-
Notifications
You must be signed in to change notification settings - Fork 21
/
features_generation.py
1052 lines (866 loc) · 51.9 KB
/
features_generation.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
# -*- coding：utf-8 -*-
# Author: juzstu
# Time: 2021/8/7 18:19
import pandas as pd
import numpy as np
from tqdm import tqdm
import os
from sklearn.feature_extraction.text import TfidfVectorizer, CountVectorizer
from sklearn.decomposition import TruncatedSVD
from collections import defaultdict
import warnings
import faiss
import pickle
from gensim.models import Word2Vec
import multiprocessing
warnings.filterwarnings('ignore')
def reduce_mem_usage(df, verbose=True):
numerics = ['int16', 'int32', 'int64', 'float16', 'float32', 'float64']
start_mem = df.memory_usage().sum() / 1024 ** 2
for col in df.columns:
col_type = df[col].dtypes
if col_type in numerics:
c_min = df[col].min()
c_max = df[col].max()
if str(col_type)[:3] == 'int':
if c_min > np.iinfo(np.int8).min and c_max < np.iinfo(np.int8).max:
df[col] = df[col].astype(np.int8)
elif c_min > np.iinfo(np.int16).min and c_max < np.iinfo(np.int16).max:
df[col] = df[col].astype(np.int16)
elif c_min > np.iinfo(np.int32).min and c_max < np.iinfo(np.int32).max:
df[col] = df[col].astype(np.int32)
elif c_min > np.iinfo(np.int64).min and c_max < np.iinfo(np.int64).max:
df[col] = df[col].astype(np.int64)
else:
if c_min > np.finfo(np.float16).min and c_max < np.finfo(np.float16).max:
df[col] = df[col].astype(np.float16)
elif c_min > np.finfo(np.float32).min and c_max < np.finfo(np.float32).max:
df[col] = df[col].astype(np.float32)
else:
df[col] = df[col].astype(np.float64)
end_mem = df.memory_usage().sum() / 1024 ** 2
if verbose:
print('Mem. usage decreased to {:5.2f} Mb ({:.1f}% reduction)'.format(
end_mem, 100 * (start_mem - end_mem) / start_mem))
return df
# 全局统计
def cnt_stat(df, group_cols, target_col, use_cnt=True, use_nunique=True):
if isinstance(group_cols, list):
col_name = '_'.join(group_cols)
else:
col_name = 'global_' + group_cols
if use_cnt:
df[f'{col_name}_cnt'] = df.groupby(group_cols)[target_col].transform('count')
if use_nunique:
df[f'{col_name}_dcnt'] = df.groupby(group_cols)[target_col].transform('nunique')
return df
def count2vec(input_values, output_num, output_prefix, seed=1024):
count_enc = CountVectorizer(lowercase=True, ngram_range=(1, 1))
count_vec = count_enc.fit_transform(input_values)
svd_tmp = TruncatedSVD(n_components=output_num, n_iter=10, random_state=seed).fit_transform(count_vec)
svd_tmp = pd.DataFrame(svd_tmp)
svd_tmp.columns = ['svd_cntvec_{}_{}'.format(output_prefix, i) for i in range(output_num)]
return svd_tmp
def tf_idf(input_values, output_num, output_prefix, seed=1024):
tfidf_enc = TfidfVectorizer(lowercase=True, ngram_range=(1, 1))
tfidf_vec = tfidf_enc.fit_transform(input_values)
svd_tmp = TruncatedSVD(n_components=output_num, n_iter=10, random_state=seed).fit_transform(tfidf_vec)
svd_tmp = pd.DataFrame(svd_tmp)
svd_tmp.columns = ['svd_tfidf_{}_{}'.format(output_prefix, i) for i in range(output_num)]
return svd_tmp
def gen_svd_df(df, group_target, num=5, seed=1024):
tfidf_tmp = tf_idf(df[group_target], num, group_target, seed)
count_tmp = count2vec(df[group_target], num, group_target, seed)
df = pd.concat([df, tfidf_tmp, count_tmp], axis=1)
del df[group_target]
return df
def list_tm(df, df_, col, cnt_, read_, comment_, like_, click_, forward_, follow_, favorite_, use_hist=False):
df_ = df_.reset_index(drop=True)
if use_hist:
cnt_dict = cnt_
read_dict = read_
comment_dict = comment_
like_dict = like_
click_dict = click_
forward_dict = forward_
follow_dict = follow_
favorite_dict = favorite_
else:
cnt_dict = defaultdict(int)
read_dict = defaultdict(int)
comment_dict = defaultdict(int)
like_dict = defaultdict(int)
click_dict = defaultdict(int)
forward_dict = defaultdict(int)
follow_dict = defaultdict(int)
favorite_dict = defaultdict(int)
list_ids = [col, 'read_comment', 'comment', 'like', 'click_avatar', 'forward', 'follow', 'favorite']
for (i, read_comment, comment, like, click_avatar, forward, follow, favorite) in df[list_ids].values:
if ';' in i:
user, ids = i.split('|')
for j in ids.split(';'):
cnt_dict[f'{user}|{j}'] += 1
read_dict[f'{user}|{j}'] += read_comment
comment_dict[f'{user}|{j}'] += comment
like_dict[f'{user}|{j}'] += like
click_dict[f'{user}|{j}'] += click_avatar
forward_dict[f'{user}|{j}'] += forward
follow_dict[f'{user}|{j}'] += follow
favorite_dict[f'{user}|{j}'] += favorite
read_dict_ = {k: v / cnt_dict[k] for k, v in read_dict.items()}
comment_dict_ = {k: v / cnt_dict[k] for k, v in comment_dict.items()}
like_dict_ = {k: v / cnt_dict[k] for k, v in like_dict.items()}
click_dict_ = {k: v / cnt_dict[k] for k, v in click_dict.items()}
forward_dict_ = {k: v / cnt_dict[k] for k, v in forward_dict.items()}
follow_dict_ = {k: v / cnt_dict[k] for k, v in follow_dict.items()}
favorite_dict_ = {k: v / cnt_dict[k] for k, v in favorite_dict.items()}
tmp_list = []
for f, d, c in df_[['feedid', 'date_', col]].values:
user, ids = c.split('|')
if ';' in ids:
for j in ids.split(';'):
tmp_list.append([user, f, d, f'{user}|{j}'])
else:
tmp_list.append([user, f, d, np.nan])
tmp_list = pd.DataFrame(tmp_list, columns=['userid', 'feedid', 'date_', col])
for c in zip(list_ids[1:], [read_dict_, comment_dict_, like_dict_, click_dict_, forward_dict_, follow_dict_, favorite_dict_]):
tmp_list[c[0]] = tmp_list[col].map(c[1])
del tmp_list[col]
stat_cols = ['min', 'max', 'mean', 'std', 'count']
tmp_list = tmp_list.groupby(['userid', 'feedid', 'date_'])[list_ids[1:]].agg(stat_cols).reset_index()
tmp_list.columns = ['userid', 'feedid', 'date_'] + [f'{col}_{i}_{j}' for i in list_ids[1:] for j in stat_cols]
for i in list_ids[1:]:
tmp_list[f'{col}_{i}_std'] = np.sqrt(
tmp_list[f'{col}_{i}_std'] ** 2 * (tmp_list[f'{col}_{i}_count'] - 1) / tmp_list[f'{col}_{i}_count'])
del tmp_list[f'{col}_{i}_count']
return tmp_list, (cnt_dict, read_dict, comment_dict, like_dict, click_dict, forward_dict, follow_dict, favorite_dict)
def save_obj(obj, path):
with open(path, 'wb') as f:
pickle.dump(obj, f, pickle.HIGHEST_PROTOCOL)
def load_obj(path):
with open(path, 'rb') as f:
return pickle.load(f)
# 多个tag, keyword的点击率统计特征
def gen_tag_mean(train_path, feed_info_path, sava_path):
user_action = pd.read_feather(train_path)
m_list = ['machine_keyword_list', 'manual_tag_list']
feed_info = pd.read_csv(feed_info_path, usecols=['feedid'] + m_list)
user_action = user_action.merge(feed_info, on='feedid', how='left')
for d in m_list:
user_action[d] = user_action['userid'].astype(str) + '|' + user_action[d].astype(str)
del feed_info
for m in m_list:
print(f'start deal {m} ...')
