token_len(all_tokens) < maximun_token:
return all_paths
else:
# Shuffle the paths
random.shuffle(list_of_paths)
new_list_of_paths = []
# check the length of the prompt
for p in list_of_paths:
tmp_all_paths = "\n".join(new_list_of_paths + [p])
tmp_all_tokens = prompt + tmp_all_paths
if model.token_len(tmp_all_tokens) > maximun_token:
return "\n".join(new_list_of_paths)
new_list_of_paths.append(p)
def num_tokens_from_message(path_string, model):
"""Returns the number of tokens used by a list of messages."""
messages = [{"role": "user", "content": path_string}]
try:
encoding = tiktoken.encoding_for_model(model)
except KeyError:
print("Warning: model not found. Using cl100k_base encoding.")
encoding = tiktoken.get_encoding("cl100k_base")
if model in ["gpt-3.5-turbo", 'gpt-3.5-turbo-16k']:
tokens_per_message = 4 # every message follows <|start|>{role/name}\n{content}<|end|>\n
elif model == "gpt-4":
tokens_per_message = 3
else:
raise NotImplementedError(f"num_tokens_from_messages() is not implemented for model {model}.")
num_tokens = 0
for message in messages:
num_tokens += tokens_per_message
for key, value in message.items():
num_tokens += len(encoding.encode(value))
num_tokens += 3 # every reply is primed with <|start|>assistant<|message|>
return num_tokens
def get_token_limit(model='gpt-4'):
"""Returns the token limitation of provided model"""
if model in ['gpt-4', 'gpt-4-0613']:
num_tokens_limit = 8192
elif model in ['gpt-3.5-turbo-16k', 'gpt-3.5-turbo-16k-0613']:
num_tokens_limit = 16384
elif model in ['gpt-3.5-turbo', 'gpt-3.5-turbo-0613', 'text-davinci-003', 'text-davinci-002']:
num_tokens_limit = 4096
else:
raise NotImplementedError(f"""get_token_limit() is not implemented for model {model}.""")
tokenizer = tiktoken.encoding_for_model(model)
return num_tokens_limit, tokenizer
def split_path_list(path_list, token_limit, model):
"""
Split the path list into several lists, each list can be fed into the model.
"""
output_list = []
current_list = []
current_token_count = 4
for path in path_list:
path += '\n'
path_token_count = num_tokens_from_message(path, model) - 4
if current_token_count + path_token_count > token_limit: # If the path makes the current list exceed the token limit
output_list.append(current_list)
current_list = [path] # Start a new list.
current_token_count = path_token_count + 4
else: # The new path fits into the current list without exceeding the limit
current_list.append(path) # Just add it there.
current_token_count += path_token_count
# Add the last list of tokens, if it's non-empty.
if current_list: # The last list not exceed the limit but no more paths
output_list.append(current_list)
return output_list
def shuffle_split_path_list(path_content_list, prompt_len, model):
"""
First shuffle the path_content list, then split the path list into a list of several lists
Each list can be directly fed into the model
"""
token_limitation, tokenizer = get_token_limit(model) # Get input token limitation for current model
token_limitation -= prompt_len + 4 # minus prompt length for path length
all_path_content = '\n'.join(path_content_list)
token_num_all_path = num_tokens_from_message(all_path_content, model)
random.shuffle(path_content_list)
if token_num_all_path > token_limitation:
list_of_paths = split_path_list(path_content_list, token_limitation, model)
else:
list_of_paths = [[path + '\n' for path in path_content_list]]
return list_of_paths
def ill_rank(pred, gt, ent2idx, q_h, q_t, q_r):
pred_ranks = np.argsort(pred)[::-1]
truth = gt[(q_h, q_r)]
truth = [t for t in truth if t != ent2idx[q_t]]
filtered_ranks = []
for i in range(len(pred_ranks)):
idx = pred_ranks[i]
