-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathutils.py
1172 lines (907 loc) · 52 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 scipy.sparse as sps
import numpy as np
import pandas as pd
import os
import shutil
import warnings
import math
import Data_manager.split_functions.split_train_validation_random_holdout as split
from HyperparameterTuning.SearchBayesianSkopt import SearchBayesianSkopt
from HyperparameterTuning.SearchSingleCase import SearchSingleCase
from Recommenders.DataIO import DataIO
from recmodels import *
from skopt.space import Real
from HyperparameterTuning.run_hyperparameter_search import runHyperparameterSearch_Collaborative
from sklearn.cluster import KMeans
from OutlierDetection.UnivariateAnomalyDetector import *
def urm_all_ones_visualized():
URM_summed = urm_visualization_all_ones_summed()
nnz_inds = URM_summed.nonzero()
URM_all_ones = sps.csr_matrix((np.ones(len(nnz_inds[0])), (nnz_inds[0], nnz_inds[1])), shape=URM_summed.shape)
return URM_all_ones
def urm_all_ones():
URM_summed = urm_all_ones_summed()
nnz_inds = URM_summed.nonzero()
URM_all_ones = sps.csr_matrix((np.ones(len(nnz_inds[0])), (nnz_inds[0], nnz_inds[1])), shape=URM_summed.shape)
return URM_all_ones
def urm_all_ones_summed():
# Import dataframe for URM matrix
root_path = "data"
interactions = pd.read_csv(os.path.join(root_path, "interactions_and_impressions.csv"), low_memory=False)
data_ICM_type = pd.read_csv(os.path.join(root_path,"data_ICM_type.csv"), low_memory=False)
interactions["Data"] = 1
interactions = interactions.drop(columns=["Impressions"])
all_items = pd.concat([interactions["ItemID"], data_ICM_type["item_id"]], ignore_index=True).unique()
n_users = len(interactions["UserID"].unique())
n_items = len(all_items) # should be considered also the items present in the ICM but not here?
URM_summed = sps.csr_matrix((interactions["Data"].values,
(interactions["UserID"].values, interactions["ItemID"].values)),
shape = (n_users, n_items))
return URM_summed
def urm_seen_or_info(value_seen=1, value_info=0.5): # so 1 if seen, 0.5 if only info page opened
assert value_info < value_seen and value_info != value_seen and value_info != 0, "value_seen = {} and value_info = {} are incompatible".format(value_seen, value_info)
root_path = "data"
interactions = pd.read_csv(os.path.join(root_path, "interactions_and_impressions.csv"), low_memory=False)
data_ICM_type = pd.read_csv(os.path.join(root_path,"data_ICM_type.csv"), low_memory=False)
interactions.loc[interactions["Data"] == 1, "Data"] = value_info
interactions.loc[interactions["Data"] == 0, "Data"] = value_seen
interactions = interactions.drop(columns=["Impressions"])
all_items = pd.concat([interactions["ItemID"], data_ICM_type["item_id"]], ignore_index=True).unique()
n_users = len(interactions["UserID"].unique())
n_items = len(all_items) # should be considered also the items present in the ICM but not here?
interactions.drop_duplicates(inplace=True)
interactions.sort_values(by=['Data', 'UserID', 'ItemID'], inplace=True)
interactions.drop_duplicates(subset = ['UserID', 'ItemID'], inplace=True)
interactions.to_csv(os.path.join(root_path, "seen-or-info.csv"))
URM_seen_or_info = sps.csr_matrix((interactions["Data"].values,
(interactions["UserID"].values, interactions["ItemID"].values)),
shape = (n_users, n_items))
return URM_seen_or_info
def urm_visualization_all_ones_summed():
# Import dataframe for URM matrix
root_path = "data"
interactions = pd.read_csv(os.path.join(root_path, "interactions_and_impressions.csv"), low_memory=False)
data_ICM_type = pd.read_csv(os.path.join(root_path,"data_ICM_type.csv"), low_memory=False)
interactions.drop(interactions[interactions["Data"] == 1].index, inplace = True)
interactions["Data"] = 1
interactions.drop(columns=["Impressions"], inplace = True)
all_items = pd.concat([interactions["ItemID"], data_ICM_type["item_id"]], ignore_index=True).unique()
n_users = len(interactions["UserID"].unique())
n_items = len(all_items) # should be considered also the items present in the ICM but not here?
URM_summed = sps.csr_matrix((interactions["Data"].values,
(interactions["UserID"].values, interactions["ItemID"].values)),
shape = (n_users, n_items))
return URM_summed
def urm_info_all_ones_summed():
# Import dataframe for URM matrix
root_path = "data"
interactions = pd.read_csv(os.path.join(root_path, "interactions_and_impressions.csv"), low_memory=False)
n_users = len(interactions["UserID"].unique())
data_ICM_type = pd.read_csv(os.path.join(root_path,"data_ICM_type.csv"), low_memory=False)
interactions.drop(interactions[interactions["Data"] == 0].index, inplace = True)
interactions["Data"] = 1
interactions.drop(columns=["Impressions"], inplace = True)
all_items = pd.concat([interactions["ItemID"], data_ICM_type["item_id"]], ignore_index=True).unique()
n_items = len(all_items) # should be considered also the items present in the ICM but not here?