cnt_u, read_u, comment_u, like_u, click_u, forward_u, follow_u, favorite_u = None, None, None, None, None, None, None, None
for d in range(2, 15):
print(f'date of {d}')
prev_tmp = user_action[user_action['date_'] == d - 1]
now_tmp = user_action[user_action['date_'] == d]
flg = True if d > 2 else False
now_tmp, all_dict = list_tm(prev_tmp, now_tmp, m, cnt_u, read_u, comment_u, like_u, click_u, forward_u, follow_u, favorite_u, use_hist=flg)
if d == 14:
save_obj(all_dict, f'{sava_path}/{m}_user_ctr_dict.pkl')
cnt_u, read_u, comment_u, like_u, click_u, forward_u, follow_u, favorite_u = all_dict
print(f'the shape of df_{m}_{d}:{now_tmp.shape[0]}')
now_tmp['userid'] = now_tmp['userid'].astype(int)
now_tmp.to_feather(f'{sava_path}/{m}_user_ctr_{d}.feather')
# 常规全局统计特征
def gen_base_feats(train_path, feed_info_path, train_save_path, base_hist_path):
user_action = pd.read_feather(train_path)
r1_test_a = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data1/test_a.csv', usecols=['userid', 'feedid'])
r1_test_b = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data1/test_b.csv', usecols=['userid', 'feedid'])
r2_test_a = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data2/test_a.csv', usecols=['userid', 'feedid'])
test = pd.concat([r1_test_a, r1_test_b, r2_test_a], axis=0, ignore_index=True)
del r1_test_a, r1_test_b, r2_test_a
test['date_'] = 15
for i in ['read_comment', 'comment', 'like', 'click_avatar', 'forward', 'follow', 'favorite']:
test[i] = -1
user_action = user_action[test.columns].append(test)
del test
feed_info = pd.read_csv(feed_info_path, usecols=['feedid', 'authorid', 'videoplayseconds'])
feed_info['vs_ratio'] = feed_info['videoplayseconds'] / feed_info.groupby('authorid')['videoplayseconds'].transform(
'mean')
user_action = user_action.merge(feed_info, on='feedid', how='left')
del feed_info
user_action = cnt_stat(user_action, 'feedid', 'userid')
user_action = cnt_stat(user_action, ['feedid', 'date_'], 'userid', use_nunique=False)
user_action = cnt_stat(user_action, 'userid', 'feedid')
user_action = cnt_stat(user_action, ['userid', 'date_'], 'feedid', use_nunique=False)
user_action = cnt_stat(user_action, 'userid', 'authorid', use_cnt=False)
user_action = cnt_stat(user_action, 'authorid', 'userid')
user_action = cnt_stat(user_action, ['authorid', 'date_'], 'userid', use_nunique=False)
user_action = cnt_stat(user_action, ['userid', 'authorid'], 'feedid', use_nunique=False)
user_action = cnt_stat(user_action, ['authorid', 'feedid'], 'userid', use_nunique=False)
user_action = cnt_stat(user_action, ['userid', 'authorid', 'date_'], 'feedid', use_nunique=False)
user_action = cnt_stat(user_action, ['authorid', 'feedid', 'date_'], 'userid', use_nunique=False)
user_action['userid_authorid_ratio'] = user_action['userid_authorid_cnt'] / user_action['global_userid_cnt']
user_action['feedid_authorid_ratio'] = user_action['authorid_feedid_cnt'] / user_action['global_authorid_cnt']
user_action['userid_date_ratio'] = user_action['userid_date__cnt'] / user_action['global_userid_cnt']
user_action['feedid_date_ratio'] = user_action['feedid_date__cnt'] / user_action['global_feedid_cnt']
user_action['authorid_date_ratio'] = user_action['authorid_date__cnt'] / user_action['global_authorid_cnt']
for t in ['feedid', 'authorid']:
user_action[f'date_first_{t}'] = user_action.groupby(t)['date_'].transform('min')
user_action[f'diff_first_{t}'] = user_action['date_'] - user_action[f'date_first_{t}']
train_ = user_action[(user_action['date_'] < 15) & (user_action['date_'] > 1)]
train_ = train_.drop(['authorid', 'date_first_feedid', 'date_first_authorid'], axis=1)
train_ = reduce_mem_usage(train_)
train_ = train_.reset_index(drop=True)
train_.to_feather(train_save_path)
feed_list = (['feedid'], ['feedid', 'global_feedid_cnt', 'global_feedid_dcnt', 'date_first_feedid'])
user_list = (['userid'], ['userid', 'global_userid_cnt', 'global_userid_dcnt'])
author_list = (['authorid'], ['authorid', 'global_authorid_cnt', 'global_authorid_dcnt', 'date_first_authorid'])
user_author_list = (['userid', 'authorid'], ['userid', 'authorid', 'userid_authorid_cnt'])
author_feed_list = (['authorid', 'feedid'], ['authorid', 'feedid', 'authorid_feedid_cnt'])
for i in [feed_list, user_list, author_list, user_author_list, author_feed_list]:
file_name = '_'.join(i[0])
tmp_df = user_action[i[1]].drop_duplicates(subset=i[0], keep='first').reset_index(drop=True)
print(f'for {file_name} now rows:', tmp_df.shape[0])
tmp_df = reduce_mem_usage(tmp_df)
tmp_df.to_feather(base_hist_path + f'_{file_name}.feather')
def gen_tm_ratio(train_path, feed_info_path, train_save_path, hist_path):
user_action = pd.read_feather(train_path)
del user_action['device'], user_action['stay'], user_action['play']
feed_info = pd.read_csv(feed_info_path, usecols=['feedid', 'authorid'])
user_action = user_action.merge(feed_info, on='feedid', how='left')
del feed_info
user_action['user_authorid'] = user_action['userid'].astype(str) + '_' + user_action['authorid'].astype(str)
print(f'train shape:{user_action.shape[0]}')
tm_cols = ['read_comment', 'comment', 'like', 'click_avatar', 'forward', 'follow', 'favorite']
tm_ids = ['userid', 'feedid', 'authorid', 'user_authorid']
for d in tqdm(tm_ids):
grp_tmp = user_action.groupby(d)[tm_cols].agg('sum').reset_index()
now_cols = [f'{d}_{i}_ratio' for i in tm_cols]