if idx not in truth and pred[idx] > pred[ent2idx[q_t]]:
filtered_ranks.append(idx)
rank = len(filtered_ranks) + 1
return rank
def harsh_rank(pred, gt, ent2idx, q_h, q_t, q_r):
pred_ranks = np.argsort(pred)[::-1]
truth = gt[(q_h, q_r)]
truth = [t for t in truth]
filtered_ranks = []
for i in range(len(pred_ranks)):
idx = pred_ranks[i]
if idx not in truth and pred[idx] >= pred[ent2idx[q_t]]:
filtered_ranks.append(idx)
rank = len(filtered_ranks) + 1
return rank
def balance_rank(pred, gt, ent2idx, q_h, q_t, q_r):
if pred[ent2idx[q_t]] != 0:
pred_ranks = np.argsort(pred)[::-1]
truth = gt[(q_h, q_r)]
truth = [t for t in truth if t != ent2idx[q_t]]
filtered_ranks = []
for i in range(len(pred_ranks)):
idx = pred_ranks[i]
if idx not in truth:
filtered_ranks.append(idx)
rank = filtered_ranks.index(ent2idx[q_t]) + 1
else:
truth = gt[(q_h, q_r)]
filtered_pred = []
for i in range(len(pred)):
if i not in truth:
filtered_pred.append(pred[i])
n_non_zero = np.count_nonzero(filtered_pred)
rank = n_non_zero + 1
return rank
def random_rank(pred, gt, ent2idx, q_h, q_t, q_r):
pred_ranks = np.argsort(pred)[::-1]
truth = gt[(q_h, q_r)]
truth = [t for t in truth if t != ent2idx[q_t]]
truth.append(ent2idx[q_t])
filtered_ranks = []
for i in range(len(pred_ranks)):
idx = pred_ranks[i]
if idx not in truth and pred[idx] >= pred[ent2idx[q_t]]:
if (pred[idx] == pred[ent2idx[q_t]]) and (np.random.uniform() < 0.5):
filtered_ranks.append(idx)
else:
filtered_ranks.append(idx)
rank = len(filtered_ranks) + 1
return rank
def load_json_data(file_path, default=None):
"""从文件加载JSON数据,如果文件不存在则返回默认值。"""
try:
if os.path.exists(file_path):
print(f"Use cache from: {file_path}")
with open(file_path, 'r') as file:
return json.load(file)
else:
print(f"File not found: {file_path}")
# 在这里添加你想要执行的操作,比如创建一个空的JSON对象并返回
return default
except Exception as e:
print(f"Error loading JSON data from {file_path}: {e}")
return default
def save_json_data(data, file_path):
"""将数据保存到JSON文件。"""
try:
with open(file_path, 'w', encoding='utf-8') as file:
json.dump(data, file, indent=4, ensure_ascii=False)
print(f"Data has been converted to JSON and saved to {file_path}")
except Exception as e:
print(f"Error saving JSON data to {file_path}: {e}")
def write_to_file(content, path):
with open(path, "w", encoding="utf-8") as fout:
fout.write(content)
def stat_ranks(rank_list, method='filter'):
hits = [1, 3, 10]
total_rank = torch.cat(rank_list)
mrr = torch.mean(1.0 / total_rank.float())
print("MRR ({}): {:.6f}".format(method, mrr.item()))
for hit in hits:
avg_count = torch.mean((total_rank <= hit).float())
print("Hits ({}) @ {}: {:.6f}".format(method, hit, avg_count.item()))
return mrr
def construct_adjacency_list_and_index(triples, relation_id_list, num_entities):
"""构造邻接矩阵列表"""
adj_list = []
relation_index = {}
triples = np.array(triples)
for i, relation_id in enumerate(relation_id_list):
idx = np.argwhere(triples[:, 1] == relation_id)
adj_list.append(ssp.csc_matrix((np.ones(len(idx), dtype=np.uint8),
(triples[:, 0][idx].squeeze(1), triples[:, 2][idx].squeeze(1))),
shape=(num_entities, num_entities)))
relation_index[relation_id] = i
return adj_list, relation_index
def incidence_matrix(adj_list):
'''
adj_list: List of sparse adjacency matrices
'''
rows, cols, dats = [], [], []
dim = adj_list[0].shape
for adj in adj_list:
adjcoo = adj.tocoo()
rows += adjcoo.row.tolist()
cols += adjcoo.col.tolist()
dats += adjcoo.data.tolist()
row = np.array(rows)
col = np.array(cols)
data = np.array(dats)
return ssp.csc_matrix((data, (row, col)), shape=dim)
def _sp_row_vec_from_idx_list(idx_list, dim):
"""Create sparse vector of dimensionality dim from a list of indices."""