URM_summed = sps.csr_matrix((interactions["Data"].values,
(interactions["UserID"].values, interactions["ItemID"].values)),
shape = (n_users, n_items))
return URM_summed
def icm_types():
# Import dataframe for ICM matrix
root_path = "data"
interactions = pd.read_csv(os.path.join(root_path, "interactions_and_impressions.csv"), low_memory=False)
data_ICM_type = pd.read_csv(os.path.join(root_path,"data_ICM_type.csv"), low_memory=False)
all_items = pd.concat([interactions["ItemID"], data_ICM_type["item_id"]], ignore_index=True).unique() # should be considered the items present in the ICM but not in interactions? should be considered the items present in interactions but not in ICM?
all_types = data_ICM_type["feature_id"].unique()
# to have features with consecutive id
mapped_id, original_id = pd.factorize(data_ICM_type["feature_id"].unique())
feature_original_ID_to_index = pd.Series(mapped_id, index=original_id)
data_ICM_type["feature_id"] = data_ICM_type["feature_id"].map(feature_original_ID_to_index)
n_items = len(all_items)
n_types = len(all_types)
ICM_csr = sps.csr_matrix((data_ICM_type["data"].values,
(data_ICM_type["item_id"].values, data_ICM_type["feature_id"].values)),
shape = (n_items, n_types))
return ICM_csr
def icm_length():
# Import dataframe for ICM matrix
root_path = "data"
max_len = 9999
interactions = pd.read_csv(os.path.join(root_path, "interactions_and_impressions.csv"), low_memory=False)
icm_length = pd.read_csv(os.path.join(root_path,"data_ICM_length.csv"), low_memory=False)
all_items = pd.concat([interactions["ItemID"], icm_length["item_id"]], ignore_index=True).unique()
icm_length.drop(columns=["feature_id"], inplace=True)
missing_items = pd.DataFrame()
missing_items["item_id"] = all_items[np.isin(all_items, icm_length["item_id"], invert=True)]
missing_items["data"] = np.nan
icm_length = pd.concat([icm_length, missing_items], ignore_index=True)
icm_length.sort_values(by=["item_id"], inplace=True, ignore_index=True)
icm_length.loc[icm_length["data"] > max_len, "data"] = np.nan
return icm_length
def icm():
# Import dataframe for URM matrix
root_path = "data"
interactions = pd.read_csv(os.path.join(root_path, "interactions_and_impressions.csv"), low_memory=False)
ICM_one_hot = pd.read_csv(os.path.join(root_path,"ICM_one_hot.csv"), low_memory=False)
index= pd.DataFrame()
index["index"] = ICM_one_hot.index
all_items_id = pd.concat([interactions["ItemID"], index["index"]], ignore_index=True).unique()
n_items = len(all_items_id)
n_features = len(ICM_one_hot.columns)
ICM_one_hot = ICM_one_hot.reindex(list(range(0, n_items))).reset_index(drop=True).fillna(0)
ICM_one_hot = sps.csr_matrix(ICM_one_hot, shape = (n_items, n_features))
return ICM_one_hot
def get_info_norm_urm(train_percentage = 0.7, seed=1234):
urm_info = urm_info_all_ones_summed()
urm_vis = urm_visualization_all_ones_summed()
urm_train_vis, _ = split.split_train_in_two_percentage_global_sample(urm_vis,
train_percentage = train_percentage,
seed=seed)
users_stats = statistics_per_user(urm_vis, urm_info)
del urm_vis
urm_train_info, urm_validation_info = split.split_train_in_two_percentage_global_sample(urm_info,
train_percentage = train_percentage,
seed=seed)
users_stats_train = statistics_per_user(urm_train_vis, urm_train_info)
del urm_train_vis
replace_outliers_univariate(urm_info,
replace_with="max",
strategy="MAD")
replace_outliers_univariate(urm_train_info,
replace_with="max",
strategy="MAD")
users_stats_train[users_stats_train["InfoInteractionCount"] == 0] = 1
urm_train_info = urm_train_info / np.array(users_stats_train["InfoInteractionCount"])[:,None]
urm_train_info = sps.csr_matrix(urm_train_info.astype(np.float))
users_stats[users_stats["InfoInteractionCount"] == 0] = 1
urm_info = urm_info / np.array(users_stats["InfoInteractionCount"])[:,None]
urm_info = sps.csr_matrix(urm_info.astype(np.float))
urm_validation_info.data = np.ones(len(urm_validation_info.data))
sps.save_npz(os.path.join("data","info_norm.npz"), urm_info)
sps.save_npz(os.path.join("data","info_norm_train.npz"), urm_train_info)
sps.save_npz(os.path.join("data","info_norm_val.npz"), urm_validation_info)
return urm_info, urm_train_info, urm_validation_info
def get_vis_norm_urm(train_percentage = 0.7, seed=1234):
urm_vis = urm_visualization_all_ones_summed()
urm_info = urm_info_all_ones_summed()
urm_train_info, _ = split.split_train_in_two_percentage_global_sample(urm_info,
train_percentage = train_percentage,
seed=seed)
users_stats = statistics_per_user(urm_vis, urm_info)
items_stats = statistics_per_item(urm_vis, urm_info)
del urm_info
urm_train_vis, urm_validation_vis = split.split_train_in_two_percentage_global_sample(urm_vis,
train_percentage = train_percentage,
seed=seed)
users_stats_train = statistics_per_user(urm_train_vis, urm_train_info)
items_stats_train = statistics_per_item(urm_train_vis, urm_train_info)