grp_tmp.columns = [d] + now_cols
grp_tmp['tmp_sum'] = grp_tmp[now_cols].sum(axis=1)
for i in now_cols:
grp_tmp[i] = grp_tmp[i] / grp_tmp['tmp_sum']
del grp_tmp['tmp_sum']
if 'user' in d:
for t in tm_cols:
tmp_df = user_action[user_action[t] == 1]
tmp_df = tmp_df.groupby(d)['date_'].agg([[f'hist_{d}_{t}_date_diff', 'max']])
tmp_df[f'hist_{d}_{t}_date_diff'] = 15 - tmp_df[f'hist_{d}_{t}_date_diff']
grp_tmp = grp_tmp.merge(tmp_df, on=d, how='left')
grp_tmp = reduce_mem_usage(grp_tmp)
grp_tmp.to_feather(f'{hist_path}_{d}.feather')
for t in tqdm(range(2, 15)):
prev_tmp = user_action[user_action['date_'] < t]
now_tmp = user_action[user_action['date_'] == t]
for d in tm_ids:
grp_tmp = prev_tmp.groupby(d)[tm_cols].agg(['sum']).reset_index()
now_cols = [f'{d}_{i}_ratio' for i in tm_cols]
grp_tmp.columns = [d] + now_cols
grp_tmp['tmp_sum'] = grp_tmp[now_cols].sum(axis=1)
for i in now_cols:
grp_tmp[i] = grp_tmp[i] / grp_tmp['tmp_sum']
del grp_tmp['tmp_sum']
if 'user' in d:
for p in tm_cols:
tmp_df = prev_tmp[user_action[p] == 1]
tmp_df = tmp_df.groupby(d)['date_'].agg([[f'hist_{d}_{p}_date_diff', 'max']])
grp_tmp = grp_tmp.merge(tmp_df, on=d, how='left')
grp_tmp[f'hist_{d}_{p}_date_diff'] = t - grp_tmp[f'hist_{d}_{p}_date_diff']
now_tmp = now_tmp.merge(grp_tmp, on=d, how='left')
if d not in ['userid', 'feedid']:
del now_tmp[d]
for c in tm_cols:
del now_tmp[c]
now_tmp.to_feather(f'{train_save_path}_{t}.feather')
# 各个ID类特征的点击率
def gen_tm_feats(train_path, feed_info_path, km_path, train_save_path, hist_path):
user_action = pd.read_feather(train_path)
feed_info = pd.read_csv(feed_info_path, usecols=['feedid', 'authorid', 'videoplayseconds', 'manual_tag_list', 'machine_tag_list'])
km = pd.read_csv(km_path)
feed_info = feed_info.merge(km, on='feedid', how='left')
feed_info['machine_tag_list'] = feed_info['machine_tag_list'].apply(
lambda x: x.split(';') if isinstance(x, str) else x)
feed_info['machine_tag_list'] = feed_info['machine_tag_list'].apply(
lambda x: ';'.join([i.split(' ')[0] for i in x]) if isinstance(x, list) else np.nan)
for m in tqdm(['manual_tag_list', 'machine_tag_list']):
feed_info[m] = feed_info[m].astype(str).apply(lambda x: ';'.join(sorted(x.split(';'))))
user_action = user_action.merge(feed_info, on='feedid', how='left')
user_action['play'] = user_action['play'] / 1000 / user_action['videoplayseconds']
user_action['stay'] = user_action['stay'] / 1000 / user_action['videoplayseconds']
del feed_info
del user_action['device'], user_action['videoplayseconds']
tm_cols = ['authorid', 'manual_tag_list', 'machine_tag_list', 'km_label']
connect_cols = ['authorid', 'manual_tag_list', 'machine_tag_list', 'km_label']
for n in tqdm(tm_cols):
user_action[f'user_{n}'] = user_action['userid'].astype(str) + '_' + user_action[n].astype(str)
connect_cols.append(f'user_{n}')
print(f'train shape:{user_action.shape[0]}')
tm_cols = ['read_comment', 'comment', 'like', 'click_avatar', 'forward', 'follow', 'favorite', 'play', 'stay']
tm_ids = ['userid', 'feedid'] + connect_cols
for d in tqdm(tm_ids):
grp_tmp = user_action.groupby(d)[tm_cols].agg('mean').reset_index()
grp_tmp.columns = [d] + [f'{d}_{i}_tm' for i in tm_cols]
grp_tmp = reduce_mem_usage(grp_tmp)
grp_tmp.to_feather(f'{hist_path}_{d}.feather')
for t in tqdm(range(2, 15)):
prev_tmp = user_action[user_action['date_'] < t]
now_tmp = user_action[user_action['date_'] == t]
for d in tm_ids:
grp_tmp = prev_tmp.groupby(d)[tm_cols].agg(['mean']).reset_index()
grp_tmp.columns = [d] + [f'{d}_{i}_tm' for i in tm_cols]
now_tmp = now_tmp.merge(grp_tmp, on=d, how='left')
if d not in ['userid', 'feedid']:
del now_tmp[d]
for c in tm_cols:
del now_tmp[c]
now_tmp.to_feather(f'{train_save_path}_{t}.feather')
def gen_svd_feats(train_path, feed_info_path, train_save_path, hist_path):
user_action = pd.read_feather(train_path)
r1_test_a = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data1/test_a.csv', usecols=['userid', 'feedid'])
r1_test_b = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data1/test_b.csv', usecols=['userid', 'feedid'])
r2_test_a = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data2/test_a.csv', usecols=['userid', 'feedid'])
test = pd.concat([r1_test_a, r1_test_b, r2_test_a], axis=0, ignore_index=True)
del r1_test_a, r1_test_b, r2_test_a
test['date_'] = 15
user_action = user_action[test.columns].append(test)
del test
m_list = ['manual_keyword_list', 'machine_keyword_list', 'manual_tag_list', 'machine_tag_list']
n_list = ['description']
# , 'ocr', 'asr', 'description_char', 'ocr_char', 'asr_char']
feed_info = pd.read_csv(feed_info_path, usecols=['feedid'] + m_list + n_list)
feed_info['machine_tag_list'] = feed_info['machine_tag_list'].apply(
lambda x: x.split(';') if isinstance(x, str) else x)
feed_info['machine_tag_list'] = feed_info['machine_tag_list'].apply(
lambda x: ';'.join([i.split(' ')[0] for i in x]) if isinstance(x, list) else np.nan)
user_action = user_action.merge(feed_info, on='feedid', how='left')
del feed_info
for m in tqdm(m_list + n_list):
user_action[m] = user_action[m].astype(str)
if m in m_list:
user_action[m] = user_action[m].apply(lambda x: x.replace(';', ' '))
user_action = gen_svd_df(user_action, m)
train_ = user_action[(user_action['date_'] < 15) & (user_action['date_'] > 1)]
train_ = reduce_mem_usage(train_)
train_ = train_.reset_index(drop=True)