shape = (1, dim)
data = np.ones(len(idx_list))
row_ind = np.zeros(len(idx_list))
col_ind = list(idx_list)
return ssp.csr_matrix((data, (row_ind, col_ind)), shape=shape)
def _get_neighbors(adj, nodes):
"""Takes a set of nodes and a graph adjacency matrix and returns a set of neighbors.
Directly copied from dgl.contrib.data.knowledge_graph"""
sp_nodes = _sp_row_vec_from_idx_list(list(nodes), adj.shape[1])
sp_neighbors = sp_nodes.dot(adj)
neighbors = set(ssp.find(sp_neighbors)[1]) # convert to set of indices
return neighbors
def _bfs_relational(adj, roots, max_nodes_per_hop=None):
"""
BFS for graphs.
Modified from dgl.contrib.data.knowledge_graph to accomodate node sampling
"""
visited = set()
current_lvl = set(roots)
next_lvl = set()
while current_lvl:
for v in current_lvl:
visited.add(v)
next_lvl = _get_neighbors(adj, current_lvl)
next_lvl -= visited # set difference
if max_nodes_per_hop and max_nodes_per_hop < len(next_lvl):
next_lvl = set(random.sample(next_lvl, max_nodes_per_hop))
yield next_lvl
current_lvl = set.union(next_lvl)
def extract_neighbors(adj, roots, h=1, max_nodes_per_hop=None):
bfs_generator = _bfs_relational(adj, roots, max_nodes_per_hop)
lvls = list()
for _ in range(h):
try:
lvls.append(next(bfs_generator))
except StopIteration:
pass
return set().union(*lvls)
def get_subgraph_nodes(root1_nei, root2_nei, ind, kind):
# 根据'kind'获取子图节点
subgraph_nei_nodes_int = root1_nei.intersection(root2_nei)
if ind[0] in subgraph_nei_nodes_int:
subgraph_nei_nodes_int.remove(ind[0])
if ind[1] in subgraph_nei_nodes_int:
subgraph_nei_nodes_int.remove(ind[1])
subgraph_nei_nodes_un = root1_nei.union(root2_nei)
if ind[0] in subgraph_nei_nodes_un:
subgraph_nei_nodes_un.remove(ind[0])
if ind[1] in subgraph_nei_nodes_un:
subgraph_nei_nodes_un.remove(ind[1])
if kind == "intersection":
subgraph_nodes = set(list(ind) + list(subgraph_nei_nodes_int))
else:
subgraph_nodes = set(list(ind) + list(subgraph_nei_nodes_un))
return list(subgraph_nodes)
def subgraph_extraction_labeling(ind, A_list, kind, h=1, max_nodes_per_hop=None):
A_incidence = incidence_matrix(A_list)
A_incidence += A_incidence.T
root1_nei = extract_neighbors(A_incidence, set([ind[0]]), h, max_nodes_per_hop)
root2_nei = extract_neighbors(A_incidence, set([ind[1]]), h, max_nodes_per_hop)
temp_entity_id = {}
subgraph_nodes = get_subgraph_nodes(root1_nei, root2_nei, ind, kind)
for idx, entity in enumerate(subgraph_nodes):
temp_entity_id[idx] = entity
subject_object_list = []
subgraph = [adj[subgraph_nodes, :][:, subgraph_nodes] for adj in A_list]
nonzero_row_indices, nonzero_col_indices = incidence_matrix(subgraph).nonzero()
for idx in range(len(nonzero_row_indices)):
subject = temp_entity_id[nonzero_row_indices[idx]]
object = temp_entity_id[nonzero_col_indices[idx]]
subject_object_list.append([subject, object])
return subject_object_list
def copy_folder_contents(source_folder, destination_folder):
# 创建目标文件夹,如果它不存在
if not os.path.exists(destination_folder):
os.makedirs(destination_folder)
# 复制每个文件和子文件夹
for item in os.listdir(source_folder):
source_item = os.path.join(source_folder, item)
destination_item = os.path.join(destination_folder, item)
if os.path.isdir(source_item):
shutil.copytree(source_item, destination_item)
else:
shutil.copy2(source_item, destination_item)
print(f"Contents of '{source_folder}' have been copied to '{destination_folder}'")
def filter_candidates(test_query, candidates, test_data):
"""
Filter out those candidates that are also answers to the test query
but not the correct answer.