del urm_train_info
icm_len = icm_length()
icm_len_train = icm_len.copy()
mask = np.isnan(icm_len_train["data"])
mask = np.transpose(mask.values)
missing_values_len_train = urm_train_vis.max(0).toarray()[0]
missing_values_len_train = missing_values_len_train[mask]
missing_values_len_train[missing_values_len_train == 0] = 1
icm_len_train.loc[np.isnan(icm_len_train["data"].values), "data"] = missing_values_len_train
users_stats_train[users_stats_train["ProfileSeen"] == 0] = 1
urm_train_vis = urm_train_vis / np.transpose(np.array(icm_len_train["data"]))
#urm_train_vis = urm_train_vis / np.array(users_stats["ProfileSeen"])[:,None]
urm_train_vis = sps.csr_matrix(urm_train_vis.astype(np.float))
mask = np.isnan(icm_len["data"])
mask = np.transpose(mask.values)
missing_values_len_train = urm_vis.max(0).toarray()[0]
missing_values_len_train = missing_values_len_train[mask]
missing_values_len_train[missing_values_len_train == 0] = 1
icm_len.loc[np.isnan(icm_len["data"].values), "data"] = missing_values_len_train
users_stats[users_stats["ProfileSeen"] == 0] = 1
urm_vis = urm_vis / np.transpose(np.array(icm_len["data"]))
#urm_vis = urm_vis / np.array(users_stats["ProfileSeen"])[:,None]
urm_vis = sps.csr_matrix(urm_vis.astype(np.float))
urm_validation_vis.data = np.ones(len(urm_validation_vis.data))
sps.save_npz(os.path.join("data","vis_norm.npz"), urm_vis)
sps.save_npz(os.path.join("data","vis_norm_train.npz"), urm_train_vis)
sps.save_npz(os.path.join("data","vis_norm_val.npz"), urm_validation_vis)
return urm_vis, urm_train_vis, urm_validation_vis
def statistics_per_user(urm_seen_train, urm_info_train):
ucm = pd.DataFrame();
profile_length_seen = np.ediff1d(sps.csr_matrix(urm_seen_train).indptr)
seen_count = np.sum(urm_seen_train, axis=1)
ucm["ProfileSeen"] = profile_length_seen
ucm["SeenInteractionCount"] = seen_count
info_count = np.sum(urm_info_train, axis=1)
profile_length_info = np.ediff1d(sps.csr_matrix(urm_info_train).indptr)
ucm["ProfileInfo"] = profile_length_info
ucm["InfoInteractionCount"] = info_count
ucm = ucm.convert_dtypes()
return ucm
def statistics_per_item(urm_seen_train, urm_info_train):
item_stats = pd.DataFrame();
profile_length_seen = np.ediff1d(sps.csr_matrix(urm_seen_train.T).indptr)
seen_count = np.sum(urm_seen_train, axis=0)
seen_count = seen_count.transpose()
item_stats["ProfileSeen"] = profile_length_seen.T
item_stats["SeenInteractionCount"] = seen_count
info_count = np.sum(urm_info_train, axis=0)
info_count = info_count.transpose()
profile_length_info = np.ediff1d(sps.csr_matrix(urm_info_train.T).indptr)
item_stats["ProfileInfo"] = profile_length_info.T
item_stats["InfoInteractionCount"] = info_count
item_stats = item_stats.convert_dtypes()
return item_stats
def remove_outliers_univariate(sparse_matrix, strategy="MAD"):
sparse_matrix = sparse_matrix.tocoo()
if strategy=="MAD":
outliers_mask = get_mad_outliers(sparse_matrix.data.reshape(-1, 1))
elif strategy=="STD":
outliers_mask = get_stddev_outliers(sparse_matrix.data.reshape(-1, 1))
outliers_mask = np.logical_not(outliers_mask)
rows = sparse_matrix.row[outliers_mask]
cols = sparse_matrix.col[outliers_mask]
data = sparse_matrix.data[outliers_mask]
return sps.csr_matrix((data, (rows, cols)), shape = sparse_matrix.shape)
def replace_outliers_univariate(sparse_matrix, replace_with="max", strategy="MAD", tol=3):
sparse_matrix = sparse_matrix.tocoo()
if strategy=="MAD":
outliers_mask = get_mad_outliers(sparse_matrix.data.reshape(-1, 1), tolerance=tol)
elif strategy=="STD":
outliers_mask = get_stddev_outliers(sparse_matrix.data.reshape(-1, 1))
not_outliers_mask = np.logical_not(outliers_mask)
if replace_with=="max":
replacing = sparse_matrix.data[not_outliers_mask].max()
else:
replacing = replace_with
sparse_matrix.data[outliers_mask] = replacing
return sparse_matrix
def get_URM_stacked(URM_csr):
URM_csr = sps.vstack([URM_csr, icm().T])
URM_csr = sps.csr_matrix(URM_csr)
return URM_csr
def get_ICM_stacked(URM_csr):
ICM_csr = sps.csr_matrix(get_URM_stacked(URM_csr).T)
return ICM_csr
def get_urm_visualization_summed_transformed(train_percentage, k=None, seed=None, transformation="minmax"):
URM = urm_visualization_all_ones_summed()
URM_all_ones = urm_all_ones()
URM_train, URM_validation = split.split_train_in_two_percentage_global_sample(URM,
train_percentage = train_percentage,
seed=seed)
_, URM_validation_all_ones = split.split_train_in_two_percentage_global_sample(URM_all_ones,
train_percentage = train_percentage,
seed=seed)
URM_val_coo = URM_validation.tocoo()
URM_val_all_ones_coo = URM_validation_all_ones.tocoo()
#assert (URM_val_coo.row == URM_val_all_ones_coo.row).all() and (URM_val_coo.col == URM_val_all_ones_coo.col).all() and (URM_val_all_ones_coo.data == np.ones(len(URM_val_all_ones_coo.data))).all(), "The validation and training set overlap!"
