File size: 42,602 Bytes
7885a28
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
import sys

import numpy as np
import pytest
from numpy.testing import assert_allclose, assert_array_equal
from scipy import sparse
from scipy.interpolate import BSpline
from scipy.sparse import random as sparse_random

from sklearn.linear_model import LinearRegression
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import (
    KBinsDiscretizer,
    PolynomialFeatures,
    SplineTransformer,
)
from sklearn.preprocessing._csr_polynomial_expansion import (
    _calc_expanded_nnz,
    _calc_total_nnz,
    _get_sizeof_LARGEST_INT_t,
)
from sklearn.utils._testing import assert_array_almost_equal
from sklearn.utils.fixes import (
    CSC_CONTAINERS,
    CSR_CONTAINERS,
    parse_version,
    sp_version,
)


@pytest.mark.parametrize("est", (PolynomialFeatures, SplineTransformer))
def test_polynomial_and_spline_array_order(est):
    """Test that output array has the given order."""
    X = np.arange(10).reshape(5, 2)

    def is_c_contiguous(a):
        return np.isfortran(a.T)

    assert is_c_contiguous(est().fit_transform(X))
    assert is_c_contiguous(est(order="C").fit_transform(X))
    assert np.isfortran(est(order="F").fit_transform(X))


@pytest.mark.parametrize(
    "params, err_msg",
    [
        ({"knots": [[1]]}, r"Number of knots, knots.shape\[0\], must be >= 2."),
        ({"knots": [[1, 1], [2, 2]]}, r"knots.shape\[1\] == n_features is violated"),
        ({"knots": [[1], [0]]}, "knots must be sorted without duplicates."),
    ],
)
def test_spline_transformer_input_validation(params, err_msg):
    """Test that we raise errors for invalid input in SplineTransformer."""
    X = [[1], [2]]

    with pytest.raises(ValueError, match=err_msg):
        SplineTransformer(**params).fit(X)


@pytest.mark.parametrize("extrapolation", ["continue", "periodic"])
def test_spline_transformer_integer_knots(extrapolation):
    """Test that SplineTransformer accepts integer value knot positions."""
    X = np.arange(20).reshape(10, 2)
    knots = [[0, 1], [1, 2], [5, 5], [11, 10], [12, 11]]
    _ = SplineTransformer(
        degree=3, knots=knots, extrapolation=extrapolation
    ).fit_transform(X)


def test_spline_transformer_feature_names():
    """Test that SplineTransformer generates correct features name."""
    X = np.arange(20).reshape(10, 2)
    splt = SplineTransformer(n_knots=3, degree=3, include_bias=True).fit(X)
    feature_names = splt.get_feature_names_out()
    assert_array_equal(
        feature_names,
        [
            "x0_sp_0",
            "x0_sp_1",
            "x0_sp_2",
            "x0_sp_3",
            "x0_sp_4",
            "x1_sp_0",
            "x1_sp_1",
            "x1_sp_2",
            "x1_sp_3",
            "x1_sp_4",
        ],
    )

    splt = SplineTransformer(n_knots=3, degree=3, include_bias=False).fit(X)
    feature_names = splt.get_feature_names_out(["a", "b"])
    assert_array_equal(
        feature_names,
        [
            "a_sp_0",
            "a_sp_1",
            "a_sp_2",
            "a_sp_3",
            "b_sp_0",
            "b_sp_1",
            "b_sp_2",
            "b_sp_3",
        ],
    )


@pytest.mark.parametrize(
    "extrapolation",
    ["constant", "linear", "continue", "periodic"],
)
@pytest.mark.parametrize("degree", [2, 3])
def test_split_transform_feature_names_extrapolation_degree(extrapolation, degree):
    """Test feature names are correct for different extrapolations and degree.

    Non-regression test for gh-25292.
    """
    X = np.arange(20).reshape(10, 2)
    splt = SplineTransformer(degree=degree, extrapolation=extrapolation).fit(X)
    feature_names = splt.get_feature_names_out(["a", "b"])
    assert len(feature_names) == splt.n_features_out_

    X_trans = splt.transform(X)
    assert X_trans.shape[1] == len(feature_names)


@pytest.mark.parametrize("degree", range(1, 5))
@pytest.mark.parametrize("n_knots", range(3, 5))
@pytest.mark.parametrize("knots", ["uniform", "quantile"])
@pytest.mark.parametrize("extrapolation", ["constant", "periodic"])
def test_spline_transformer_unity_decomposition(degree, n_knots, knots, extrapolation):
    """Test that B-splines are indeed a decomposition of unity.

    Splines basis functions must sum up to 1 per row, if we stay in between boundaries.
    """
    X = np.linspace(0, 1, 100)[:, None]
    # make the boundaries 0 and 1 part of X_train, for sure.
    X_train = np.r_[[[0]], X[::2, :], [[1]]]
    X_test = X[1::2, :]

    if extrapolation == "periodic":
        n_knots = n_knots + degree  # periodic splines require degree < n_knots

    splt = SplineTransformer(
        n_knots=n_knots,
        degree=degree,
        knots=knots,
        include_bias=True,
        extrapolation=extrapolation,
    )
    splt.fit(X_train)
    for X in [X_train, X_test]:
        assert_allclose(np.sum(splt.transform(X), axis=1), 1)


@pytest.mark.parametrize(["bias", "intercept"], [(True, False), (False, True)])
def test_spline_transformer_linear_regression(bias, intercept):
    """Test that B-splines fit a sinusodial curve pretty well."""
    X = np.linspace(0, 10, 100)[:, None]
    y = np.sin(X[:, 0]) + 2  # +2 to avoid the value 0 in assert_allclose
    pipe = Pipeline(
        steps=[
            (
                "spline",
                SplineTransformer(
                    n_knots=15,
                    degree=3,
                    include_bias=bias,
                    extrapolation="constant",
                ),
            ),
            ("ols", LinearRegression(fit_intercept=intercept)),
        ]
    )
    pipe.fit(X, y)
    assert_allclose(pipe.predict(X), y, rtol=1e-3)