Parameters:
test_query (np.ndarray): test_query
candidates (dict): answer candidates with corresponding confidence scores
test_data (np.ndarray): test dataset
Returns:
candidates (dict): filtered candidates
"""
other_answers = test_data[
(test_data[:, 0] == test_query[0])
* (test_data[:, 1] == test_query[1])
* (test_data[:, 2] != test_query[2])
* (test_data[:, 3] == test_query[3])
]
if len(other_answers):
objects = other_answers[:, 2]
for obj in objects:
candidates.pop(obj, None)
return candidates
def calculate_rank(test_query_answer, candidates, num_entities, setting="best"):
"""
Calculate the rank of the correct answer for a test query.
Depending on the setting, the average/best/worst rank is taken if there
are several candidates with the same confidence score.
Parameters:
test_query_answer (int): test query answer
candidates (dict): answer candidates with corresponding confidence scores
num_entities (int): number of entities in the dataset
setting (str): "average", "best", or "worst"
Returns:
rank (int): rank of the correct answer
"""
rank = num_entities
if test_query_answer in candidates:
conf = candidates[test_query_answer]
all_confs = list(candidates.values())
all_confs = sorted(all_confs, reverse=True)
ranks = [idx for idx, x in enumerate(all_confs) if x == conf]
try:
if setting == "average":
rank = (ranks[0] + ranks[-1]) // 2 + 1
elif setting == "best":
rank = ranks[0] + 1
elif setting == "worst":
rank = ranks[-1] + 1
except Exception as e:
ranks
return rank
def get_top_k_with_index(similarity_file_path, top_k):
with open(similarity_file_path, 'rb') as f:
loaded_arr = pickle.load(f)
# 获取每一行的最大 top_k 个值的索引
top_k_indices = np.argsort(loaded_arr, axis=1)[:, -top_k:][:, ::-1]
# 利用高级索引获取对应的值
rows = np.arange(loaded_arr.shape[0])[:, None] # 生成行索引
top_k_values = loaded_arr[rows, top_k_indices]
# 将结果以字典形式存储
result_dict = {}
for i in range(loaded_arr.shape[0]):
result_dict[i] = {index: value for index, value in zip(top_k_indices[i], top_k_values[i])}
return result_dict
def get_candicates_by_timestamp(test_query, bkg, interval):
timestamp_id = test_query[3]
min_timestamp_id = timestamp_id - interval
mask = (bkg[:, 3] >= min_timestamp_id) * (bkg[:, 3] < timestamp_id)
# 先筛选出时间戳符合条件的条目
time_filtered = bkg[mask]
candidates = select_canicates_based_timestamp(time_filtered, timestamp_id)
return candidates
def select_canicates_based_timestamp_normal(time_filtered, target_timestamp_id, min_score, max_score):
# 按照第四列(时间戳)降序排列
time_filtered_sorted = time_filtered[time_filtered[:, 3].argsort()[::-1]]
# 获取每个目标实体的第一个(最大)时间戳的索引
unique_targets, indices = np.unique(time_filtered_sorted[:, 2], return_index=True)
max_timestamps = time_filtered_sorted[indices, 3]
array = 1.0 / (target_timestamp_id - max_timestamps)
a = min_score
b = max_score
# 找到数组的最小值和最大值
min_val = min(array)
max_val = max(array)
# 使用 Min-Max 归一化将数组归一化到 [a, b] 区间
normalized_array = [
a + (x - min_val) * (b - a) / (max_val - min_val) if max_val != min_val else a
for x in array
]
# 创建一个字典,将每个目标实体映射到其最大时间戳
candidates = dict(zip(unique_targets, normalized_array))
return candidates