if k is not None and k > 0:
URM_train = scale_URM_per_clusters(URM_train, k, transformation)
URM = scale_URM_per_clusters(URM, k, transformation)
else:
URM_train = transform_sparse_matrix(URM_train, transformation)
URM = transform_sparse_matrix(URM, transformation)
return URM, URM_train, URM_validation_all_ones
def get_urm_info_summed_transformed():
return
def get_URM_all_ones_summed_transformed(train_percentage, k=None, seed=None, transformation="minmax"):
URM = urm_all_ones_summed()
URM_all_ones = urm_all_ones()
URM_train, URM_validation = split.split_train_in_two_percentage_global_sample(URM,
train_percentage = train_percentage,
seed=seed)
_, URM_validation_all_ones = split.split_train_in_two_percentage_global_sample(URM_all_ones,
train_percentage = train_percentage,
seed=seed)
URM_val_coo = URM_validation.tocoo()
URM_val_all_ones_coo = URM_validation_all_ones.tocoo()
assert (URM_val_coo.row == URM_val_all_ones_coo.row).all() and (URM_val_coo.col == URM_val_all_ones_coo.col).all() and (URM_val_all_ones_coo.data == np.ones(len(URM_val_all_ones_coo.data))).all(), "The validation and training set overlap!"
if k is not None and k > 0:
URM_train = scale_URM_per_clusters(URM_train, k, transformation)
URM = scale_URM_per_clusters(URM, k, transformation)
else:
URM_train = transform_sparse_matrix(URM_train, transformation)
URM = transform_sparse_matrix(URM, transformation)
return URM, URM_train, URM_validation_all_ones
def transform_sparse_matrix(sp_matrix, transformation):
sp_matrix_coo = sp_matrix.tocoo()
if (transformation == "logistic"):
cluster["Data"] = cluster["Data"].apply(logistic_scale_implicit_rating)
elif (transformation == "tanh"):
cluster["Data"] = cluster["Data"].apply(tanh_scale_implicit_rating)
elif (transformation == "minmax"):
transformed_data = range_scaling(sp_matrix_coo.data)
elif (transformation == "std"):
transformed_data = std_norm(sp_matrix_coo.data)
elif (transformation == "robust"):
transformed_data = robust_norm(sp_matrix_coo.data)
else:
print("Wrong transformation type!")
transformed_matrix = sps.csr_matrix((transformed_data,
(sp_matrix_coo.row, sp_matrix_coo.col)),
shape = sp_matrix_coo.shape)
return transformed_matrix
def scale_URM_per_clusters(URM, k, transformation):
df = get_df_from_urm(URM)
clusters_list = get_clusters_dfs_from_df(df, k)
clusters = []
for cluster in clusters_list:
global max_implicit_rating
global min_implicit_rating
max_implicit_rating = cluster["Data"].max()
min_implicit_rating = cluster["Data"].min()
if (transformation == "logistic"):
cluster["Data"] = cluster["Data"].apply(logistic_scale_implicit_rating)
elif (transformation == "tanh"):
cluster["Data"] = cluster["Data"].apply(tanh_scale_implicit_rating)
elif (transformation == "minmax"):
cluster["Data"] = cluster["Data"].apply(ranged_min_max_scale_implicit_rating)
else:
print("Wrong scale type!")