train_.to_feather(train_save_path)
del user_action['userid'], user_action['date_']
user_action = user_action.drop_duplicates(subset='feedid', keep='first')
user_action = reduce_mem_usage(user_action)
user_action = user_action.reset_index(drop=True)
user_action.to_feather(hist_path)
# 原始512emb svd降维
def svd_feed(feed_emb_path, save_path, output_num=10, seed=1024):
feed_emb = pd.read_csv(feed_emb_path)
feed_emb['feed_embedding'] = feed_emb['feed_embedding'].apply(
lambda x: np.array(x.strip().split(' '), dtype='float32'))
feed_embedding = np.array([i for i in feed_emb['feed_embedding'].values])
svd_tmp = TruncatedSVD(n_components=output_num, n_iter=10, random_state=seed).fit_transform(feed_embedding)
svd_tmp = pd.DataFrame(svd_tmp)
svd_tmp.columns = ['feed_emb_{}'.format(i) for i in range(output_num)]
svd_tmp = pd.concat([feed_emb[['feedid']], svd_tmp], axis=1)
svd_tmp.to_feather(save_path)
# 原始512emb聚类获取类别特征，便于后续的target encoding
def get_kmeans_label(feed_emb_path, save_path, num):
feed_emb = pd.read_csv(feed_emb_path)
feed_emb['feed_embedding'] = feed_emb['feed_embedding'].apply(
lambda x: np.array(x.strip().split(' '), dtype='float32'))
feed_embedding = np.array([i for i in feed_emb['feed_embedding'].values])
del feed_emb['feed_embedding']
km = faiss.Kmeans(512, num, niter=500, verbose=True, gpu=False)
km.train(feed_embedding)
print('train finished.')
_, labels = km.index.search(feed_embedding, 1)
feed_emb['km_label'] = labels
feed_emb.to_csv(save_path, index=False)
def w2v_feat(df, feat, length, w2v_model_path):
if os.path.exists(w2v_model_path):
print('find existing w2v model: {}'.format(w2v_model_path))
model = Word2Vec.load(w2v_model_path)
else:
print('start word2vec, use cpu count: {} ...'.format(multiprocessing.cpu_count()))
model = Word2Vec(df[feat].values, vector_size=length, window=5, min_count=1, sg=1, hs=0, negative=5,
workers=multiprocessing.cpu_count() * 2, epochs=10, seed=1)
model.save(w2v_model_path)
return model
# 用户浏览feed聚合后进行word2vec，获取feed与用户的交互emd特征
def get_user_feed_w2v(train_path, w2v_path, save_path, num):
user_action = pd.read_feather(train_path)
r1_test_a = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data1/test_a.csv', usecols=['userid', 'feedid'])
r1_test_b = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data1/test_b.csv', usecols=['userid', 'feedid'])
r2_test_a = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data2/test_a.csv', usecols=['userid', 'feedid'])
test = pd.concat([r1_test_a, r1_test_b, r2_test_a], axis=0, ignore_index=True)
del r1_test_a, r1_test_b, r2_test_a
test['date_'] = 15
user_action = user_action[test.columns].append(test)
del test
user_action['feedid'] = user_action['feedid'].astype(str)
feed_list = user_action['feedid'].unique()
user_action = user_action.groupby('userid')['feedid'].agg(list).reset_index()
model = w2v_feat(user_action, 'feedid', num, w2v_path)
tmp_list = []
for x in tqdm(feed_list):
tmp_list.append(model.wv[x])
tmp_list = pd.DataFrame(tmp_list, columns=[f'fu_w2v_{w}' for w in range(num)])
tmp_list['feedid'] = feed_list
tmp_list['feedid'] = tmp_list['feedid'].astype(int)
tmp_list = reduce_mem_usage(tmp_list)
tmp_list.to_feather(save_path)
def get_user_feed_svd(train_path, save_path):
user_action = pd.read_feather(train_path)
r1_test_a = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data1/test_a.csv', usecols=['userid', 'feedid'])
r1_test_b = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data1/test_b.csv', usecols=['userid', 'feedid'])
r2_test_a = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data2/test_a.csv', usecols=['userid', 'feedid'])
test = pd.concat([r1_test_a, r1_test_b, r2_test_a], axis=0, ignore_index=True)
del r1_test_a, r1_test_b, r2_test_a
test['date_'] = 15
user_action = user_action[test.columns].append(test)
del test
user_action['userid'] = user_action['userid'].astype(str)
user_action = user_action.groupby('feedid')['userid'].agg(lambda x: ' '.join(list(x))).reset_index()
user_action.columns = ['feedid', 'ulist']
user_action = gen_svd_df(user_action, 'ulist')
user_action = reduce_mem_usage(user_action)
user_action.to_feather(save_path)
# 当前feed与历史feed emb均值做差，获取emb的差异性特征
def gen_hist_emb_diff(train_path, emb_path, train_save_path, hist_path):
user_action = pd.read_feather(train_path)
del user_action['device'], user_action['play'], user_action['stay']
feed_emb = pd.read_feather(emb_path)
emb_cols = [e for e in feed_emb.columns if e != 'feedid']
hist_cols = [f'hist_mean_{c}' for c in emb_cols]
user_action = user_action.merge(feed_emb, on='feedid', how='left')
del feed_emb
tm_cols = ['read_comment', 'comment', 'like', 'click_avatar', 'forward', 'follow', 'favorite']
for d in tqdm(tm_cols):
grp_tmp = user_action[user_action[d] == 1].groupby('userid')[emb_cols].agg('mean').reset_index()
grp_tmp.columns = ['userid'] + [f'hist_mean_{d}_{i}' for i in emb_cols]
grp_tmp = reduce_mem_usage(grp_tmp)
grp_tmp.to_feather(f'{hist_path}_{d}_emb_mean.feather')
train_df = []
for t in tqdm(range(2, 15)):
prev_tmp = user_action[user_action['date_'] < t]
now_tmp = user_action[user_action['date_'] == t]
for d in tm_cols:
grp_tmp = prev_tmp[prev_tmp[d] == 1].groupby('userid')[emb_cols].agg('mean').reset_index()
grp_tmp.columns = ['userid'] + hist_cols
now_tmp = now_tmp.merge(grp_tmp, on='userid', how='left')
for i in emb_cols:
now_tmp[f'{d}_{i}'] = now_tmp[i] - now_tmp[f'hist_mean_{i}']
del now_tmp[f'hist_mean_{i}']
del now_tmp[d]
now_tmp = now_tmp.drop(emb_cols, axis=1)
train_df.append(now_tmp)
train_ = pd.concat(train_df, axis=0, ignore_index=True)
train_['feedid'] = train_['feedid'].astype(int)
print(f'train shape:{train_.shape[0]}')
train_ = reduce_mem_usage(train_)
train_.to_feather(train_save_path)
def gen_hist_w2v_mean(train_path, emb_path, save_train_path):
user_action = pd.read_feather(train_path)
user_action = user_action[['userid', 'feedid', 'date_', 'like']]
feed_info = pd.read_feather(emb_path)
emb_cols = [e for e in feed_info.columns if e != 'feedid']
user_action = user_action.merge(feed_info, on='feedid', how='left')
del feed_info
tm_cols = ['like']
train_df = []
for t in tqdm(range(2, 15)):
prev_tmp = user_action[user_action['date_'] < t]
now_tmp = user_action[user_action['date_'] == t][['userid', 'feedid', 'date_']]
for d in tm_cols:
grp_tmp = prev_tmp[prev_tmp[d] == 1].groupby('userid')[emb_cols].agg('mean').reset_index()
grp_tmp.columns = ['userid'] + [f'hist_mean_{d}_{i}' for i in emb_cols]
now_tmp = now_tmp.merge(grp_tmp, on='userid', how='left')
train_df.append(now_tmp)
train_ = pd.concat(train_df, axis=0, ignore_index=True)
print(f'train shape:{train_.shape}')
train_ = reduce_mem_usage(train_)
train_.to_feather(save_train_path)
def gen_feats(base_path, ori_train_path, save_path):
ori_feed_path = f'{base_path}/feed_info.csv'
ori_emb_path = f'{base_path}/feed_embeddings.csv'
b_path = f'{save_path}/base'
s_path = f'{save_path}/svd'
t_path = f'{save_path}/base_tm'
dec_path = f'{save_path}/feed_emb_dec.feather'
km_path = f'{save_path}/feed_km.csv'
fu_w2v_path = f'{save_path}/user_feed.w2v'
feed_w2v_path = f'{save_path}/fu_w2v.feather'
w_path1 = f'{save_path}/diff_emb'
w_path2 = f'{save_path}/diff_fu'
l_path = f'{save_path}/hist_mean_w2v'
gen_base_feats(ori_train_path, ori_feed_path, f'{b_path}_train.feather', f'{b_path}_hist')
print('the generation of base feats finished.')
gen_tag_mean(ori_train_path, ori_feed_path, save_path)
print('the generation of tag ctr feats finished.')
gen_svd_feats(ori_train_path, ori_feed_path, f'{s_path}_train.feather', f'{s_path}_hist.feather')
print('the generation of svd feats finished.')
get_kmeans_label(ori_emb_path, km_path, 1000)
print('the generation of emb kmeans label finished.')
svd_feed(ori_emb_path, dec_path)
print('origin emb svd finished.')
get_user_feed_w2v(ori_train_path, fu_w2v_path, feed_w2v_path, 10)
print('the generation of w2v feats finished.')
gen_tm_feats(ori_train_path, ori_feed_path, km_path, f'{t_path}_train', f'{t_path}_hist')
print('the generation of target mean feats finished.')
gen_hist_emb_diff(ori_train_path, dec_path, f'{w_path1}_train.feather', f'{w_path1}_hist')
gen_hist_emb_diff(ori_train_path, feed_w2v_path, f'{w_path2}_train.feather', f'{w_path2}_hist')
print('the generation of hist emb diff feats finished.')
get_user_feed_svd(ori_train_path, f'{save_path}/svd_under_feed.feather')
print('the generation of hist user feed svd finished.')
gen_hist_w2v_mean(ori_train_path, feed_w2v_path, f'{l_path}_train.feather')
print('the generation of hist emb mean feats finished.')
def get_ori_train():
train2 = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data2/user_action.csv')
train1 = pd.read_csv('./wbdc2021/data/wedata/wechat_algo_data1/user_action.csv')
train2 = train2.append(train1)
del train1
train2 = train2.reset_index(drop=True)
train2 = reduce_mem_usage(train2)
print(train2.shape)
train2.to_feather('./feats/user_action.feather')
train2 = train2.sort_values('date_')
train2 = train2.drop_duplicates(subset=['userid', 'feedid'], keep='first').reset_index(drop=True)
print(train2.shape)
train2.to_feather('./feats/user_action_drop_repeat.feather')
def merge_file(path):
tmp_list = []
for i in tqdm(range(2, 15)):
tmp_df = pd.read_feather(f'{path}_{i}.feather')
tmp_list.append(tmp_df)
tmp_list = pd.concat(tmp_list, axis=0, ignore_index=True)
print(f'train shape:{tmp_list.shape[0]}')
tmp_list = reduce_mem_usage(tmp_list)
tmp_list.to_feather(f'{path}_all.feather')
def tag_ctr_filter(path, tag):
cnt_dict, read_dict, comment_dict, like_dict, click_dict, forward_dict, follow_dict, favorite_dict = load_obj(f'{path}/{tag}_user_ctr_dict.pkl')
tmp_list = []
for d in [read_dict, comment_dict, like_dict, click_dict, forward_dict, follow_dict, favorite_dict]:
d = {k: v / cnt_dict[k] for k, v in d.items() if v > 0}
tmp_list.append(d)
save_obj(tmp_list, f'{path}/{tag}_user_ctr_7dict.pkl')
def load_train_file(base_path):
for_train = pd.read_feather(f'{base_path}/base_train.feather')
all_kw_tm = pd.read_feather(f'{base_path}/machine_keyword_list_user_ctr_all.feather')
all_tag_tm = pd.read_feather(f'{base_path}/manual_tag_list_user_ctr_all.feather')
train_svd = pd.read_feather(f'{base_path}/svd_train.feather')
train_base_tm = pd.read_feather(f'{base_path}/base_tm_train_all.feather')
train_emb = pd.read_feather(f'{base_path}/diff_emb_train.feather')
train_w2v = pd.read_feather(f'{base_path}/diff_fu_train.feather')
fu_w2v = pd.read_feather(f'{base_path}/fu_w2v.feather')
train_wm = pd.read_feather(f'{base_path}/hist_mean_w2v_train.feather')
feed_svd = pd.read_feather(f'{base_path}/svd_under_feed.feather')
for_train = for_train.merge(train_base_tm, on=['feedid', 'userid', 'date_'], how='left')
del train_base_tm
for_train = for_train.merge(train_svd, on=['feedid', 'userid', 'date_'], how='left')
del train_svd
for_train = for_train.merge(all_tag_tm, on=['feedid', 'userid', 'date_'], how='left')
del all_tag_tm
for_train = for_train.merge(all_kw_tm, on=['feedid', 'userid', 'date_'], how='left')
del all_kw_tm
for_train = for_train.merge(train_emb, on=['feedid', 'userid', 'date_'], how='left')
del train_emb
for_train = for_train.merge(fu_w2v, on='feedid', how='left')
del fu_w2v
for_train = for_train.merge(train_w2v, on=['feedid', 'userid', 'date_'], how='left')
del train_w2v
for_train = for_train.merge(feed_svd, on='feedid', how='left')
del feed_svd
for_train = for_train.merge(train_wm, on=['feedid', 'userid', 'date_'], how='left')
del train_wm
for_train.to_feather(f'{base_path}/round2_train_feats.feather')
def load_test_file(base_path, test_path):
feed_info = pd.read_feather(f'{base_path}/feed_info.feather')
for_test = pd.read_csv(test_path)
for_test = for_test.merge(feed_info, on='feedid', how='left')
base_list = [['feedid'], ['userid'], ['authorid'], ['userid', 'authorid'], ['authorid', 'feedid']]
for b in base_list:
col_name = '_'.join(b) + '_date__cnt'
for_test[col_name] = for_test.groupby(b)['device'].transform('count')
del for_test['device']
feed_list = ['feedid']
user_list = ['userid']
author_list = ['authorid']
user_author_list = ['userid', 'authorid']
author_feed_list = ['authorid', 'feedid']
for i in [feed_list, user_list, author_list, user_author_list, author_feed_list]:
file_name = '_'.join(i)
tmp_df = pd.read_feather(f'{base_path}/base_hist_{file_name}.feather')
for_test = for_test.merge(tmp_df, on=i, how='left')
# 冷启动填充
for n in ['feedid', 'authorid']:
for_test.loc[for_test[f'global_{n}_cnt'].isnull(), f'global_{n}_cnt'] = for_test.loc[
for_test[f'global_{n}_cnt'].isnull(), f'{n}_date__cnt']
for_test.loc[for_test[f'global_{n}_dcnt'].isnull(), f'global_{n}_dcnt'] = for_test.loc[
for_test[f'global_{n}_dcnt'].isnull(), f'{n}_date__cnt']
for n in ['userid_authorid', 'authorid_feedid']:
for_test.loc[for_test[f'{n}_cnt'].isnull(), f'{n}_cnt'] = for_test.loc[
for_test[f'{n}_cnt'].isnull(), f'{n}_date__cnt']
for_test['userid_authorid_ratio'] = for_test['userid_authorid_cnt'] / for_test['global_userid_cnt']
for_test['feedid_authorid_ratio'] = for_test['authorid_feedid_cnt'] / for_test['global_authorid_cnt']
for_test['userid_date_ratio'] = for_test['userid_date__cnt'] / for_test['global_userid_cnt']
for_test['feedid_date_ratio'] = for_test['feedid_date__cnt'] / for_test['global_feedid_cnt']
for_test['authorid_date_ratio'] = for_test['authorid_date__cnt'] / for_test['global_authorid_cnt']
for t in ['feedid', 'authorid']:
for_test[f'diff_first_{t}'] = 15 - for_test[f'date_first_{t}']
del for_test[f'date_first_{t}']
for_test[f'diff_first_{t}'] = for_test[f'diff_first_{t}'].fillna(0)
m_list = ['machine_keyword_list', 'manual_tag_list']
label_list = ['read_comment', 'comment', 'like', 'click_avatar', 'forward', 'follow', 'favorite']
for m in m_list:
print(m)
tmp_list = []
list_ids = [f'tmp_{m}'] + label_list
for_test[f'tmp_{m}'] = for_test['userid'].astype(str) + '|' + for_test[m].astype(str)
read_, comment_, like_, click_, forward_, follow_, favorite_ = load_obj(f'{base_path}/{m}_user_ctr_7dict.pkl')
for f, c in tqdm(for_test[['feedid', f'tmp_{m}']].values):
user, ids = c.split('|')
if ';' in ids:
for j in ids.split(';'):
tmp_list.append([user, f, f'{user}|{j}'])
else:
tmp_list.append([user, f, np.nan])
tmp_list = pd.DataFrame(tmp_list, columns=['userid', 'feedid', f'tmp_{m}'])
for c in zip(list_ids[1:], [read_, comment_, like_, click_, forward_, follow_, favorite_]):
tmp_list[c[0]] = tmp_list[f'tmp_{m}'].map(c[1]).fillna(0)
del tmp_list[f'tmp_{m}'], for_test[f'tmp_{m}']
stat_cols = ['min', 'max', 'mean', 'std', 'count']
tmp_list = tmp_list.groupby(['userid', 'feedid'])[list_ids[1:]].agg(stat_cols).reset_index()
tmp_list.columns = ['userid', 'feedid'] + [f'{m}_{i}_{j}' for i in list_ids[1:] for j in stat_cols]
for i in list_ids[1:]:
tmp_list[f'{m}_{i}_std'] = np.sqrt(
tmp_list[f'{m}_{i}_std'] ** 2 * (tmp_list[f'{m}_{i}_count'] - 1) / tmp_list[f'{m}_{i}_count'])
del tmp_list[f'{m}_{i}_count']
tmp_list['userid'] = tmp_list['userid'].astype(int)
for_test = for_test.merge(tmp_list, on=['userid', 'feedid'], how='left')
del for_test['machine_keyword_list']
test_svd = pd.read_feather(f'{base_path}/svd_hist.feather')
for_test = for_test.merge(test_svd, on='feedid', how='left')
svd_dict = {'svd_tfidf_manual_keyword_list_0': 0.292,
'svd_tfidf_manual_keyword_list_1': 0.04446,
'svd_tfidf_manual_keyword_list_2': 0.0377,
'svd_tfidf_manual_keyword_list_3': 0.009636,
'svd_tfidf_manual_keyword_list_4': 0.01441,
'svd_cntvec_manual_keyword_list_0': 0.292,
'svd_cntvec_manual_keyword_list_1': 0.11945,
'svd_cntvec_manual_keyword_list_2': 0.08264,
'svd_cntvec_manual_keyword_list_3': 0.002274,
'svd_cntvec_manual_keyword_list_4': 0.04883,
'svd_tfidf_machine_keyword_list_0': 0.6016,
'svd_tfidf_machine_keyword_list_1': 0.0257,
'svd_tfidf_machine_keyword_list_2': 0.0202,
'svd_tfidf_machine_keyword_list_3': 0.00846,
'svd_tfidf_machine_keyword_list_4': 0.01064,
'svd_cntvec_machine_keyword_list_0': 0.04886,
'svd_cntvec_machine_keyword_list_1': 0.6016,
'svd_cntvec_machine_keyword_list_2': 0.0418,
'svd_cntvec_machine_keyword_list_3': 0.0287,
'svd_cntvec_machine_keyword_list_4': 0.03296,
'svd_tfidf_manual_tag_list_0': 0.1848,
'svd_tfidf_manual_tag_list_1': 0.0705,
'svd_tfidf_manual_tag_list_2': 0.04874,
'svd_tfidf_manual_tag_list_3': 0.06216,
'svd_tfidf_manual_tag_list_4': -0.01532,
'svd_cntvec_manual_tag_list_0': 0.7734,
'svd_cntvec_manual_tag_list_1': 0.04358,
'svd_cntvec_manual_tag_list_2': -0.02547,
'svd_cntvec_manual_tag_list_3': -0.03415,
'svd_cntvec_manual_tag_list_4': 0.0614,
'svd_tfidf_machine_tag_list_0': 0.2861,
'svd_tfidf_machine_tag_list_1': -0.0294,
'svd_tfidf_machine_tag_list_2': 0.0158,
'svd_tfidf_machine_tag_list_3': 0.04144,
'svd_tfidf_machine_tag_list_4': -0.0001535,
'svd_cntvec_machine_tag_list_0': 1.076,
'svd_cntvec_machine_tag_list_1': 0.04974,
'svd_cntvec_machine_tag_list_2': -0.0243,
'svd_cntvec_machine_tag_list_3': 0.009315,
'svd_cntvec_machine_tag_list_4': 0.02483,
'svd_tfidf_description_0': 0.1698,
'svd_tfidf_description_1': 0.02164,
'svd_tfidf_description_2': -0.02055,
'svd_tfidf_description_3': -0.0198,
'svd_tfidf_description_4': 0.01101,
'svd_cntvec_description_0': 3.06,
'svd_cntvec_description_1': -1.725,
'svd_cntvec_description_2': 0.444,
'svd_cntvec_description_3': 0.01752,
'svd_cntvec_description_4': -0.2338}
for k, v in svd_dict.items():
for_test[k] = for_test[k].fillna(v)
del test_svd
tm_cols = ['authorid', 'manual_tag_list', 'machine_tag_list', 'km_label']
connect_cols = ['authorid', 'manual_tag_list', 'machine_tag_list', 'km_label']
for n in tqdm(tm_cols):
for_test[f'user_{n}'] = for_test['userid'].astype(str) + '_' + for_test[n].astype(str)
connect_cols.append(f'user_{n}')
tm_ids = ['userid', 'feedid'] + connect_cols
for d in tqdm(tm_ids):
grp_tmp = pd.read_feather(f'{base_path}/base_tm_hist_{d}.feather')
for_test = for_test.merge(grp_tmp, on=d, how='left')
if d not in ['userid', 'feedid', 'user_authorid', 'authorid']:
del for_test[d]
fu_w2v = pd.read_feather(f'{base_path}/fu_w2v.feather')
feed_emb = pd.read_feather(f'{base_path}/feed_emb_dec.feather')
fu_dict = {'fu_w2v_0': -0.4192, 'fu_w2v_1': -0.7144, 'fu_w2v_2': -1.131, 'fu_w2v_3': -0.279, 'fu_w2v_4': 1.125,
'fu_w2v_5': 0.841, 'fu_w2v_6': -0.3801, 'fu_w2v_7': 0.1775, 'fu_w2v_8': -1.9795, 'fu_w2v_9': -1.745}
for mode, emb in zip(['emb', 'fu'], [feed_emb, fu_w2v]):
emb_cols = [e for e in emb.columns if e != 'feedid']
for_test = for_test.merge(emb, on='feedid', how='left')
if mode == 'fu':
for k, v in fu_dict.items():
for_test[k] = for_test[k].fillna(v)
for d in label_list:
grp_tmp = pd.read_feather(f'{base_path}/diff_{mode}_hist_{d}_emb_mean.feather')
for_test = for_test.merge(grp_tmp, on='userid', how='left')
for i in emb_cols:
for_test[f'{d}_{i}'] = for_test[i] - for_test[f'hist_mean_{d}_{i}']
if d != 'like' or mode == 'emb':
del for_test[f'hist_mean_{d}_{i}']
if mode == 'emb':
for_test = for_test.drop(emb_cols, axis=1)
feed_svd = pd.read_feather(f'{base_path}/svd_under_feed.feather')
for_test = for_test.merge(feed_svd, on='feedid', how='left')
ulist_dict = {'svd_tfidf_ulist_0': 0.03958, 'svd_tfidf_ulist_1': -0.0002842, 'svd_tfidf_ulist_2': 0.001081,
'svd_tfidf_ulist_3': 0.002897, 'svd_tfidf_ulist_4': 0.01572, 'svd_cntvec_ulist_0': 0.972,
'svd_cntvec_ulist_1': -0.02539, 'svd_cntvec_ulist_2': -0.1755, 'svd_cntvec_ulist_3': -0.4963,
'svd_cntvec_ulist_4': 0.107}
for k, v in ulist_dict.items():
for_test[k] = for_test[k].fillna(v)
del feed_svd
for_test.to_feather(f'{base_path}/cbt_round2_test_feats.feather')
def split_train_valid():
train_set = pd.read_feather(feat_path + '/round2_train_feats.feather')
print('origin shape', train_set.shape)
label_cols = ['read_comment', 'comment', 'like', 'click_avatar', 'forward', 'follow', 'favorite']
use_cols = [i for i in train_set.columns if i not in ['feedid', 'userid', 'date_'] + label_cols]
valid_set = train_set[train_set['date_'] == 14]
train_set = train_set[train_set['date_'] < 14]
emb_comment_cols = ['comment_feed_emb_0', 'comment_feed_emb_1', 'comment_feed_emb_2', 'comment_feed_emb_3',
'comment_feed_emb_4', 'comment_feed_emb_5', 'comment_feed_emb_6', 'comment_feed_emb_7',
'comment_feed_emb_8', 'comment_feed_emb_9']
fu_comment_cols = ['comment_fu_w2v_0', 'comment_fu_w2v_1', 'comment_fu_w2v_2', 'comment_fu_w2v_3',
'comment_fu_w2v_4', 'comment_fu_w2v_5', 'comment_fu_w2v_6', 'comment_fu_w2v_7',
'comment_fu_w2v_8', 'comment_fu_w2v_9']
read_comment_cols = [i for i in use_cols if ('like_feed_emb' not in i) and ('click_avatar_feed_emb' not in i) and (
'forward_feed_emb' not in i) and ('follow_feed_emb' not in i) and ('favorite_feed_emb' not in i)
and ('like_fu' not in i) and ('click_avatar_fu' not in i) and ('forward_fu' not in i)
and ('follow_fu' not in i) and ('favorite_fu' not in i) and i not in emb_comment_cols
and i not in fu_comment_cols]
comment_cols = [i for i in use_cols if ('read_comment_feed_emb' not in i) and ('click_avatar_feed_emb' not in i) and (
'forward_feed_emb' not in i) and ('follow_feed_emb' not in i) and ('favorite_feed_emb' not in i) and (
'like_feed_emb' not in i)
and ('read_comment_fu' not in i) and ('click_avatar_fu' not in i) and ('forward_fu' not in i)
and ('follow_fu' not in i) and ('favorite_fu' not in i) and ('like_fu' not in i)]
like_cols = [i for i in use_cols if ('read_comment_feed_emb' not in i) and ('click_avatar_feed_emb' not in i) and (
'forward_feed_emb' not in i) and ('follow_feed_emb' not in i) and ('favorite_feed_emb' not in i)
and ('read_comment_fu' not in i) and ('click_avatar_fu' not in i) and ('forward_fu' not in i)
and ('follow_fu' not in i) and ('favorite_fu' not in i) and i not in emb_comment_cols
and i not in fu_comment_cols]
click_avatar_cols = [i for i in use_cols if ('like_feed_emb' not in i) and ('read_comment_feed_emb' not in i) and (
'forward_feed_emb' not in i) and ('follow_feed_emb' not in i) and ('favorite_feed_emb' not in i)
and ('like_fu' not in i) and ('read_comment_fu' not in i) and ('forward_fu' not in i)
and ('follow_fu' not in i) and ('favorite_fu' not in i) and i not in emb_comment_cols
and i not in fu_comment_cols]
forward_cols = [i for i in use_cols if ('like_feed_emb' not in i) and ('click_avatar_feed_emb' not in i) and (
'read_comment_feed_emb' not in i) and ('follow_feed_emb' not in i) and ('favorite_feed_emb' not in i)
and ('like_fu' not in i) and ('click_avatar_fu' not in i) and ('read_comment_fu' not in i)
and ('follow_fu' not in i) and ('favorite_fu' not in i) and i not in emb_comment_cols
and i not in fu_comment_cols]
follow_cols = [i for i in use_cols if ('like_feed_emb' not in i) and ('click_avatar_feed_emb' not in i) and (
'read_comment_feed_emb' not in i) and ('forward_feed_emb' not in i) and ('favorite_feed_emb' not in i)
and ('like_fu' not in i) and ('click_avatar_fu' not in i) and ('read_comment_fu' not in i)
and ('forward_fu' not in i) and ('favorite_fu' not in i) and i not in emb_comment_cols
and i not in fu_comment_cols]
favorite_cols = [i for i in use_cols if ('like_feed_emb' not in i) and ('click_avatar_feed_emb' not in i) and (
'read_comment_feed_emb' not in i) and ('follow_feed_emb' not in i) and ('forward_feed_emb' not in i)
and ('like_fu' not in i) and ('click_avatar_fu' not in i) and ('read_comment_fu' not in i)
and ('follow_fu' not in i) and ('forward_fu' not in i) and i not in emb_comment_cols
and i not in fu_comment_cols]
train_feats = [read_comment_cols, comment_cols, like_cols, click_avatar_cols, forward_cols, follow_cols, favorite_cols]
for i, j in zip(label_cols, train_feats):
train_set['num_label'] = train_set.groupby('userid')[i].transform('nunique')
tmp_train = train_set[train_set['num_label'] == 2]
del tmp_train['num_label']
tmp_train = tmp_train[['feedid', 'userid', 'date_'] + [i] + j].reset_index(drop=True)
print(f'{i} now train shape:', tmp_train.shape)
tmp_train.to_feather(f'{feat_path}/for_train_{i}.feather')
del tmp_train
valid_set['num_label'] = valid_set.groupby('userid')[i].transform('nunique')
tmp_valid = valid_set[valid_set['num_label'] == 2]
tmp_valid = tmp_valid[['feedid', 'userid', 'date_'] + [i] + j].reset_index(drop=True)
print(f'{i} now valid shape:', tmp_valid.shape)
tmp_valid.to_feather(f'{feat_path}/for_valid_{i}.feather')
del tmp_valid
def split_all():
train_set = pd.read_feather(feat_path + '/round2_train_feats.feather')
print('origin shape', train_set.shape)
label_cols = ['read_comment', 'comment', 'like', 'click_avatar', 'forward', 'follow', 'favorite']
use_cols = [i for i in train_set.columns if i not in ['feedid', 'userid', 'date_'] + label_cols]
emb_comment_cols = ['comment_feed_emb_0', 'comment_feed_emb_1', 'comment_feed_emb_2', 'comment_feed_emb_3',
'comment_feed_emb_4', 'comment_feed_emb_5', 'comment_feed_emb_6', 'comment_feed_emb_7',
'comment_feed_emb_8', 'comment_feed_emb_9']
fu_comment_cols = ['comment_fu_w2v_0', 'comment_fu_w2v_1', 'comment_fu_w2v_2', 'comment_fu_w2v_3',
'comment_fu_w2v_4', 'comment_fu_w2v_5', 'comment_fu_w2v_6', 'comment_fu_w2v_7',
'comment_fu_w2v_8', 'comment_fu_w2v_9']
read_comment_cols = [i for i in use_cols if ('like_feed_emb' not in i) and ('click_avatar_feed_emb' not in i) and (
'forward_feed_emb' not in i) and ('follow_feed_emb' not in i) and ('favorite_feed_emb' not in i)
and ('like_fu' not in i) and ('click_avatar_fu' not in i) and ('forward_fu' not in i)
and ('follow_fu' not in i) and ('favorite_fu' not in i) and i not in emb_comment_cols
and i not in fu_comment_cols]
comment_cols = [i for i in use_cols if
('read_comment_feed_emb' not in i) and ('click_avatar_feed_emb' not in i) and (
'forward_feed_emb' not in i) and ('follow_feed_emb' not in i) and (
'favorite_feed_emb' not in i) and (
'like_feed_emb' not in i)
and ('read_comment_fu' not in i) and ('click_avatar_fu' not in i) and ('forward_fu' not in i)
and ('follow_fu' not in i) and ('favorite_fu' not in i) and ('like_fu' not in i)]
like_cols = [i for i in use_cols if ('read_comment_feed_emb' not in i) and ('click_avatar_feed_emb' not in i) and (
'forward_feed_emb' not in i) and ('follow_feed_emb' not in i) and ('favorite_feed_emb' not in i)
and ('read_comment_fu' not in i) and ('click_avatar_fu' not in i) and ('forward_fu' not in i)
and ('follow_fu' not in i) and ('favorite_fu' not in i) and i not in emb_comment_cols
and i not in fu_comment_cols]
click_avatar_cols = [i for i in use_cols if ('like_feed_emb' not in i) and ('read_comment_feed_emb' not in i) and (
'forward_feed_emb' not in i) and ('follow_feed_emb' not in i) and ('favorite_feed_emb' not in i)
and ('like_fu' not in i) and ('read_comment_fu' not in i) and ('forward_fu' not in i)
and ('follow_fu' not in i) and ('favorite_fu' not in i) and i not in emb_comment_cols
and i not in fu_comment_cols]
forward_cols = [i for i in use_cols if ('like_feed_emb' not in i) and ('click_avatar_feed_emb' not in i) and (