@pytest.mark.parametrize(
    ["knots", "n_knots", "sample_weight", "expected_knots"],
    [
        ("uniform", 3, None, np.array([[0, 2], [3, 8], [6, 14]])),
        (
            "uniform",
            3,
            np.array([0, 0, 1, 1, 0, 3, 1]),
            np.array([[2, 2], [4, 8], [6, 14]]),
        ),
        ("uniform", 4, None, np.array([[0, 2], [2, 6], [4, 10], [6, 14]])),
        ("quantile", 3, None, np.array([[0, 2], [3, 3], [6, 14]])),
        (
            "quantile",
            3,
            np.array([0, 0, 1, 1, 0, 3, 1]),
            np.array([[2, 2], [5, 8], [6, 14]]),
        ),
    ],
)
def test_spline_transformer_get_base_knot_positions(
    knots, n_knots, sample_weight, expected_knots
):
    """Check the behaviour to find knot positions with and without sample_weight."""
    X = np.array([[0, 2], [0, 2], [2, 2], [3, 3], [4, 6], [5, 8], [6, 14]])
    base_knots = SplineTransformer._get_base_knot_positions(
        X=X, knots=knots, n_knots=n_knots, sample_weight=sample_weight
    )
    assert_allclose(base_knots, expected_knots)


@pytest.mark.parametrize(["bias", "intercept"], [(True, False), (False, True)])
def test_spline_transformer_periodic_linear_regression(bias, intercept):
    """Test that B-splines fit a periodic curve pretty well."""

    # "+ 3" to avoid the value 0 in assert_allclose
    def f(x):
        return np.sin(2 * np.pi * x) - np.sin(8 * np.pi * x) + 3

    X = np.linspace(0, 1, 101)[:, None]
    pipe = Pipeline(
        steps=[
            (
                "spline",
                SplineTransformer(
                    n_knots=20,
                    degree=3,
                    include_bias=bias,
                    extrapolation="periodic",
                ),
            ),
            ("ols", LinearRegression(fit_intercept=intercept)),
        ]
    )
    pipe.fit(X, f(X[:, 0]))

    # Generate larger array to check periodic extrapolation
    X_ = np.linspace(-1, 2, 301)[:, None]
    predictions = pipe.predict(X_)
    assert_allclose(predictions, f(X_[:, 0]), atol=0.01, rtol=0.01)
    assert_allclose(predictions[0:100], predictions[100:200], rtol=1e-3)


def test_spline_transformer_periodic_spline_backport():
    """Test that the backport of extrapolate="periodic" works correctly"""
    X = np.linspace(-2, 3.5, 10)[:, None]
    degree = 2

    # Use periodic extrapolation backport in SplineTransformer
    transformer = SplineTransformer(
        degree=degree, extrapolation="periodic", knots=[[-1.0], [0.0], [1.0]]
    )
    Xt = transformer.fit_transform(X)

    # Use periodic extrapolation in BSpline
    coef = np.array([[1.0, 0.0], [0.0, 1.0], [1.0, 0.0], [0.0, 1.0]])
    spl = BSpline(np.arange(-3, 4), coef, degree, "periodic")
    Xspl = spl(X[:, 0])
    assert_allclose(Xt, Xspl)


def test_spline_transformer_periodic_splines_periodicity():
    """Test if shifted knots result in the same transformation up to permutation."""
    X = np.linspace(0, 10, 101)[:, None]

    transformer_1 = SplineTransformer(
        degree=3,
        extrapolation="periodic",
        knots=[[0.0], [1.0], [3.0], [4.0], [5.0], [8.0]],
    )

    transformer_2 = SplineTransformer(
        degree=3,
        extrapolation="periodic",
        knots=[[1.0], [3.0], [4.0], [5.0], [8.0], [9.0]],
    )

    Xt_1 = transformer_1.fit_transform(X)
    Xt_2 = transformer_2.fit_transform(X)

    assert_allclose(Xt_1, Xt_2[:, [4, 0, 1, 2, 3]])


@pytest.mark.parametrize("degree", [3, 5])
def test_spline_transformer_periodic_splines_smoothness(degree):
    """Test that spline transformation is smooth at first / last knot."""
    X = np.linspace(-2, 10, 10_000)[:, None]

    transformer = SplineTransformer(
        degree=degree,
        extrapolation="periodic",
        knots=[[0.0], [1.0], [3.0], [4.0], [5.0], [8.0]],
    )
    Xt = transformer.fit_transform(X)

    delta = (X.max() - X.min()) / len(X)
    tol = 10 * delta

    dXt = Xt
    # We expect splines of degree `degree` to be (`degree`-1) times
    # continuously differentiable. I.e. for d = 0, ..., `degree` - 1 the d-th
    # derivative should be continuous. This is the case if the (d+1)-th
    # numerical derivative is reasonably small (smaller than `tol` in absolute
    # value). We thus compute d-th numeric derivatives for d = 1, ..., `degree`
    # and compare them to `tol`.
    #
    # Note that the 0-th derivative is the function itself, such that we are
    # also checking its continuity.
    for d in range(1, degree + 1):
        # Check continuity of the (d-1)-th derivative
        diff = np.diff(dXt, axis=0)
        assert np.abs(diff).max() < tol
        # Compute d-th numeric derivative
        dXt = diff / delta