clusters.append(cluster)
URM = pd.concat(clusters)
URM = URM.sort_values(by=['UserID',"ItemID"])
URM.drop(columns=["cluster"], inplace = True)
data_ICM_type = pd.read_csv(os.path.join("data","data_ICM_type.csv"), low_memory=False)
all_items = pd.concat([URM["ItemID"], data_ICM_type["item_id"]], ignore_index=True).unique()
n_users = len(URM["UserID"].unique())
n_items = len(all_items)
URM_scaled = sps.csr_matrix((URM["Data"].values,
(URM["UserID"].values, URM["ItemID"].values)),
shape = (n_users, n_items))
return URM_scaled
def get_df_from_urm(URM):
coo = URM.tocoo(copy=False)
df = pd.DataFrame({'UserID': coo.row, 'ItemID': coo.col, 'Data': coo.data}
)[['UserID', 'ItemID', 'Data']].sort_values(['UserID', 'ItemID']
).reset_index(drop=True)
return df
def get_clusters_dfs_from_df(df, k):
df["Data"] = df.groupby(["UserID", "ItemID"])["Data"].transform("sum")
df.drop_duplicates(inplace=True)
df_copy = df.copy()
df_copy.drop(columns=["ItemID"], inplace = True)
df_copy["Data"] = df_copy.groupby(["UserID"])["Data"].transform("sum")
df_copy.drop_duplicates(inplace=True)
df_copy.reset_index(inplace=True)
# Extract the data values
X = df_copy['Data'].values.reshape(-1, 1)
# Create a KMeans model with 2 clusters
kmeans = KMeans(n_clusters=k)
# Fit the model to the data
kmeans.fit(X)
# Add a new column to the DataFrame with the cluster labels
df_copy['cluster'] = kmeans.labels_
df_merged = pd.merge(df[["UserID","ItemID","Data"]], df_copy[["UserID","cluster"]], on='UserID')
# Create an empty list to store the separate datasets
clustered_data = []
# Iterate through each cluster
for i in range(k):
# Create a new dataframe for the current cluster
cluster_df = df_merged[df_merged['cluster'] == i]
# Save the dataframe to the list
clustered_data.append(cluster_df)
return clustered_data
def logistic(x):
return 1 / (1 + math.exp(-x))
def logistic_scale_implicit_rating(implicit_rating, alpha=1, min_rating=0, max_rating=5):
normalized_rating = implicit_rating/max_implicit_rating
return min_rating + (max_rating - min_rating) * logistic(1 * normalized_rating)
def tanh_scale_implicit_rating(implicit_rating, alpha=1, min_rating=0, max_rating=5):
return min_rating + (max_rating - min_rating) * math.tanh(alpha * implicit_rating)
def range_scaling(x, a=1, b=5):
max_x = np.max(x)
min_x = np.min(x)
return ((b - a)/(max_x - min_x)) * (x - min_x) + a
def ranged_min_max_scale_implicit_rating(implicit_rating):
return (4/(max_implicit_rating - min_implicit_rating))*(implicit_rating-min_implicit_rating)+1
def std_norm(x):
mean_x = np.mean(x)
stddev_x = np.std(x)
return (x - mean_x) / stddev_x
def clusterize(X, k):
kmeans = KMeans(n_clusters=k)
kmeans.fit(X)
return kmeans.labels_
def get_ucm():
urm_visualizations = urm_visualization_all_ones_summed()
urm_info = urm_info_all_ones_summed()
urm_train_vis, _ = split.split_train_in_two_percentage_global_sample(urm_visualizations,
train_percentage = 0.7,
seed=1234)
urm_train_info, _ = split.split_train_in_two_percentage_global_sample(urm_info,
train_percentage = 0.7,
seed=1234)
ucm = statistics_per_user(urm_train_vis, urm_train_info)
return ucm
def robust_norm(x):
x = x.reshape(-1, 1)
from sklearn.preprocessing import RobustScaler
transformer = RobustScaler().fit(x)
tmp = transformer.transform(x)
return tmp.reshape(1, -1)[0]
def get_data_global_sample(dataset_version, train_percentage = 0.70, setSeed=False, k=None, transformation=None, value_seen=None, value_info=None):
if setSeed == True:
seed = 1234
else:
seed = None
#warnings.filterwarnings('ignore')
URM_csr = urm_all_ones()
ICM = icm()
URM_train, URM_validation = split.split_train_in_two_percentage_global_sample(URM_csr,
train_percentage = train_percentage,
seed=seed)
if (dataset_version == "interactions-all-ones"):
return URM_csr, URM_train, URM_validation, ICM
elif (dataset_version == "stacked"):
URM_stacked = get_URM_stacked(URM_csr)
URM_stacked_train = get_URM_stacked(URM_train)
ICM_stacked = get_ICM_stacked(URM_train)
ICM_stacked_train = get_ICM_stacked(URM_train)
return URM_csr, URM_train, URM_validation, URM_stacked, URM_stacked_train, ICM_stacked, ICM_stacked_train
elif (dataset_version == "interactions-summed"):
return urm_all_ones_summed(), icm_types()
elif (dataset_version == "custom"):
assert value_seen is not None and value_info is not None, "value_seen or value_info is None!"
URM = urm_seen_or_info(value_seen, value_info)
URM_train, URM_validation = split.split_train_in_two_percentage_global_sample(URM,
train_percentage = train_percentage,
seed=seed)
return URM, URM_train, URM_validation, ICM
elif (dataset_version == "interactions-summed-transformed"):
assert transformation is not None, "transformation is None!"
URM_csr, URM_train, URM_validation = get_URM_all_ones_summed_transformed(train_percentage, k, seed, transformation)
return URM_csr, URM_train, URM_validation, ICM
elif (dataset_version == "visualizations-summed-transformed"):
assert transformation is not None, "transformation is None!"