def select_canicates_based_timestamp(time_filtered, target_timestamp_id):
# 按照第四列(时间戳)降序排列
time_filtered_sorted = time_filtered[time_filtered[:, 3].argsort()[::-1]]
# 获取每个目标实体的第一个(最大)时间戳的索引
unique_targets, indices = np.unique(time_filtered_sorted[:, 2], return_index=True)
max_timestamps = time_filtered_sorted[indices, 3]
# 创建一个字典,将每个目标实体映射到其最大时间戳
candidates = dict(zip(unique_targets, 1.0 / (target_timestamp_id - max_timestamps)))
return candidates
def get_candicates_by_source_with_timestamp(test_query, bkg, interval):
source_id = test_query[0]
timestamp_id = test_query[3]
min_timestamp_id = timestamp_id - interval
mask = (bkg[:, 3] >= min_timestamp_id) * (bkg[:, 3] < timestamp_id)
# 先筛选出时间戳符合条件的条目
time_filtered = bkg[mask]
# 进行源实体的匹配
target_mask = time_filtered[:, 0] == source_id
source_mask = time_filtered[:, 2] == source_id
candicates_target = time_filtered[target_mask][:, 2]
candicates_source = time_filtered[source_mask][:, 0]
# 合并候选实体数组
candicates = np.hstack((candicates_target, candicates_source))
if len(candicates) == 0:
candicates = select_canicates_based_timestamp(time_filtered, timestamp_id)
else:
counter = Counter(candicates)
candicates = dict(counter)
return candicates
def expand_candidates(candidates, data, interval, target_timestamp_id):
min_timestamp_id = target_timestamp_id - interval
train_n = np.array(data.train_idx)
valid_n = np.array(data.valid_idx)
analysis_bkg = np.vstack((valid_n, train_n))
mask = analysis_bkg[:, 3] >= min_timestamp_id
temp_dict = {}
for bkg_target_id, bkg_timestamp_id in analysis_bkg[mask][:, 2:4]:
if bkg_target_id in candidates:
continue
if bkg_target_id not in temp_dict:
temp_dict[bkg_target_id] = bkg_timestamp_id
curr_timestamp_id = temp_dict[bkg_target_id]
if curr_timestamp_id < bkg_timestamp_id:
temp_dict[bkg_target_id] = bkg_timestamp_id
pro = 0.0
for value in temp_dict.values():
pro = pro + value
temp_temp_dict = {}
for key, value in temp_dict.items():
temp_temp_dict[key] = value / pro
if len(temp_temp_dict.values()) == 0:
return candidates
X_min = min(list(temp_temp_dict.values()))
X_max = max(list(temp_temp_dict.values()))
if X_max == X_min:
return candidates
b = max(list(candidates.values()))
a = min(list(candidates.values()))
temp_3_dict = {}
for key, value in temp_temp_dict.items():
# temp_3_dict[key] = (a + (b - a) * (value - X_min) / (X_max - X_min))
temp_3_dict[key] = 0.2*(a + (b - a) * (math.log(value) - math.log(X_min)) / (math.log(X_max) - math.log(X_min)))
merged_dict = {**candidates, **temp_3_dict}
return merged_dict
def expand_candidates_auto(candidates, bkg, interval, test_query):
is_exist = True
if len(candidates) == 0:
is_exist = False
timestamp_id = test_query[3]
min_timestamp_id = timestamp_id - interval
mask = (bkg[:, 3] >= min_timestamp_id) * (bkg[:, 3] < timestamp_id)
# 先筛选出时间戳符合条件的条目
time_filtered = bkg[mask]
candidates_with_max_timestamp = select_canicates_based_timestamp(time_filtered, timestamp_id)
exist_candidates_set = set(list(candidates.keys()))
added_candidates_set = set(list(candidates_with_max_timestamp.keys()))
share_candidates = exist_candidates_set.intersection(added_candidates_set)
if not share_candidates:
benchmark_rate = 1.0
else:
benchmark_rate = min(candidates[share] / candidates_with_max_timestamp[share] for share in share_candidates)
candidates_with_max_timestamp = {key: value * benchmark_rate for key, value in candidates_with_max_timestamp.items()}
merge_dict = {**candidates_with_max_timestamp, **candidates}
return merge_dict, is_exist
def expand_candidates_with_freq_weight(candidates, bkg, interval, test_query, freq_weight):
if len(candidates) == 1:
return candidates, True
source_id = test_query[0]
target_timestamp_id = test_query[3]
if interval == 0:
time_mask = bkg[:, 3] < target_timestamp_id
else:
min_timestamp_id = target_timestamp_id - interval
time_mask = (bkg[:, 3] >= min_timestamp_id) * (bkg[:, 3] < target_timestamp_id)
target_mask = bkg[:, 0] == source_id
source_mask = bkg[:, 2] == source_id
combined_mask = target_mask | source_mask
mask = combined_mask * time_mask
subgraph = bkg[mask]
if len(subgraph) != 0:
if len(candidates) == 0:
min_score = 0
max_score = 1
else:
min_score = min(list(candidates.values()))
max_score = max(list(candidates.values()))
candidates_with_max_timestamp_id = select_canicates_based_timestamp_normal(subgraph, target_timestamp_id,
min_score,
max_score)
else:
if len(candidates) == 0:
mask = bkg[:, 3] < target_timestamp_id
time_filtered = bkg[mask]
candidates_with_max_timestamp_id = select_canicates_based_timestamp(time_filtered, target_timestamp_id)
return candidates_with_max_timestamp_id, False
else:
return candidates, True
# 获取两个字典的全部键集合
all_keys = set(candidates_with_max_timestamp_id.keys()).union(set(candidates.keys()))
# 使用字典的 get 方法设置默认值为 0
merge_dict = {
key: (1 - freq_weight) * candidates_with_max_timestamp_id.get(key, 0) + freq_weight * candidates.get(key, 0)
for key in all_keys
}
return merge_dict, True
def expand_candidates_auto_with_freq_weight(candidates, bkg, interval, test_query, freq_weight):
is_exist = True
if len(candidates) == 0:
is_exist = False
timestamp_id = test_query[3]
min_timestamp_id = timestamp_id - interval
mask = (bkg[:, 3] >= min_timestamp_id) * (bkg[:, 3] < timestamp_id)
# 先筛选出时间戳符合条件的条目
time_filtered = bkg[mask]
candidates_with_max_timestamp = select_canicates_based_timestamp(time_filtered, timestamp_id)
exist_candidates_set = set(list(candidates.keys()))
added_candidates_set = set(list(candidates_with_max_timestamp.keys()))
share_candidates = exist_candidates_set.intersection(added_candidates_set)
if not share_candidates:
benchmark_rate = 1.0
candidates_with_max_timestamp = {key: value * benchmark_rate for key, value in
candidates_with_max_timestamp.items()}
merge_dict = candidates_with_max_timestamp
else:
merge_dict = {**candidates_with_max_timestamp, **candidates}
for cand_id in share_candidates:
# merge_dict[cand_id] = (1 - freq_weight) * candidates[cand_id] + freq_weight * candidates_with_max_timestamp[
# cand_id]
merge_dict[cand_id] = candidates[cand_id] + candidates_with_max_timestamp[
cand_id]
return merge_dict, is_exist
def expand_candidates_with_source(candidates, bkg, interval, test_query, freq_weight):
exist_in_neighbors = True
is_has_neighbors = True
is_exist = True