    # As degree `degree` splines are not `degree` times continuously
    # differentiable at the knots, the `degree + 1`-th numeric derivative
    # should have spikes at the knots.
    diff = np.diff(dXt, axis=0)
    assert np.abs(diff).max() > 1


@pytest.mark.parametrize(["bias", "intercept"], [(True, False), (False, True)])
@pytest.mark.parametrize("degree", [1, 2, 3, 4, 5])
def test_spline_transformer_extrapolation(bias, intercept, degree):
    """Test that B-spline extrapolation works correctly."""
    # we use a straight line for that
    X = np.linspace(-1, 1, 100)[:, None]
    y = X.squeeze()

    # 'constant'
    pipe = Pipeline(
        [
            [
                "spline",
                SplineTransformer(
                    n_knots=4,
                    degree=degree,
                    include_bias=bias,
                    extrapolation="constant",
                ),
            ],
            ["ols", LinearRegression(fit_intercept=intercept)],
        ]
    )
    pipe.fit(X, y)
    assert_allclose(pipe.predict([[-10], [5]]), [-1, 1])

    # 'linear'
    pipe = Pipeline(
        [
            [
                "spline",
                SplineTransformer(
                    n_knots=4,
                    degree=degree,
                    include_bias=bias,
                    extrapolation="linear",
                ),
            ],
            ["ols", LinearRegression(fit_intercept=intercept)],
        ]
    )
    pipe.fit(X, y)
    assert_allclose(pipe.predict([[-10], [5]]), [-10, 5])

    # 'error'
    splt = SplineTransformer(
        n_knots=4, degree=degree, include_bias=bias, extrapolation="error"
    )
    splt.fit(X)
    msg = "X contains values beyond the limits of the knots"
    with pytest.raises(ValueError, match=msg):
        splt.transform([[-10]])
    with pytest.raises(ValueError, match=msg):
        splt.transform([[5]])


def test_spline_transformer_kbindiscretizer():
    """Test that a B-spline of degree=0 is equivalent to KBinsDiscretizer."""
    rng = np.random.RandomState(97531)
    X = rng.randn(200).reshape(200, 1)
    n_bins = 5
    n_knots = n_bins + 1

    splt = SplineTransformer(
        n_knots=n_knots, degree=0, knots="quantile", include_bias=True
    )
    splines = splt.fit_transform(X)

    kbd = KBinsDiscretizer(n_bins=n_bins, encode="onehot-dense", strategy="quantile")
    kbins = kbd.fit_transform(X)

    # Though they should be exactly equal, we test approximately with high
    # accuracy.
    assert_allclose(splines, kbins, rtol=1e-13)


@pytest.mark.skipif(
    sp_version < parse_version("1.8.0"),
    reason="The option `sparse_output` is available as of scipy 1.8.0",
)
@pytest.mark.parametrize("degree", range(1, 3))
@pytest.mark.parametrize("knots", ["uniform", "quantile"])
@pytest.mark.parametrize(
    "extrapolation", ["error", "constant", "linear", "continue", "periodic"]
)
@pytest.mark.parametrize("include_bias", [False, True])
def test_spline_transformer_sparse_output(
    degree, knots, extrapolation, include_bias, global_random_seed
):
    rng = np.random.RandomState(global_random_seed)
    X = rng.randn(200).reshape(40, 5)

    splt_dense = SplineTransformer(
        degree=degree,
        knots=knots,
        extrapolation=extrapolation,
        include_bias=include_bias,
        sparse_output=False,
    )
    splt_sparse = SplineTransformer(
        degree=degree,
        knots=knots,
        extrapolation=extrapolation,
        include_bias=include_bias,
        sparse_output=True,
    )

    splt_dense.fit(X)
    splt_sparse.fit(X)

    X_trans_sparse = splt_sparse.transform(X)
    X_trans_dense = splt_dense.transform(X)
    assert sparse.issparse(X_trans_sparse) and X_trans_sparse.format == "csr"
    assert_allclose(X_trans_dense, X_trans_sparse.toarray())

    # extrapolation regime
    X_min = np.amin(X, axis=0)
    X_max = np.amax(X, axis=0)
    X_extra = np.r_[
        np.linspace(X_min - 5, X_min, 10), np.linspace(X_max, X_max + 5, 10)
    ]
    if extrapolation == "error":
        msg = "X contains values beyond the limits of the knots"
        with pytest.raises(ValueError, match=msg):
            splt_dense.transform(X_extra)
        msg = "Out of bounds"
        with pytest.raises(ValueError, match=msg):
            splt_sparse.transform(X_extra)
    else:
        assert_allclose(
            splt_dense.transform(X_extra), splt_sparse.transform(X_extra).toarray()
        )


@pytest.mark.skipif(
    sp_version >= parse_version("1.8.0"),
    reason="The option `sparse_output` is available as of scipy 1.8.0",
)
def test_spline_transformer_sparse_output_raise_error_for_old_scipy():
    """Test that SplineTransformer with sparse=True raises for scipy<1.8.0."""
    X = [[1], [2]]
    with pytest.raises(ValueError, match="scipy>=1.8.0"):
        SplineTransformer(sparse_output=True).fit(X)