URM_csr, URM_train, URM_validation = get_urm_visualization_summed_transformed(train_percentage, k, seed, transformation)
return URM_csr, URM_train, URM_validation, ICM
else:
print("Wrong dataset name. Try: \n - interactions-all-ones \n - stacked \n - interactions-summed \n - interactions-summed-transformed")
def get_data_user_wise(dataset_version, train_percentage = 0.70):
#warnings.filterwarnings('ignore')
URM_csr = urm_all_ones()
ICM = icm()
URM_train, URM_validation = split.split_train_in_two_percentage_user_wise(URM_csr, train_percentage = train_percentage)
if (dataset_version == "interactions-all-ones"):
return URM_csr, URM_train, URM_validation, ICM
elif (dataset_version == "stacked"):
URM_stacked = get_URM_stacked(URM_csr)
URM_train, URM_validation = split.split_train_in_two_percentage_user_wise(URM_stacked, train_percentage = train_percentage)
ICM_stacked = get_ICM_stacked(URM_csr)
return URM_stacked, URM_train, URM_validation, ICM_stacked
elif (dataset_version == "interactions-summed"):
return urm_all_ones_summed(), icm_types()
else:
print("Wrong dataset name. Try: \n - interactions-all-ones \n - stacked \n - interactions-summed")
def get_users_for_submission():
root_path = "data"
users = pd.read_csv(os.path.join(root_path, "data_target_users_test.csv"))
return users["user_id"]
def global_effects(URM_biased, shrink_user=5000, shrink_item=3000):
# 1) global average
mu = URM_biased.data.sum(dtype=np.float32) / URM_biased.nnz
URM_unbiased = URM_biased.copy()
URM_unbiased.data = URM_unbiased.data - mu
# 2) item average bias
# compute the number of non-zero elements for each column
col_nnz = np.ediff1d(sps.csc_matrix(URM_biased).indptr)
item_bias = URM_unbiased.sum(axis=0) / (col_nnz + shrink_item)
item_bias = np.asarray(item_bias).ravel() # converts 2-d matrix to 1-d array without anycopy
item_bias[col_nnz==0] = -np.inf
nz = URM_unbiased.nonzero()
URM_unbiased[nz] -= item_bias[nz[1]]
# 3) user average bias
# This computes the mean of the row excluding the missing values
row_nnz = np.ediff1d(sps.csr_matrix(URM_unbiased).indptr)
user_bias = URM_unbiased.sum(axis=1).ravel() / (row_nnz + shrink_user)
user_bias = np.asarray(user_bias).ravel()
user_bias[row_nnz==0] = -np.inf
# 4) summ all the contributes
URM_unbiased[nz] = user_bias[nz[0]] + item_bias[nz[1]] + mu
return URM_unbiased
def bayesian_search(recommender_class, recommender_input_args, hyperparameters_range_dictionary, evaluator_validation, dataset_version="interactions-all-ones", n_cases = 60, perc_random_starts = 0.3, metric_to_optimize = "MAP", cutoff_to_optimize = 10, cust_output_folder=None, block_size=None, resume_from_saved=False, ICM=None, ICM_name=None):
n_random_starts = int(n_cases * perc_random_starts)
output_folder_path = get_hyperparams_search_output_folder(recommender_class, dataset_version=dataset_version, custom_folder_name=cust_output_folder)
if recommender_class is TopPopRec:
recommender_input_args_local = recommender_input_args
urm = recommender_input_args_local.CONSTRUCTOR_POSITIONAL_ARGS[0]
hyperparameterSearch = SearchSingleCase(recommender_class, evaluator_validation=evaluator_validation)
hyperparameterSearch.search(recommender_input_args,
fit_hyperparameters_values={},
metric_to_optimize = metric_to_optimize,
cutoff_to_optimize = cutoff_to_optimize,
output_folder_path = output_folder_path,
output_file_name_root = recommender_class.RECOMMENDER_NAME,
resume_from_saved = resume_from_saved,
save_model = "best",
)
elif recommender_class is ItemKNNCBFRec or recommender_class is ItemKNNCFRec or recommender_class is UserKNNCFRec:
if recommender_class is ItemKNNCFRec:
recommender_class = ItemKNNCFRecommender
knn_cf = True
elif recommender_class is UserKNNCFRec:
recommender_class = UserKNNCFRecommender
knn_cf = True
elif recommender_class is ItemKNNCBFRec:
recommender_class = ItemKNNCBFRecommender
knn_cbf = True
else:
knn_cf = False
if knn_cf: #if knn_cf:
recommender_input_args_local = recommender_input_args
urm = recommender_input_args_local.CONSTRUCTOR_POSITIONAL_ARGS[0] # the URM
runHyperparameterSearch_Collaborative(recommender_class,
urm,
n_cases = n_cases,
n_random_starts = n_random_starts,
resume_from_saved = resume_from_saved,
evaluator_validation = evaluator_validation,
metric_to_optimize = metric_to_optimize,
cutoff_to_optimize = cutoff_to_optimize,
output_folder_path = output_folder_path)
elif knn_cbf:
recommender_input_args_local = recommender_input_args
urm = recommender_input_args_local.CONSTRUCTOR_POSITIONAL_ARGS[0] # the URM
runHyperparameterSearch_Content(recommender_class,
urm,
ICM_object = ICM,
ICM_name = ICM_name,
n_cases = n_cases,
n_random_starts = n_random_starts,
resume_from_saved = resume_from_saved,
evaluator_validation = evaluator_validation,
metric_to_optimize = metric_to_optimize,
cutoff_to_optimize = cutoff_to_optimize,
output_folder_path = output_folder_path,
parallelizeKNN = True,
allow_weighting = True,
allow_bias_ICM = True)
else:
hyperparameterSearch = SearchBayesianSkopt(recommender_class, evaluator_validation=evaluator_validation, block_size = block_size)
hyperparameterSearch._set_skopt_params(n_jobs=-1)
hyperparameterSearch.search(recommender_input_args,