if len(candidates) == 0:
is_exist = False
timestamp_id = test_query[3]
source_id = test_query[0]
target_id = test_query[2]
min_timestamp_id = timestamp_id - interval
mask = (bkg[:, 3] >= min_timestamp_id) * (bkg[:, 3] < timestamp_id)
# 先筛选出时间戳符合条件的条目
time_filtered = bkg[mask]
# 进行源实体的匹配
target_mask = time_filtered[:, 0] == source_id
source_mask = time_filtered[:, 2] == source_id
candicates_target_with_timestamp = time_filtered[target_mask][:, [2,3]]
candicates_source_with_timestamp = time_filtered[source_mask][:, [0,3]]
# 合并候选实体数组
candicates_with_neighbor = np.hstack((candicates_target_with_timestamp[:,0], candicates_source_with_timestamp[:,0]))
if len(candicates_with_neighbor) == 0:
result_dict = select_canicates_based_timestamp(time_filtered, timestamp_id)
is_has_neighbors = False
exist_in_neighbors = False
else:
unique_neighbors = np.unique(candicates_with_neighbor)
if target_id not in unique_neighbors:
exist_in_neighbors = False
exist_candidates_set = set(list(candidates.keys()))
added_candidates_set = set(unique_neighbors.tolist())
share_candidates = exist_candidates_set.intersection(added_candidates_set)
# 创建一个包含初始数据的DataFrame
data = np.vstack((candicates_target_with_timestamp,candicates_source_with_timestamp))
df = pd.DataFrame(data, columns=['neighbor', 'timestamp'])
# 计算最大值10000 - 第二列的数值 的倒数
df['timestamp'] = freq_weight * (1 / (timestamp_id - df['timestamp']))
# 使用groupby根据第一列进行分组,并找到每个分组中第二列的最大值
result = df.loc[df.groupby('neighbor')['timestamp'].idxmax()].reset_index(drop=True)
result_dict = result.set_index('neighbor')['timestamp'].to_dict()
# if len(share_candidates) == 0:
# benchmark_rate = 1.0
# else:
# benchmark_rate = max(candidates[share] / result_dict[share] for share in share_candidates)
# benchmark_rate = sum(candidates[share] / result_dict[share] for share in share_candidates)/len(share_candidates)
#
# result_dict = {key: value * benchmark_rate for key, value in result_dict.items()}
# merge_dict = {k: freq_weight * result_dict.get(k, 0) + candidates.get(k, 0) for k in set(result_dict) | set(candidates)}
merge_dict = {**result_dict, **candidates}
# merge_dict = {**result_dict}
return merge_dict, is_exist, is_has_neighbors, exist_in_neighbors
def expand_candidates_with_relation(candidates, bkg, interval, test_query, freq_weight):
exist_in_neighbors = True
is_has_neighbors = True
is_exist = True
if len(candidates) == 0:
is_exist = False
timestamp_id = test_query[3]
source_id = test_query[0]
relation_id = test_query[1]
target_id = test_query[2]
min_timestamp_id = timestamp_id - interval
mask = (bkg[:, 3] >= min_timestamp_id) * (bkg[:, 3] < timestamp_id)
# 先筛选出时间戳符合条件的条目
time_filtered = bkg[mask]
# 进行源实体的匹配
target_mask = time_filtered[:, 1] == relation_id
source_mask = time_filtered[:, 1] == relation_id
candicates_target_with_timestamp = time_filtered[target_mask][:, [2,3]]
candicates_source_with_timestamp = time_filtered[source_mask][:, [0,3]]
# 合并候选实体数组
candicates_with_neighbor = np.hstack((candicates_target_with_timestamp[:,0], candicates_source_with_timestamp[:,0]))
if len(candicates_with_neighbor) == 0:
result_dict = select_canicates_based_timestamp(time_filtered, timestamp_id)
is_has_neighbors = False
exist_in_neighbors = False
else:
unique_neighbors = np.unique(candicates_with_neighbor)
if target_id not in unique_neighbors:
exist_in_neighbors = False
# 创建一个包含初始数据的DataFrame
data = np.vstack((candicates_target_with_timestamp,candicates_source_with_timestamp))
df = pd.DataFrame(data, columns=['neighbor', 'timestamp'])
# 计算最大值10000 - 第二列的数值 的倒数
df['timestamp'] = freq_weight * (1 / (timestamp_id - df['timestamp']))
# 使用groupby根据第一列进行分组,并找到每个分组中第二列的最大值
result = df.loc[df.groupby('neighbor')['timestamp'].idxmax()].reset_index(drop=True)
result_dict = result.set_index('neighbor')['timestamp'].to_dict()
# merge_dict = {k: freq_weight * result_dict.get(k, 0) + candidates.get(k, 0) for k in set(result_dict) | set(candidates)}
merge_dict = {**result_dict, **candidates}
# merge_dict = {**result_dict}
return merge_dict, is_exist, is_has_neighbors, exist_in_neighbors
def remove_candidates(candidates, data, interval, target_timestamp_id):
min_timestamp_id = target_timestamp_id - interval
train_n = np.array(data.train_idx)
valid_n = np.array(data.valid_idx)
analysis_bkg = np.vstack((valid_n, train_n))
mask = analysis_bkg[:, 3] >= min_timestamp_id
candidates_id = analysis_bkg[mask][:, 2]
temp_dict = candidates.copy()
for key in candidates.keys():
if key not in candidates_id:
del temp_dict[key]
return temp_dict
def data_analysis(test_query, analysis_bkg_all):
# Source Entity Existence Analysis
source_id = test_query[0]
# Use boolean indexing for a quick check without creating a new array
source_exists_in_bkg = np.any(analysis_bkg_all[:, 0] == source_id)
target_exists_in_bkg = np.any(analysis_bkg_all[:, 2] == source_id)
# If source entity does not exist in either train or validation set
if not (source_exists_in_bkg or target_exists_in_bkg):
return 1
return 0
def get_win_subgraph(test_data, data, learn_edges, window, win_start=0):
unique_timestamp_id = np.unique(test_data[:,3])
win_subgraph = {}
for timestamp_id in unique_timestamp_id:
subgraph = ra.get_window_edges(data.all_idx, timestamp_id - win_start, learn_edges, window)
win_subgraph[timestamp_id] = subgraph
return win_subgraph
def calculate_hours_between_dates_pandas(start_dates, end_dates):
"""
使用 pandas 计算多对日期之间的小时差。
参数:
start_dates (list of str): 开始日期,格式为 "YYYY-MM-DD" 的字符串列表。
end_dates (list of str): 结束日期,格式为 "YYYY-MM-DD" 的字符串列表。
返回:
list of float: 日期对之间的小时差列表。
"""
# 将列表转换为 pandas Series
start_series = pd.to_datetime(start_dates)
end_series = pd.to_datetime(end_dates)
# 计算小时差
difference = (end_series - start_series).astype('timedelta64[h]').astype(int)
return difference.tolist()
def merge_scores_optimized(dict_A, dict_B, model_weight):
"""
Normalize the scores in two dictionaries to a 0-1 range and merge the scores for all keys.
Args:
dict_A (dict): Scores from the first dictionary.
dict_B (dict): Scores from the second dictionary.
Returns:
dict: Merged scores dictionary for all keys from both dictionaries.
"""
# Normalize the scores in both dictionaries
normalized_A = normalize_scores(dict_A)
normalized_B = normalize_scores(dict_B)