@pytest.mark.parametrize("n_knots", [5, 10])
@pytest.mark.parametrize("include_bias", [True, False])
@pytest.mark.parametrize("degree", [3, 4])
@pytest.mark.parametrize(
    "extrapolation", ["error", "constant", "linear", "continue", "periodic"]
)
@pytest.mark.parametrize("sparse_output", [False, True])
def test_spline_transformer_n_features_out(
    n_knots, include_bias, degree, extrapolation, sparse_output
):
    """Test that transform results in n_features_out_ features."""
    if sparse_output and sp_version < parse_version("1.8.0"):
        pytest.skip("The option `sparse_output` is available as of scipy 1.8.0")

    splt = SplineTransformer(
        n_knots=n_knots,
        degree=degree,
        include_bias=include_bias,
        extrapolation=extrapolation,
        sparse_output=sparse_output,
    )
    X = np.linspace(0, 1, 10)[:, None]
    splt.fit(X)

    assert splt.transform(X).shape[1] == splt.n_features_out_


@pytest.mark.parametrize(
    "params, err_msg",
    [
        ({"degree": (-1, 2)}, r"degree=\(min_degree, max_degree\) must"),
        ({"degree": (0, 1.5)}, r"degree=\(min_degree, max_degree\) must"),
        ({"degree": (3, 2)}, r"degree=\(min_degree, max_degree\) must"),
        ({"degree": (1, 2, 3)}, r"int or tuple \(min_degree, max_degree\)"),
    ],
)
def test_polynomial_features_input_validation(params, err_msg):
    """Test that we raise errors for invalid input in PolynomialFeatures."""
    X = [[1], [2]]

    with pytest.raises(ValueError, match=err_msg):
        PolynomialFeatures(**params).fit(X)


@pytest.fixture()
def single_feature_degree3():
    X = np.arange(6)[:, np.newaxis]
    P = np.hstack([np.ones_like(X), X, X**2, X**3])
    return X, P


@pytest.mark.parametrize(
    "degree, include_bias, interaction_only, indices",
    [
        (3, True, False, slice(None, None)),
        (3, False, False, slice(1, None)),
        (3, True, True, [0, 1]),
        (3, False, True, [1]),
        ((2, 3), True, False, [0, 2, 3]),
        ((2, 3), False, False, [2, 3]),
        ((2, 3), True, True, [0]),
        ((2, 3), False, True, []),
    ],
)
@pytest.mark.parametrize("X_container", [None] + CSR_CONTAINERS + CSC_CONTAINERS)
def test_polynomial_features_one_feature(
    single_feature_degree3,
    degree,
    include_bias,
    interaction_only,
    indices,
    X_container,
):
    """Test PolynomialFeatures on single feature up to degree 3."""
    X, P = single_feature_degree3
    if X_container is not None:
        X = X_container(X)
    tf = PolynomialFeatures(
        degree=degree, include_bias=include_bias, interaction_only=interaction_only
    ).fit(X)
    out = tf.transform(X)
    if X_container is not None:
        out = out.toarray()
    assert_allclose(out, P[:, indices])
    if tf.n_output_features_ > 0:
        assert tf.powers_.shape == (tf.n_output_features_, tf.n_features_in_)


@pytest.fixture()
def two_features_degree3():
    X = np.arange(6).reshape((3, 2))
    x1 = X[:, :1]
    x2 = X[:, 1:]
    P = np.hstack(
        [
            x1**0 * x2**0,  # 0
            x1**1 * x2**0,  # 1
            x1**0 * x2**1,  # 2
            x1**2 * x2**0,  # 3
            x1**1 * x2**1,  # 4
            x1**0 * x2**2,  # 5
            x1**3 * x2**0,  # 6
            x1**2 * x2**1,  # 7
            x1**1 * x2**2,  # 8
            x1**0 * x2**3,  # 9
        ]
    )
    return X, P


@pytest.mark.parametrize(
    "degree, include_bias, interaction_only, indices",
    [
        (2, True, False, slice(0, 6)),
        (2, False, False, slice(1, 6)),
        (2, True, True, [0, 1, 2, 4]),
        (2, False, True, [1, 2, 4]),
        ((2, 2), True, False, [0, 3, 4, 5]),
        ((2, 2), False, False, [3, 4, 5]),
        ((2, 2), True, True, [0, 4]),
        ((2, 2), False, True, [4]),
        (3, True, False, slice(None, None)),
        (3, False, False, slice(1, None)),
        (3, True, True, [0, 1, 2, 4]),
        (3, False, True, [1, 2, 4]),
        ((2, 3), True, False, [0, 3, 4, 5, 6, 7, 8, 9]),
        ((2, 3), False, False, slice(3, None)),
        ((2, 3), True, True, [0, 4]),
        ((2, 3), False, True, [4]),
        ((3, 3), True, False, [0, 6, 7, 8, 9]),
        ((3, 3), False, False, [6, 7, 8, 9]),
        ((3, 3), True, True, [0]),
        ((3, 3), False, True, []),  # would need 3 input features
    ],
)
@pytest.mark.parametrize("X_container", [None] + CSR_CONTAINERS + CSC_CONTAINERS)
def test_polynomial_features_two_features(
    two_features_degree3,
    degree,
    include_bias,
    interaction_only,
    indices,
    X_container,
):
    """Test PolynomialFeatures on 2 features up to degree 3."""
    X, P = two_features_degree3
    if X_container is not None:
        X = X_container(X)
    tf = PolynomialFeatures(
        degree=degree, include_bias=include_bias, interaction_only=interaction_only
    ).fit(X)
    out = tf.transform(X)
    if X_container is not None:
        out = out.toarray()
    assert_allclose(out, P[:, indices])
    if tf.n_output_features_ > 0:
        assert tf.powers_.shape == (tf.n_output_features_, tf.n_features_in_)


def test_polynomial_feature_names():
    X = np.arange(30).reshape(10, 3)
    poly = PolynomialFeatures(degree=2, include_bias=True).fit(X)
    feature_names = poly.get_feature_names_out()
    assert_array_equal(
        ["1", "x0", "x1", "x2", "x0^2", "x0 x1", "x0 x2", "x1^2", "x1 x2", "x2^2"],
        feature_names,
    )
    assert len(feature_names) == poly.transform(X).shape[1]

    poly = PolynomialFeatures(degree=3, include_bias=False).fit(X)
    feature_names = poly.get_feature_names_out(["a", "b", "c"])
    assert_array_equal(
        [
            "a",
            "b",
            "c",
            "a^2",
            "a b",
            "a c",
            "b^2",
            "b c",
            "c^2",
            "a^3",
            "a^2 b",
            "a^2 c",
            "a b^2",
            "a b c",
            "a c^2",
            "b^3",
            "b^2 c",
            "b c^2",
            "c^3",
        ],
        feature_names,
    )
    assert len(feature_names) == poly.transform(X).shape[1]

    poly = PolynomialFeatures(degree=(2, 3), include_bias=False).fit(X)
    feature_names = poly.get_feature_names_out(["a", "b", "c"])
    assert_array_equal(
        [
            "a^2",
            "a b",
            "a c",
            "b^2",
            "b c",
            "c^2",
            "a^3",
            "a^2 b",
            "a^2 c",
            "a b^2",
            "a b c",
            "a c^2",
            "b^3",
            "b^2 c",
            "b c^2",
            "c^3",
        ],
        feature_names,
    )
    assert len(feature_names) == poly.transform(X).shape[1]

    poly = PolynomialFeatures(
        degree=(3, 3), include_bias=True, interaction_only=True
    ).fit(X)
    feature_names = poly.get_feature_names_out(["a", "b", "c"])
    assert_array_equal(["1", "a b c"], feature_names)
    assert len(feature_names) == poly.transform(X).shape[1]

    # test some unicode
    poly = PolynomialFeatures(degree=1, include_bias=True).fit(X)
    feature_names = poly.get_feature_names_out(["\u0001F40D", "\u262e", "\u05d0"])
    assert_array_equal(["1", "\u0001F40D", "\u262e", "\u05d0"], feature_names)


@pytest.mark.parametrize(
    ["deg", "include_bias", "interaction_only", "dtype"],
    [
        (1, True, False, int),
        (2, True, False, int),
        (2, True, False, np.float32),
        (2, True, False, np.float64),
        (3, False, False, np.float64),
        (3, False, True, np.float64),
        (4, False, False, np.float64),
        (4, False, True, np.float64),
    ],
)
@pytest.mark.parametrize("csc_container", CSC_CONTAINERS)
def test_polynomial_features_csc_X(
    deg, include_bias, interaction_only, dtype, csc_container
):
    rng = np.random.RandomState(0)
    X = rng.randint(0, 2, (100, 2))
    X_csc = csc_container(X)

    est = PolynomialFeatures(
        deg, include_bias=include_bias, interaction_only=interaction_only
    )
    Xt_csc = est.fit_transform(X_csc.astype(dtype))
    Xt_dense = est.fit_transform(X.astype(dtype))

    assert sparse.issparse(Xt_csc) and Xt_csc.format == "csc"
    assert Xt_csc.dtype == Xt_dense.dtype
    assert_array_almost_equal(Xt_csc.toarray(), Xt_dense)


@pytest.mark.parametrize(
    ["deg", "include_bias", "interaction_only", "dtype"],
    [
        (1, True, False, int),
        (2, True, False, int),
        (2, True, False, np.float32),
        (2, True, False, np.float64),
        (3, False, False, np.float64),
        (3, False, True, np.float64),
    ],
)
@pytest.mark.parametrize("csr_container", CSR_CONTAINERS)
def test_polynomial_features_csr_X(
    deg, include_bias, interaction_only, dtype, csr_container
):
    rng = np.random.RandomState(0)
    X = rng.randint(0, 2, (100, 2))
    X_csr = csr_container(X)

    est = PolynomialFeatures(
        deg, include_bias=include_bias, interaction_only=interaction_only
    )
    Xt_csr = est.fit_transform(X_csr.astype(dtype))
    Xt_dense = est.fit_transform(X.astype(dtype, copy=False))

    assert sparse.issparse(Xt_csr) and Xt_csr.format == "csr"
    assert Xt_csr.dtype == Xt_dense.dtype
    assert_array_almost_equal(Xt_csr.toarray(), Xt_dense)


@pytest.mark.parametrize("n_features", [1, 4, 5])
@pytest.mark.parametrize(
    "min_degree, max_degree", [(0, 1), (0, 2), (1, 3), (0, 4), (3, 4)]
)
@pytest.mark.parametrize("interaction_only", [True, False])
@pytest.mark.parametrize("include_bias", [True, False])
@pytest.mark.parametrize("csr_container", CSR_CONTAINERS)
def test_num_combinations(
    n_features, min_degree, max_degree, interaction_only, include_bias, csr_container
):
    """
    Test that n_output_features_ is calculated correctly.
    """
    x = csr_container(([1], ([0], [n_features - 1])))
    est = PolynomialFeatures(
        degree=max_degree,
        interaction_only=interaction_only,
        include_bias=include_bias,
    )
    est.fit(x)
    num_combos = est.n_output_features_

    combos = PolynomialFeatures._combinations(
        n_features=n_features,
        min_degree=0,
        max_degree=max_degree,
        interaction_only=interaction_only,
        include_bias=include_bias,
    )
    assert num_combos == sum([1 for _ in combos])


@pytest.mark.parametrize(
    ["deg", "include_bias", "interaction_only", "dtype"],
    [
        (2, True, False, np.float32),
        (2, True, False, np.float64),
        (3, False, False, np.float64),
        (3, False, True, np.float64),
    ],
)
@pytest.mark.parametrize("csr_container", CSR_CONTAINERS)
def test_polynomial_features_csr_X_floats(
    deg, include_bias, interaction_only, dtype, csr_container
):
    X_csr = csr_container(sparse_random(1000, 10, 0.5, random_state=0))
    X = X_csr.toarray()

    est = PolynomialFeatures(
        deg, include_bias=include_bias, interaction_only=interaction_only
    )
    Xt_csr = est.fit_transform(X_csr.astype(dtype))
    Xt_dense = est.fit_transform(X.astype(dtype))

    assert sparse.issparse(Xt_csr) and Xt_csr.format == "csr"
    assert Xt_csr.dtype == Xt_dense.dtype
    assert_array_almost_equal(Xt_csr.toarray(), Xt_dense)


@pytest.mark.parametrize(
    ["zero_row_index", "deg", "interaction_only"],
    [
        (0, 2, True),
        (1, 2, True),
        (2, 2, True),
        (0, 3, True),
        (1, 3, True),
        (2, 3, True),
        (0, 2, False),
        (1, 2, False),
        (2, 2, False),
        (0, 3, False),
        (1, 3, False),
        (2, 3, False),
    ],
)
@pytest.mark.parametrize("csr_container", CSR_CONTAINERS)
def test_polynomial_features_csr_X_zero_row(
    zero_row_index, deg, interaction_only, csr_container
):
    X_csr = csr_container(sparse_random(3, 10, 1.0, random_state=0))
    X_csr[zero_row_index, :] = 0.0
    X = X_csr.toarray()

    est = PolynomialFeatures(deg, include_bias=False, interaction_only=interaction_only)
    Xt_csr = est.fit_transform(X_csr)
    Xt_dense = est.fit_transform(X)

    assert sparse.issparse(Xt_csr) and Xt_csr.format == "csr"
    assert Xt_csr.dtype == Xt_dense.dtype
    assert_array_almost_equal(Xt_csr.toarray(), Xt_dense)


# This degree should always be one more than the highest degree supported by
# _csr_expansion.
@pytest.mark.parametrize(
    ["include_bias", "interaction_only"],
    [(True, True), (True, False), (False, True), (False, False)],
)
@pytest.mark.parametrize("csr_container", CSR_CONTAINERS)
def test_polynomial_features_csr_X_degree_4(
    include_bias, interaction_only, csr_container
):
    X_csr = csr_container(sparse_random(1000, 10, 0.5, random_state=0))
    X = X_csr.toarray()

    est = PolynomialFeatures(
        4, include_bias=include_bias, interaction_only=interaction_only
    )
    Xt_csr = est.fit_transform(X_csr)
    Xt_dense = est.fit_transform(X)

    assert sparse.issparse(Xt_csr) and Xt_csr.format == "csr"
    assert Xt_csr.dtype == Xt_dense.dtype
    assert_array_almost_equal(Xt_csr.toarray(), Xt_dense)


@pytest.mark.parametrize(
    ["deg", "dim", "interaction_only"],
    [
        (2, 1, True),
        (2, 2, True),
        (3, 1, True),
        (3, 2, True),
        (3, 3, True),
        (2, 1, False),
        (2, 2, False),
        (3, 1, False),
        (3, 2, False),
        (3, 3, False),
    ],
)
@pytest.mark.parametrize("csr_container", CSR_CONTAINERS)
def test_polynomial_features_csr_X_dim_edges(deg, dim, interaction_only, csr_container):
    X_csr = csr_container(sparse_random(1000, dim, 0.5, random_state=0))
    X = X_csr.toarray()

    est = PolynomialFeatures(deg, interaction_only=interaction_only)
    Xt_csr = est.fit_transform(X_csr)
    Xt_dense = est.fit_transform(X)

    assert sparse.issparse(Xt_csr) and Xt_csr.format == "csr"
    assert Xt_csr.dtype == Xt_dense.dtype
    assert_array_almost_equal(Xt_csr.toarray(), Xt_dense)


@pytest.mark.parametrize("interaction_only", [True, False])
@pytest.mark.parametrize("include_bias", [True, False])
@pytest.mark.parametrize("csr_container", CSR_CONTAINERS)
def test_csr_polynomial_expansion_index_overflow_non_regression(
    interaction_only, include_bias, csr_container
):
    """Check the automatic index dtype promotion to `np.int64` when needed.

    This ensures that sufficiently large input configurations get
    properly promoted to use `np.int64` for index and indptr representation
    while preserving data integrity. Non-regression test for gh-16803.

    Note that this is only possible for Python runtimes with a 64 bit address
    space. On 32 bit platforms, a `ValueError` is raised instead.
    """

    def degree_2_calc(d, i, j):
        if interaction_only:
            return d * i - (i**2 + 3 * i) // 2 - 1 + j
        else:
            return d * i - (i**2 + i) // 2 + j

    n_samples = 13
    n_features = 120001
    data_dtype = np.float32
    data = np.arange(1, 5, dtype=np.int64)
    row = np.array([n_samples - 2, n_samples - 2, n_samples - 1, n_samples - 1])
    # An int64 dtype is required to avoid overflow error on Windows within the
    # `degree_2_calc` function.
    col = np.array(
        [n_features - 2, n_features - 1, n_features - 2, n_features - 1], dtype=np.int64
    )
    X = csr_container(
        (data, (row, col)),
        shape=(n_samples, n_features),
        dtype=data_dtype,
    )
    pf = PolynomialFeatures(
        interaction_only=interaction_only, include_bias=include_bias, degree=2
    )

    # Calculate the number of combinations a-priori, and if needed check for
    # the correct ValueError and terminate the test early.
    num_combinations = pf._num_combinations(
        n_features=n_features,
        min_degree=0,
        max_degree=2,
        interaction_only=pf.interaction_only,
        include_bias=pf.include_bias,
    )
    if num_combinations > np.iinfo(np.intp).max:
        msg = (
            r"The output that would result from the current configuration would have"
            r" \d* features which is too large to be indexed"
        )
        with pytest.raises(ValueError, match=msg):
            pf.fit(X)
        return
    X_trans = pf.fit_transform(X)
    row_nonzero, col_nonzero = X_trans.nonzero()
    n_degree_1_features_out = n_features + include_bias
    max_degree_2_idx = (
        degree_2_calc(n_features, col[int(not interaction_only)], col[1])
        + n_degree_1_features_out
    )

    # Account for bias of all samples except last one which will be handled
    # separately since there are distinct data values before it
    data_target = [1] * (n_samples - 2) if include_bias else []
    col_nonzero_target = [0] * (n_samples - 2) if include_bias else []

    for i in range(2):
        x = data[2 * i]
        y = data[2 * i + 1]
        x_idx = col[2 * i]
        y_idx = col[2 * i + 1]
        if include_bias:
            data_target.append(1)
            col_nonzero_target.append(0)
        data_target.extend([x, y])
        col_nonzero_target.extend(
            [x_idx + int(include_bias), y_idx + int(include_bias)]
        )
        if not interaction_only:
            data_target.extend([x * x, x * y, y * y])
            col_nonzero_target.extend(
                [
                    degree_2_calc(n_features, x_idx, x_idx) + n_degree_1_features_out,
                    degree_2_calc(n_features, x_idx, y_idx) + n_degree_1_features_out,
                    degree_2_calc(n_features, y_idx, y_idx) + n_degree_1_features_out,
                ]
            )
        else:
            data_target.extend([x * y])
            col_nonzero_target.append(
                degree_2_calc(n_features, x_idx, y_idx) + n_degree_1_features_out
            )

    nnz_per_row = int(include_bias) + 3 + 2 * int(not interaction_only)

    assert pf.n_output_features_ == max_degree_2_idx + 1
    assert X_trans.dtype == data_dtype
    assert X_trans.shape == (n_samples, max_degree_2_idx + 1)
    assert X_trans.indptr.dtype == X_trans.indices.dtype == np.int64
    # Ensure that dtype promotion was actually required:
    assert X_trans.indices.max() > np.iinfo(np.int32).max

    row_nonzero_target = list(range(n_samples - 2)) if include_bias else []
    row_nonzero_target.extend(
        [n_samples - 2] * nnz_per_row + [n_samples - 1] * nnz_per_row
    )

    assert_allclose(X_trans.data, data_target)
    assert_array_equal(row_nonzero, row_nonzero_target)
    assert_array_equal(col_nonzero, col_nonzero_target)


@pytest.mark.parametrize(
    "degree, n_features",
    [
        # Needs promotion to int64 when interaction_only=False
        (2, 65535),
        (3, 2344),
        # This guarantees that the intermediate operation when calculating
        # output columns would overflow a C-long, hence checks that python-
        # longs are being used.
        (2, int(np.sqrt(np.iinfo(np.int64).max) + 1)),
        (3, 65535),
        # This case tests the second clause of the overflow check which
        # takes into account the value of `n_features` itself.
        (2, int(np.sqrt(np.iinfo(np.int64).max))),
    ],
)
@pytest.mark.parametrize("interaction_only", [True, False])
@pytest.mark.parametrize("include_bias", [True, False])
@pytest.mark.parametrize("csr_container", CSR_CONTAINERS)
def test_csr_polynomial_expansion_index_overflow(
    degree, n_features, interaction_only, include_bias, csr_container
):
    """Tests known edge-cases to the dtype promotion strategy and custom
    Cython code, including a current bug in the upstream
    `scipy.sparse.hstack`.
    """
    data = [1.0]
    # Use int32 indices as much as we can
    indices_dtype = np.int32 if n_features - 1 <= np.iinfo(np.int32).max else np.int64
    row = np.array([0], dtype=indices_dtype)
    col = np.array([n_features - 1], dtype=indices_dtype)

    # First degree index
    expected_indices = [
        n_features - 1 + int(include_bias),
    ]
    # Second degree index
    expected_indices.append(n_features * (n_features + 1) // 2 + expected_indices[0])
    # Third degree index
    expected_indices.append(
        n_features * (n_features + 1) * (n_features + 2) // 6 + expected_indices[1]
    )

    X = csr_container((data, (row, col)))
    pf = PolynomialFeatures(
        interaction_only=interaction_only, include_bias=include_bias, degree=degree
    )

    # Calculate the number of combinations a-priori, and if needed check for
    # the correct ValueError and terminate the test early.
    num_combinations = pf._num_combinations(
        n_features=n_features,
        min_degree=0,
        max_degree=degree,
        interaction_only=pf.interaction_only,
        include_bias=pf.include_bias,
    )
    if num_combinations > np.iinfo(np.intp).max:
        msg = (
            r"The output that would result from the current configuration would have"
            r" \d* features which is too large to be indexed"
        )
        with pytest.raises(ValueError, match=msg):
            pf.fit(X)
        return

    # In SciPy < 1.8, a bug occurs when an intermediate matrix in
    # `to_stack` in `hstack` fits within int32 however would require int64 when
    # combined with all previous matrices in `to_stack`.
    if sp_version < parse_version("1.8.0"):
        has_bug = False
        max_int32 = np.iinfo(np.int32).max
        cumulative_size = n_features + include_bias
        for deg in range(2, degree + 1):
            max_indptr = _calc_total_nnz(X.indptr, interaction_only, deg)
            max_indices = _calc_expanded_nnz(n_features, interaction_only, deg) - 1
            cumulative_size += max_indices + 1
            needs_int64 = max(max_indices, max_indptr) > max_int32
            has_bug |= not needs_int64 and cumulative_size > max_int32
        if has_bug:
            msg = r"In scipy versions `<1.8.0`, the function `scipy.sparse.hstack`"
            with pytest.raises(ValueError, match=msg):
                X_trans = pf.fit_transform(X)
            return

    # When `n_features>=65535`, `scipy.sparse.hstack` may not use the right
    # dtype for representing indices and indptr if `n_features` is still
    # small enough so that each block matrix's indices and indptr arrays
    # can be represented with `np.int32`. We test `n_features==65535`
    # since it is guaranteed to run into this bug.
    if (
        sp_version < parse_version("1.9.2")
        and n_features == 65535
        and degree == 2
        and not interaction_only
    ):  # pragma: no cover
        msg = r"In scipy versions `<1.9.2`, the function `scipy.sparse.hstack`"
        with pytest.raises(ValueError, match=msg):
            X_trans = pf.fit_transform(X)
        return
    X_trans = pf.fit_transform(X)

    expected_dtype = np.int64 if num_combinations > np.iinfo(np.int32).max else np.int32
    # Terms higher than first degree
    non_bias_terms = 1 + (degree - 1) * int(not interaction_only)
    expected_nnz = int(include_bias) + non_bias_terms
    assert X_trans.dtype == X.dtype
    assert X_trans.shape == (1, pf.n_output_features_)
    assert X_trans.indptr.dtype == X_trans.indices.dtype == expected_dtype
    assert X_trans.nnz == expected_nnz

    if include_bias:
        assert X_trans[0, 0] == pytest.approx(1.0)
    for idx in range(non_bias_terms):
        assert X_trans[0, expected_indices[idx]] == pytest.approx(1.0)

    offset = interaction_only * n_features
    if degree == 3:
        offset *= 1 + n_features
    assert pf.n_output_features_ == expected_indices[degree - 1] + 1 - offset


@pytest.mark.parametrize("interaction_only", [True, False])
@pytest.mark.parametrize("include_bias", [True, False])
@pytest.mark.parametrize("csr_container", CSR_CONTAINERS)
def test_csr_polynomial_expansion_too_large_to_index(
    interaction_only, include_bias, csr_container
):
    n_features = np.iinfo(np.int64).max // 2
    data = [1.0]
    row = [0]
    col = [n_features - 1]
    X = csr_container((data, (row, col)))
    pf = PolynomialFeatures(
        interaction_only=interaction_only, include_bias=include_bias, degree=(2, 2)
    )
    msg = (
        r"The output that would result from the current configuration would have \d*"
        r" features which is too large to be indexed"
    )
    with pytest.raises(ValueError, match=msg):
        pf.fit(X)
    with pytest.raises(ValueError, match=msg):
        pf.fit_transform(X)


@pytest.mark.parametrize("sparse_container", CSR_CONTAINERS + CSC_CONTAINERS)
def test_polynomial_features_behaviour_on_zero_degree(sparse_container):
    """Check that PolynomialFeatures raises error when degree=0 and include_bias=False,
    and output a single constant column when include_bias=True
    """
    X = np.ones((10, 2))
    poly = PolynomialFeatures(degree=0, include_bias=False)
    err_msg = (
        "Setting degree to zero and include_bias to False would result in"
        " an empty output array."
    )
    with pytest.raises(ValueError, match=err_msg):
        poly.fit_transform(X)

    poly = PolynomialFeatures(degree=(0, 0), include_bias=False)
    err_msg = (
        "Setting both min_degree and max_degree to zero and include_bias to"
        " False would result in an empty output array."
    )
    with pytest.raises(ValueError, match=err_msg):
        poly.fit_transform(X)

    for _X in [X, sparse_container(X)]:
        poly = PolynomialFeatures(degree=0, include_bias=True)
        output = poly.fit_transform(_X)
        # convert to dense array if needed
        if sparse.issparse(output):
            output = output.toarray()
        assert_array_equal(output, np.ones((X.shape[0], 1)))


def test_sizeof_LARGEST_INT_t():
    # On Windows, scikit-learn is typically compiled with MSVC that
    # does not support int128 arithmetic (at the time of writing):
    # https://stackoverflow.com/a/6761962/163740
    if sys.platform == "win32" or (
        sys.maxsize <= 2**32 and sys.platform != "emscripten"
    ):
        expected_size = 8
    else:
        expected_size = 16

    assert _get_sizeof_LARGEST_INT_t() == expected_size


@pytest.mark.xfail(
    sys.platform == "win32",
    reason=(
        "On Windows, scikit-learn is typically compiled with MSVC that does not support"
        " int128 arithmetic (at the time of writing)"
    ),
    run=True,
)
@pytest.mark.parametrize("csr_container", CSR_CONTAINERS)
def test_csr_polynomial_expansion_windows_fail(csr_container):
    # Minimum needed to ensure integer overflow occurs while guaranteeing an
    # int64-indexable output.
    n_features = int(np.iinfo(np.int64).max ** (1 / 3) + 3)
    data = [1.0]
    row = [0]
    col = [n_features - 1]

    # First degree index
    expected_indices = [
        n_features - 1,
    ]
    # Second degree index
    expected_indices.append(
        int(n_features * (n_features + 1) // 2 + expected_indices[0])
    )
    # Third degree index
    expected_indices.append(
        int(n_features * (n_features + 1) * (n_features + 2) // 6 + expected_indices[1])
    )

    X = csr_container((data, (row, col)))
    pf = PolynomialFeatures(interaction_only=False, include_bias=False, degree=3)
    if sys.maxsize <= 2**32:
        msg = (
            r"The output that would result from the current configuration would"
            r" have \d*"
            r" features which is too large to be indexed"
        )
        with pytest.raises(ValueError, match=msg):
            pf.fit_transform(X)
    else:
        X_trans = pf.fit_transform(X)
        for idx in range(3):
            assert X_trans[0, expected_indices[idx]] == pytest.approx(1.0)