hyperparameter_search_space = hyperparameters_range_dictionary,
n_cases = n_cases,
n_random_starts = n_random_starts,
save_model = "best",
resume_from_saved = resume_from_saved,
output_folder_path = output_folder_path, # Where to save the results
output_file_name_root = recommender_class.RECOMMENDER_NAME, # How to call the files
metric_to_optimize = metric_to_optimize,
cutoff_to_optimize = cutoff_to_optimize)
def optimization_terminated(recommender, dataset_version, override = False):
recommendations_folder_root = "recommendations"
recommendations_folder_root = os.path.join(recommendations_folder_root, dataset_version)
recommendations_folder_root = os.path.join(recommendations_folder_root, recommender.RECOMMENDER_NAME)
recommendations_folder = os.path.join(recommendations_folder_root, recommender.RECOMMENDER_VERSION)
if ((not os.path.exists(recommendations_folder))):
try:
os.makedirs(recommendations_folder)
except OSError:
recommendations_folder = os.path.join(recommendations_folder_root, recommender.RECOMMENDER_VERSION[:255])
os.makedirs(recommendations_folder)
if (len(os.listdir(recommendations_folder)) == 0) or override:
hyperparam_search_folder = os.path.join(recommendations_folder_root, "hyperparams_search")
if os.path.exists(hyperparam_search_folder):
train_folder = os.path.join(recommendations_folder, "optimization")
if not os.path.exists(train_folder):
os.makedirs(train_folder)
copy_all_files(hyperparam_search_folder, train_folder, remove_source_folder=False)
else:
print("Error! It already exists the folder " + recommendations_folder)
def submission(recommender, dataset_version, override = False):
optimization_terminated(recommender, dataset_version, override = override)
recommendations_folder_root = "recommendations"
recommendations_folder_root = os.path.join(recommendations_folder_root, dataset_version)
recommendations_folder_root = os.path.join(recommendations_folder_root, recommender.RECOMMENDER_NAME)
recommendations_folder_version = get_folder_best_model(recommender.__class__, dataset_version)
rec_file_path = os.path.join(recommendations_folder_version, "recommendations.csv")
if (os.path.exists(rec_file_path) and override) or (not os.path.exists(rec_file_path)):
users = get_users_for_submission()
tmp = recommender.recommend(user_id_array=users, cutoff=10)
well_formatted = []
for i in tmp:
well_formatted.append( " ".join([str(x) for x in i]))
submission = pd.DataFrame()
submission["user_id"] = users
submission["item_list"] = well_formatted
submission.to_csv(rec_file_path, index=False)
hyperparam_search_folder = os.path.join(recommendations_folder_root, "hyperparams_search")
if os.path.exists(hyperparam_search_folder):
shutil.rmtree(hyperparam_search_folder)
submission_folder = os.path.join(recommendations_folder_version, "best")
if not os.path.exists(submission_folder):
os.makedirs(submission_folder)
recommender.save_model(folder_path=submission_folder)
else:
print("Error! It already exists the file ", rec_file_path)
def hybrid_submission(recommender, dataset_version, override = False):
optimization_terminated(recommender, dataset_version, override = override)
recommendations_folder_root = "recommendations"
recommendations_folder_root = os.path.join(recommendations_folder_root, dataset_version)
recommendations_folder_root = os.path.join(recommendations_folder_root, recommender.RECOMMENDER_NAME)
recommendations_folder_version = os.path.join(recommendations_folder_root, recommender.RECOMMENDER_VERSION)
rec_file_path = os.path.join(recommendations_folder_version, "recommendations.csv")
if (os.path.exists(rec_file_path) and override) or (not os.path.exists(rec_file_path)):
users = get_users_for_submission()
tmp = recommender.recommend(user_id_array=users, cutoff=10)
well_formatted = []
for i in tmp:
well_formatted.append( " ".join([str(x) for x in i]))
submission = pd.DataFrame()
submission["user_id"] = users
submission["item_list"] = well_formatted
submission.to_csv(rec_file_path, index=False)
else:
print("Error! It already exists the file ", rec_file_path)
def get_hyperparams_search_output_folder(recommender_class, dataset_version="interactions-all-ones", custom_folder_name=None):
folder = "recommendations"
output_folder_path = os.path.join(folder, dataset_version)
output_folder_path = os.path.join(output_folder_path, recommender_class.RECOMMENDER_NAME)
if custom_folder_name != None:
hyper_search_folder = custom_folder_name
else:
hyper_search_folder = "hyperparams_search"
output_folder_path = os.path.join(output_folder_path, hyper_search_folder)
# If directory does not exist, create
if not os.path.exists(output_folder_path):
os.makedirs(output_folder_path)
return output_folder_path + "/"
def choose_num_train_epochs(recommender, validation_evaluator, max_epochs=10, valid_every_n = 1, metric = "MAP", cutoff=10, **fit_args):
validation_results = pd.DataFrame(columns=["epoch", metric])
for i in range(int(max_epochs/valid_every_n)):
recommender.fit(epochs=valid_every_n, **fit_args)
results_df, result_string = validation_evaluator.evaluateRecommender(recommender)
print(result_string)
validation_results.loc[i] = [int((i+1) * valid_every_n), results_df.iloc[0][metric]]
maxidx = validation_results.idxmax()
return validation_results.iat[maxidx[metric], 0], validation_results
def get_best_model_hyperparameters(recommender_class, dataset_version="interactions-all-ones", optimization=True, metric="MAP", custom_folder_name = None):
folder = "recommendations"
## During Bayesian Search
if optimization:
folder = os.path.join(folder, dataset_version)
folder = os.path.join(folder, recommender_class.RECOMMENDER_NAME)
if custom_folder_name == None:
hyp_search_folder = "hyperparams_search"
else:
hyp_search_folder = custom_folder_name
folder = os.path.join(folder, hyp_search_folder)
## After Bayesian Search
else:
folder = get_folder_best_model(recommender_class, dataset_version)
folder = os.path.join(folder, "optimization")
data_loader = DataIO(folder_path = folder)
hyperparams_file = recommender_class.RECOMMENDER_NAME + "_metadata.zip"
if os.path.exists(os.path.join(folder, hyperparams_file)):
search_metadata = data_loader.load_data(hyperparams_file)
return search_metadata["hyperparameters_best"]
else:
return {}
def get_best_res_on_validation(recommender_class, dataset_version="interactions-all-ones", optimization=False, metric="MAP", custom_folder_name = None):
folder = "recommendations"
## During Bayesian Search
if optimization:
folder = os.path.join("recommendations", dataset_version)
folder = os.path.join(folder, recommender_class.RECOMMENDER_NAME)
if custom_folder_name == None:
hyp_search_folder = "hyperparams_search"
else:
hyp_search_folder = custom_folder_name
folder = os.path.join(folder, hyp_search_folder)
## After Bayesian Search
else:
folder = get_folder_best_model(recommender_class, dataset_version)
folder = os.path.join(folder, "optimization")
data_loader = DataIO(folder_path = folder)
hyperparams_file = recommender_class.RECOMMENDER_NAME + "_metadata.zip"
if os.path.exists(os.path.join(folder, hyperparams_file)):
search_metadata = data_loader.load_data(hyperparams_file)
return search_metadata["result_on_validation_best"][metric]
else:
return {}
####################### AFTER HYPERPARAMS SEARCH OF "recommender_class"
def get_folder_best_model(recommender_class, dataset_version="interactions-all-ones"):
folder = "recommendations"
folder = os.path.join(folder, dataset_version)
folder = os.path.join(folder, recommender_class.RECOMMENDER_NAME)
list_dir_no_hid = listdir_nohidden(folder)
if (len(list_dir_no_hid) == 1 and list_dir_no_hid[0] != "hyperparams_search"):
return os.path.join(folder, list_dir_no_hid[0])
if (len(list_dir_no_hid) >= 2):
i = 0
while i < len(list_dir_no_hid):
if list_dir_no_hid[i][:18] != "hyperparams_search":
return os.path.join(folder, list_dir_no_hid[i])
i += 1
print("Error: not present best model folder")
def save_item_scores(recommender_class, URM, user_id_array, dataset_version, fast=True, on_validation=True, new_item_scores_file_name_root=None):
'''on_validation must be true if URM is the URM_train (so the URM after splitting)
'''
folder = get_folder_best_model(recommender_class, dataset_version)
if new_item_scores_file_name_root is None:
file_name = ""
else:
file_name = new_item_scores_file_name_root
file_name += "item_scores"
scores_file = os.path.join(folder, file_name + ".npy")
if not os.path.exists(scores_file) or recommender_class is DiffStructHybridRecommender:
assert (URM is not None) and (user_id_array is not None)
kwargs = {}
if recommender_class is DiffStructHybridRecommender:
kwargs = {"load_scores_from_saved": True,
"recs_on_urm_splitted": on_validation,
"user_id_array_val": user_id_array,
"new_item_scores_file_name_root": new_item_scores_file_name_root}
recommender = load_best_model(URM,
recommender_class,
dataset_version=dataset_version,
optimization=on_validation,
**kwargs)
item_scores = recommender._compute_item_score(user_id_array)
recommender._print("Saving item_score in file '{}'".format(scores_file))
if fast:
np.save(scores_file, item_scores, allow_pickle=False)
else:
data_dict_to_save = {"item_scores": item_scores}
dataIO = DataIO(folder_path=folder)
dataIO.save_data(file_name=file_name, data_dict_to_save = data_dict_to_save)
recommender._print("Saving complete")
def load_item_scores(recommender_class, dataset_version, fast = True, new_item_scores_file_name_root=None):
folder = get_folder_best_model(recommender_class, dataset_version)
if new_item_scores_file_name_root == None:
new_item_scores_file_name_root = ""
file_name = new_item_scores_file_name_root + "item_scores"
if not fast:
dataIO = DataIO(folder_path=folder)
return dataIO.load_data(file_name=file_name)
else:
return np.load(os.path.join(folder, file_name + ".npy"), allow_pickle=False)
def fit_best_recommender(recommender_class, URM, dataset_version, optimization=True, **kwargs):
best_hyperparameters = get_best_model_hyperparameters(recommender_class, dataset_version, optimization)
recommender = recommender_class(*get_kwargs_constructor(recommender_class, URM, dataset_version, optimization), **kwargs)
recommender.fit(**best_hyperparameters)
return recommender
def load_best_model(URM, rec_class, dataset_version="interactions-all-ones", optimization=False, **kwargs):
rec = rec_class(*get_kwargs_constructor(rec_class, URM, dataset_version, optimization), **kwargs)
if optimization: