-
Notifications
You must be signed in to change notification settings - Fork 85
/
convert.py
5393 lines (4621 loc) · 197 KB
/
convert.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
# BSD 3-Clause License; see https://github.com/scikit-hep/awkward-1.0/blob/main/LICENSE
from __future__ import absolute_import
import numbers
import json
import collections
import math
import os
import threading
import distutils.version
import glob
import re
try:
from collections.abc import Iterable, Sized
from collections.abc import MutableMapping
except ImportError:
from collections import Iterable, Sized
from collections import MutableMapping
import awkward as ak
np = ak.nplike.NumpyMetadata.instance()
numpy = ak.nplike.Numpy.instance()
def _regularize_path(path):
if isinstance(path, getattr(os, "PathLike", ())):
path = os.fspath(path)
elif hasattr(path, "__fspath__"):
path = path.__fspath__()
elif path.__class__.__module__ == "pathlib":
import pathlib
if isinstance(path, pathlib.Path):
path = str(path)
if isinstance(path, str):
path = os.path.expanduser(path)
return path
def from_numpy(
array, regulararray=False, recordarray=True, highlevel=True, behavior=None
):
"""
Args:
array (np.ndarray): The NumPy array to convert into an Awkward Array.
This array can be a np.ma.MaskedArray.
regulararray (bool): If True and the array is multidimensional,
the dimensions are represented by nested #ak.layout.RegularArray
nodes; if False and the array is multidimensional, the dimensions
are represented by a multivalued #ak.layout.NumpyArray.shape.
If the array is one-dimensional, this has no effect.
recordarray (bool): If True and the array is a NumPy structured array
(dtype.names is not None), the fields are represented by an
#ak.layout.RecordArray; if False and the array is a structured
array, the structure is left in the #ak.layout.NumpyArray `format`,
which some functions do not recognize.
highlevel (bool): If True, return an #ak.Array; otherwise, return
a low-level #ak.layout.Content subclass.
behavior (None or dict): Custom #ak.behavior for the output array, if
high-level.
Converts a NumPy array into an Awkward Array.
The resulting layout may involve the following #ak.layout.Content types
(only):
* #ak.layout.NumpyArray
* #ak.layout.ByteMaskedArray or #ak.layout.UnmaskedArray if the
`array` is an np.ma.MaskedArray.
* #ak.layout.RegularArray if `regulararray=True`.
* #ak.layout.RecordArray if `recordarray=True`.
See also #ak.to_numpy and #ak.from_cupy.
"""
def recurse(array, mask):
if regulararray and len(array.shape) > 1:
return ak.layout.RegularArray(
recurse(array.reshape((-1,) + array.shape[2:]), mask),
array.shape[1],
array.shape[0],
)
if len(array.shape) == 0:
array = array.reshape(1)
if array.dtype.kind == "S":
asbytes = array.reshape(-1)
itemsize = asbytes.dtype.itemsize
starts = numpy.arange(0, len(asbytes) * itemsize, itemsize, dtype=np.int64)
stops = starts + numpy.char.str_len(asbytes)
data = ak.layout.ListArray64(
ak.layout.Index64(starts),
ak.layout.Index64(stops),
ak.layout.NumpyArray(
asbytes.view("u1"), parameters={"__array__": "byte"}
),
parameters={"__array__": "bytestring"},
)
for i in range(len(array.shape) - 1, 0, -1):
data = ak.layout.RegularArray(data, array.shape[i], array.shape[i - 1])
elif array.dtype.kind == "U":
asbytes = numpy.char.encode(array.reshape(-1), "utf-8", "surrogateescape")
itemsize = asbytes.dtype.itemsize
starts = numpy.arange(0, len(asbytes) * itemsize, itemsize, dtype=np.int64)
stops = starts + numpy.char.str_len(asbytes)
data = ak.layout.ListArray64(
ak.layout.Index64(starts),
ak.layout.Index64(stops),
ak.layout.NumpyArray(
asbytes.view("u1"), parameters={"__array__": "char"}
),
parameters={"__array__": "string"},
)
for i in range(len(array.shape) - 1, 0, -1):
data = ak.layout.RegularArray(data, array.shape[i], array.shape[i - 1])
else:
data = ak.layout.NumpyArray(array)
if mask is None:
return data
elif mask is False or (isinstance(mask, np.bool_) and not mask):
# NumPy's MaskedArray with mask == False is an UnmaskedArray
if len(array.shape) == 1:
return ak.layout.UnmaskedArray(data)
else:
def attach(x):
if isinstance(x, ak.layout.NumpyArray):
return ak.layout.UnmaskedArray(x)
else:
return ak.layout.RegularArray(attach(x.content), x.size, len(x))
return attach(data.toRegularArray())
else:
# NumPy's MaskedArray is a ByteMaskedArray with valid_when=False
return ak.layout.ByteMaskedArray(
ak.layout.Index8(mask), data, valid_when=False
)
if isinstance(array, numpy.ma.MaskedArray):
mask = numpy.ma.getmask(array)
array = numpy.ma.getdata(array)
if isinstance(mask, np.ndarray) and len(mask.shape) > 1:
regulararray = True
mask = mask.reshape(-1)
else:
mask = None
if not recordarray or array.dtype.names is None:
layout = recurse(array, mask)
else:
contents = []
for name in array.dtype.names:
contents.append(recurse(array[name], mask))
layout = ak.layout.RecordArray(contents, array.dtype.names)
return ak._util.maybe_wrap(layout, behavior, highlevel)
def to_numpy(array, allow_missing=True):
"""
Converts `array` (many types supported, including all Awkward Arrays and
Records) into a NumPy array, if possible.
If the data are numerical and regular (nested lists have equal lengths
in each dimension, as described by the #type), they can be losslessly
converted to a NumPy array and this function returns without an error.
Otherwise, the function raises an error. It does not create a NumPy
array with dtype `"O"` for `np.object_` (see the
[note on object_ type](https://docs.scipy.org/doc/numpy/reference/arrays.scalars.html#arrays-scalars-built-in))
since silent conversions to dtype `"O"` arrays would not only be a
significant performance hit, but would also break functionality, since
nested lists in a NumPy `"O"` array are severed from the array and
cannot be sliced as dimensions.
If `array` is a scalar, it is converted into a NumPy scalar.
If `allow_missing` is True; NumPy
[masked arrays](https://docs.scipy.org/doc/numpy/reference/maskedarray.html)
are a possible result; otherwise, missing values (None) cause this
function to raise an error.
See also #ak.from_numpy and #ak.to_cupy.
"""
if isinstance(array, (bool, str, bytes, numbers.Number)):
return numpy.array([array])[0]
elif ak._util.py27 and isinstance(array, ak._util.unicode):
return numpy.array([array])[0]
elif isinstance(array, np.ndarray):
return array
elif isinstance(array, ak.highlevel.Array):
return to_numpy(array.layout, allow_missing=allow_missing)
elif isinstance(array, ak.highlevel.Record):
out = array.layout
return to_numpy(out.array[out.at : out.at + 1], allow_missing=allow_missing)[0]
elif isinstance(array, ak.highlevel.ArrayBuilder):
return to_numpy(array.snapshot().layout, allow_missing=allow_missing)
elif isinstance(array, ak.layout.ArrayBuilder):
return to_numpy(array.snapshot(), allow_missing=allow_missing)
elif ak.operations.describe.parameters(array).get("__array__") == "bytestring":
return numpy.array(
[
ak.behaviors.string.ByteBehavior(array[i]).__bytes__()
for i in range(len(array))
]
)
elif ak.operations.describe.parameters(array).get("__array__") == "string":
return numpy.array(
[
ak.behaviors.string.CharBehavior(array[i]).__str__()
for i in range(len(array))
]
)
elif (
str(ak.operations.describe.type(array)) == "datetime64"
or str(ak.operations.describe.type(array)) == "timedelta64"
):
return array
elif isinstance(array, ak.partition.PartitionedArray):
tocat = [to_numpy(x, allow_missing=allow_missing) for x in array.partitions]
if any(isinstance(x, numpy.ma.MaskedArray) for x in tocat):
return numpy.ma.concatenate(tocat)
else:
return numpy.concatenate(tocat)
elif isinstance(array, ak._util.virtualtypes):
return to_numpy(array.array, allow_missing=True)
elif isinstance(array, ak._util.unknowntypes):
return numpy.array([])
elif isinstance(array, ak._util.indexedtypes):
return to_numpy(array.project(), allow_missing=allow_missing)
elif isinstance(array, ak._util.uniontypes):
contents = [
to_numpy(array.project(i), allow_missing=allow_missing)
for i in range(array.numcontents)
]
if any(isinstance(x, numpy.ma.MaskedArray) for x in contents):
try:
out = numpy.ma.concatenate(contents)
except Exception:
raise ValueError(
"cannot convert {0} into numpy.ma.MaskedArray".format(array)
+ ak._util.exception_suffix(__file__)
)
else:
try:
out = numpy.concatenate(contents)
except Exception:
raise ValueError(
"cannot convert {0} into np.ndarray".format(array)
+ ak._util.exception_suffix(__file__)
)
tags = numpy.asarray(array.tags)
for tag, content in enumerate(contents):
mask = tags == tag
out[mask] = content
return out
elif isinstance(array, ak.layout.UnmaskedArray):
content = to_numpy(array.content, allow_missing=allow_missing)
if allow_missing:
return numpy.ma.MaskedArray(content)
else:
return content
elif isinstance(array, ak._util.optiontypes):
content = to_numpy(array.project(), allow_missing=allow_missing)
shape = list(content.shape)
shape[0] = len(array)
data = numpy.empty(shape, dtype=content.dtype)
mask0 = numpy.asarray(array.bytemask()).view(np.bool_)
if mask0.any():
if allow_missing:
mask = numpy.broadcast_to(
mask0.reshape((shape[0],) + (1,) * (len(shape) - 1)), shape
)
if isinstance(content, numpy.ma.MaskedArray):
mask1 = numpy.ma.getmaskarray(content)
mask = mask.copy()
mask[~mask0] |= mask1
data[~mask0] = content
return numpy.ma.MaskedArray(data, mask)
else:
raise ValueError(
"ak.to_numpy cannot convert 'None' values to "
"np.ma.MaskedArray unless the "
"'allow_missing' parameter is set to True"
+ ak._util.exception_suffix(__file__)
)
else:
if allow_missing:
return numpy.ma.MaskedArray(content)
else:
return content
elif isinstance(array, ak.layout.RegularArray):
out = to_numpy(array.content, allow_missing=allow_missing)
head, tail = out.shape[0], out.shape[1:]
if array.size == 0:
shape = (0, 0) + tail
else:
shape = (head // array.size, array.size) + tail
return out[: shape[0] * array.size].reshape(shape)
elif isinstance(array, ak._util.listtypes):
return to_numpy(array.toRegularArray(), allow_missing=allow_missing)
elif isinstance(array, ak._util.recordtypes):
if array.numfields == 0:
return numpy.empty(len(array), dtype=[])
contents = [
to_numpy(array.field(i), allow_missing=allow_missing)
for i in range(array.numfields)
]
if any(len(x.shape) != 1 for x in contents):
raise ValueError(
"cannot convert {0} into np.ndarray".format(array)
+ ak._util.exception_suffix(__file__)
)
out = numpy.empty(
len(contents[0]),
dtype=[(str(n), x.dtype) for n, x in zip(array.keys(), contents)],
)
for n, x in zip(array.keys(), contents):
out[n] = x
return out
elif isinstance(array, ak.layout.NumpyArray):
out = ak.nplike.of(array).asarray(array)
if type(out).__module__.startswith("cupy."):
return out.get()
else:
return out
elif isinstance(array, ak.layout.Content):
raise AssertionError(
"unrecognized Content type: {0}".format(type(array))
+ ak._util.exception_suffix(__file__)
)
elif isinstance(array, Iterable):
return numpy.asarray(array)
else:
raise ValueError(
"cannot convert {0} into np.ndarray".format(array)
+ ak._util.exception_suffix(__file__)
)
def from_cupy(array, regulararray=False, highlevel=True, behavior=None):
"""
Args:
array (cp.ndarray): The CuPy array to convert into an Awkward Array.
regulararray (bool): If True and the array is multidimensional,
the dimensions are represented by nested #ak.layout.RegularArray
nodes; if False and the array is multidimensional, the dimensions
are represented by a multivalued #ak.layout.NumpyArray.shape.
If the array is one-dimensional, this has no effect.
highlevel (bool): If True, return an #ak.Array; otherwise, return
a low-level #ak.layout.Content subclass.
behavior (None or dict): Custom #ak.behavior for the output array, if
high-level.
Converts a CuPy array into an Awkward Array.
The resulting layout may involve the following #ak.layout.Content types
(only):
* #ak.layout.NumpyArray
* #ak.layout.RegularArray if `regulararray=True`.
See also #ak.to_cupy, #ak.from_numpy and #ak.from_jax.
"""
def recurse(array):
if regulararray and len(array.shape) > 1:
return ak.layout.RegularArray(
recurse(array.reshape((-1,) + array.shape[2:])),
array.shape[1],
array.shape[0],
)
if len(array.shape) == 0:
data = ak.layout.NumpyArray.from_cupy(array.reshape(1))
else:
data = ak.layout.NumpyArray.from_cupy(array)
return data
layout = recurse(array)
return ak._util.maybe_wrap(layout, behavior, highlevel)
def to_cupy(array):
"""
Converts `array` (many types supported) into a CuPy array, if possible.
If the data are numerical and regular (nested lists have equal lengths
in each dimension, as described by the #type), they can be losslessly
converted to a CuPy array and this function returns without an error.
Otherwise, the function raises an error.
If `array` is a scalar, it is converted into a CuPy scalar.
See also #ak.from_cupy and #ak.to_numpy.
"""
cupy = ak.nplike.Cupy.instance()
np = ak.nplike.NumpyMetadata.instance()
if isinstance(array, (bool, numbers.Number)):
return cupy.array([array])[0]
elif isinstance(array, cupy.ndarray):
return array
elif isinstance(array, np.ndarray):
return cupy.asarray(array)
elif isinstance(array, ak.highlevel.Array):
return to_cupy(array.layout)
elif isinstance(array, ak.highlevel.Record):
raise ValueError(
"CuPy does not support record structures"
+ ak._util.exception_suffix(__file__)
)
elif isinstance(array, ak.highlevel.ArrayBuilder):
return to_cupy(array.snapshot().layout)
elif isinstance(array, ak.layout.ArrayBuilder):
return to_cupy(array.snapshot())
elif (
ak.operations.describe.parameters(array).get("__array__") == "bytestring"
or ak.operations.describe.parameters(array).get("__array__") == "string"
):
raise ValueError(
"CuPy does not support arrays of strings"
+ ak._util.exception_suffix(__file__)
)
elif isinstance(array, ak.partition.PartitionedArray):
return cupy.concatenate([to_cupy(x) for x in array.partitions])
elif isinstance(array, ak._util.virtualtypes):
return to_cupy(array.array)
elif isinstance(array, ak._util.unknowntypes):
return cupy.array([])
elif isinstance(array, ak._util.indexedtypes):
return to_cupy(array.project())
elif isinstance(array, ak._util.uniontypes):
contents = [to_cupy(array.project(i)) for i in range(array.numcontents)]
out = cupy.concatenate(contents)
tags = cupy.asarray(array.tags)
for tag, content in enumerate(contents):
mask = tags == tag
out[mask] = content
return out
elif isinstance(array, ak.layout.UnmaskedArray):
return to_cupy(array.content)
elif isinstance(array, ak._util.optiontypes):
content = to_cupy(array.project())
shape = list(content.shape)
shape[0] = len(array)
mask0 = cupy.asarray(array.bytemask()).view(np.bool_)
if mask0.any():
raise ValueError(
"CuPy does not support masked arrays"
+ ak._util.exception_suffix(__file__)
)
else:
return content
elif isinstance(array, ak.layout.RegularArray):
out = to_cupy(array.content)
head, tail = out.shape[0], out.shape[1:]
shape = (head // array.size, array.size) + tail
return out[: shape[0] * array.size].reshape(shape)
elif isinstance(array, ak._util.listtypes):
return to_cupy(array.toRegularArray())
elif isinstance(array, ak._util.recordtypes):
raise ValueError(
"CuPy does not support record structures"
+ ak._util.exception_suffix(__file__)
)
elif isinstance(array, ak.layout.NumpyArray):
return array.to_cupy()
elif isinstance(array, ak.layout.Content):
raise AssertionError(
"unrecognized Content type: {0}".format(type(array))
+ ak._util.exception_suffix(__file__)
)
elif isinstance(array, Iterable):
return cupy.asarray(array)
else:
raise ValueError(
"cannot convert {0} into cp.ndarray".format(array)
+ ak._util.exception_suffix(__file__)
)
def from_jax(array, regulararray=False, highlevel=True, behavior=None):
"""
Args:
array (jax.numpy.array): The `jax.numpy.array` to convert into an Awkward Array.
regulararray (bool): If True and the array is multidimensional,
the dimensions are represented by nested #ak.layout.RegularArray
nodes; if False and the array is multidimensional, the dimensions
are represented by a multivalued #ak.layout.NumpyArray.shape.
If the array is one-dimensional, this has no effect.
highlevel (bool): If True, return an #ak.Array; otherwise, return
a low-level #ak.layout.Content subclass.
behavior (None or dict): Custom #ak.behavior for the output array, if
high-level.
Converts a JAX array into an Awkward Array.
The resulting layout may involve the following #ak.layout.Content types
(only):
* #ak.layout.NumpyArray
* #ak.layout.RegularArray if `regulararray=True`.
See also #ak.from_cupy and #ak.from_numpy.
"""
def recurse(array):
if regulararray and len(array.shape) > 1:
return ak.layout.RegularArray(
recurse(array.reshape((-1,) + array.shape[2:])),
array.shape[1],
array.shape[0],
)
if len(array.shape) == 0:
data = ak.layout.NumpyArray.from_jax(array.reshape(1))
else:
data = ak.layout.NumpyArray.from_jax(array)
return data
layout = recurse(array)
return ak._util.maybe_wrap(layout, behavior, highlevel)
def to_jax(array):
"""
Converts `array` (many types supported) into a JAX array, if possible.
If the data are numerical and regular (nested lists have equal lengths
in each dimension, as described by the #type), they can be losslessly
converted to a CuPy array and this function returns without an error.
Otherwise, the function raises an error.
If `array` is a scalar, it is converted into a JAX scalar.
See also #ak.to_cupy, #ak.from_jax and #ak.to_numpy.
"""
try:
import jax
except ImportError:
raise ImportError(
"""to use {0}, you must install jax:
pip install jax jaxlib
"""
)
if isinstance(array, (bool, numbers.Number)):
return jax.numpy.array([array])[0]
elif isinstance(array, jax.numpy.ndarray):
return array
elif isinstance(array, np.ndarray):
return jax.numpy.asarray(array)
elif isinstance(array, ak.highlevel.Array):
return to_jax(array.layout)
elif isinstance(array, ak.highlevel.Record):
raise ValueError(
"JAX does not support record structures"
+ ak._util.exception_suffix(__file__)
)
elif isinstance(array, ak.highlevel.ArrayBuilder):
return to_jax(array.snapshot().layout)
elif isinstance(array, ak.layout.ArrayBuilder):
return to_jax(array.snapshot())
elif (
ak.operations.describe.parameters(array).get("__array__") == "bytestring"
or ak.operations.describe.parameters(array).get("__array__") == "string"
):
raise ValueError(
"JAX does not support arrays of strings"
+ ak._util.exception_suffix(__file__)
)
elif isinstance(array, ak.partition.PartitionedArray):
return jax.numpy.concatenate([to_jax(x) for x in array.partitions])
elif isinstance(array, ak._util.virtualtypes):
return to_jax(array.array)
elif isinstance(array, ak._util.unknowntypes):
return jax.numpy.array([])
elif isinstance(array, ak._util.indexedtypes):
return to_jax(array.project())
elif isinstance(array, ak._util.uniontypes):
array = array.simplify()
if isinstance(array, ak._util.uniontypes):
raise ValueError(
"cannot convert {0} into jax.numpy.array".format(array)
+ ak._util.exception_suffix(__file__)
)
return to_jax(array)
elif isinstance(array, ak.layout.UnmaskedArray):
return to_jax(array.content)
elif isinstance(array, ak._util.optiontypes):
content = to_jax(array.project())
shape = list(content.shape)
shape[0] = len(array)
mask0 = jax.numpy.asarray(array.bytemask()).view(np.bool_)
if mask0.any():
raise ValueError(
"JAX does not support masked arrays"
+ ak._util.exception_suffix(__file__)
)
else:
return content
elif isinstance(array, ak.layout.RegularArray):
out = to_jax(array.content)
head, tail = out.shape[0], out.shape[1:]
shape = (head // array.size, array.size) + tail
return out[: shape[0] * array.size].reshape(shape)
elif isinstance(array, ak._util.listtypes):
return to_jax(array.toRegularArray())
elif isinstance(array, ak._util.recordtypes):
raise ValueError(
"JAX does not support record structures"
+ ak._util.exception_suffix(__file__)
)
elif isinstance(array, ak.layout.NumpyArray):
return array.to_jax()
elif isinstance(array, ak.layout.Content):
raise AssertionError(
"unrecognized Content type: {0}".format(type(array))
+ ak._util.exception_suffix(__file__)
)
elif isinstance(array, Iterable):
return jax.numpy.asarray(array)
else:
raise ValueError(
"cannot convert {0} into jax.numpy.array".format(array)
+ ak._util.exception_suffix(__file__)
)
def kernels(*arrays):
"""
Returns the names of the kernels library used by `arrays`. May be
* `"cpu"` for `libawkward-cpu-kernels.so`;
* `"cuda"` for `libawkward-cuda-kernels.so`;
* `"mixed"` if any of the arrays have different labels within their
structure or any arrays have different labels from each other;
* None if the objects are not Awkward, NumPy, or CuPy arrays (e.g.
Python numbers, booleans, strings).
Mixed arrays can't be used in any operations, and two arrays on different
devices can't be used in the same operation.
To use `"cuda"`, the package
[awkward-cuda-kernels](https://pypi.org/project/awkward-cuda-kernels)
be installed, either by
pip install awkward-cuda-kernels
or as an optional dependency with
pip install awkward[cuda] --upgrade
It is only available for Linux as a binary wheel, and only supports Nvidia
GPUs (it is written in CUDA).
See #ak.to_kernels.
"""
libs = set()
for array in arrays:
layout = ak.operations.convert.to_layout(
array,
allow_record=True,
allow_other=True,
)
if isinstance(
layout, (ak.layout.Content, ak.layout.Record, ak.partition.PartitionedArray)
):
libs.add(layout.kernels)
elif isinstance(layout, ak.nplike.numpy.ndarray):
libs.add("cpu")
elif type(layout).__module__.startswith("cupy."):
libs.add("cuda")
if libs == set():
return None
elif libs == set(["cpu"]):
return "cpu"
elif libs == set(["cuda"]):
return "cuda"
else:
return "mixed"
def to_kernels(array, kernels, highlevel=True, behavior=None):
"""
Args:
array: Data to convert to a specified `kernels` set.
kernels (`"cpu"` or `"cuda"`): If `"cpu"`, the array structure is
recursively copied (if need be) to main memory for use with
the default `libawkward-cpu-kernels.so`; if `"cuda"`, the
structure is copied to the GPU(s) for use with
`libawkward-cuda-kernels.so`.
highlevel (bool): If True, return an #ak.Array; otherwise, return
a low-level #ak.layout.Content subclass.
behavior (None or dict): Custom #ak.behavior for the output array, if
high-level.
Converts an array from `"cpu"`, `"cuda"`, or `"mixed"` kernels to `"cpu"`
or `"cuda"`.
An array is `"mixed"` if some components are set to use `"cpu"` kernels and
others are set to use `"cuda"` kernels. Mixed arrays can't be used in any
operations, and two arrays set to different kernels can't be used in the
same operation.
Any components that are already in the desired kernels library are viewed,
rather than copied, so this operation can be an inexpensive way to ensure
that an array is ready for a particular library.
To use `"cuda"`, the package
[awkward-cuda-kernels](https://pypi.org/project/awkward-cuda-kernels)
be installed, either by
pip install awkward-cuda-kernels
or as an optional dependency with
pip install awkward[cuda] --upgrade
It is only available for Linux as a binary wheel, and only supports Nvidia
GPUs (it is written in CUDA).
See #ak.kernels.
"""
arr = ak.to_layout(array)
out = arr.copy_to(kernels)
return ak._util.maybe_wrap_like(out, array, behavior, highlevel)
def from_iter(
iterable, highlevel=True, behavior=None, allow_record=True, initial=1024, resize=1.5
):
"""
Args:
iterable (Python iterable): Data to convert into an Awkward Array.
highlevel (bool): If True, return an #ak.Array; otherwise, return
a low-level #ak.layout.Content subclass.
behavior (None or dict): Custom #ak.behavior for the output array, if
high-level.
allow_record (bool): If True, the outermost element may be a record
(returning #ak.Record or #ak.layout.Record type, depending on
`highlevel`); if False, the outermost element must be an array.
initial (int): Initial size (in bytes) of buffers used by
#ak.layout.ArrayBuilder (see #ak.layout.ArrayBuilderOptions).
resize (float): Resize multiplier for buffers used by
#ak.layout.ArrayBuilder (see #ak.layout.ArrayBuilderOptions);
should be strictly greater than 1.
Converts Python data into an Awkward Array.
Internally, this function uses #ak.layout.ArrayBuilder (see the high-level
#ak.ArrayBuilder documentation for a more complete description), so it
has the same flexibility and the same constraints. Any heterogeneous
and deeply nested Python data can be converted, but the output will never
have regular-typed array lengths.
The following Python types are supported.
* bool, including `np.bool_`: converted into #ak.layout.NumpyArray.
* int, including `np.integer`: converted into #ak.layout.NumpyArray.
* float, including `np.floating`: converted into #ak.layout.NumpyArray.
* bytes: converted into #ak.layout.ListOffsetArray with parameter
`"__array__"` equal to `"bytestring"` (unencoded bytes).
* str: converted into #ak.layout.ListOffsetArray with parameter
`"__array__"` equal to `"string"` (UTF-8 encoded string).
* tuple: converted into #ak.layout.RecordArray without field names
(i.e. homogeneously typed, uniform sized tuples).
* dict: converted into #ak.layout.RecordArray with field names
(i.e. homogeneously typed records with the same sets of fields).
* iterable, including np.ndarray: converted into
#ak.layout.ListOffsetArray.
See also #ak.to_list.
"""
if isinstance(iterable, dict):
if allow_record:
return from_iter(
[iterable],
highlevel=highlevel,
behavior=behavior,
initial=initial,
resize=resize,
)[0]
else:
raise ValueError(
"cannot produce an array from a dict"
+ ak._util.exception_suffix(__file__)
)
out = ak.layout.ArrayBuilder(initial=initial, resize=resize)
for x in iterable:
out.fromiter(x)
layout = out.snapshot()
return ak._util.maybe_wrap(layout, behavior, highlevel)
def to_list(array):
"""
Converts `array` (many types supported, including all Awkward Arrays and
Records) into Python objects.
Awkward Array types have the following Pythonic translations.
* #ak.types.PrimitiveType: converted into bool, int, float.
* #ak.types.OptionType: missing values are converted into None.
* #ak.types.ListType: converted into list.
* #ak.types.RegularType: also converted into list. Python (and JSON)
forms lose information about the regularity of list lengths.
* #ak.types.ListType with parameter `"__array__"` equal to
`"__bytestring__"`: converted into bytes.
* #ak.types.ListType with parameter `"__array__"` equal to
`"__string__"`: converted into str.
* #ak.types.RecordArray without field names: converted into tuple.
* #ak.types.RecordArray with field names: converted into dict.
* #ak.types.UnionArray: Python data are naturally heterogeneous.
See also #ak.from_iter and #ak.Array.tolist.
"""
if array is None or isinstance(array, (bool, str, bytes, numbers.Number)):
return array
elif ak._util.py27 and isinstance(array, ak._util.unicode):
return array
elif isinstance(array, np.ndarray):
return array.tolist()
elif isinstance(array, ak.behaviors.string.ByteBehavior):
return array.__bytes__()
elif isinstance(array, ak.behaviors.string.CharBehavior):
return array.__str__()
elif ak.operations.describe.parameters(array).get("__array__") == "byte":
return ak.behaviors.string.CharBehavior(array).__bytes__()
elif ak.operations.describe.parameters(array).get("__array__") == "char":
return ak.behaviors.string.CharBehavior(array).__str__()
elif isinstance(array, np.datetime64) or isinstance(array, np.timedelta64):
return array
elif isinstance(array, ak.highlevel.Array):
return [to_list(x) for x in array]
elif isinstance(array, ak.highlevel.Record):
return to_list(array.layout)
elif isinstance(array, ak.highlevel.ArrayBuilder):
return to_list(array.snapshot())
elif isinstance(array, ak.layout.Record) and array.istuple:
return tuple(to_list(x) for x in array.fields())
elif isinstance(array, ak.layout.Record):
return {n: to_list(x) for n, x in array.fielditems()}
elif isinstance(array, ak.layout.ArrayBuilder):
return [to_list(x) for x in array.snapshot()]
elif isinstance(array, ak.layout.NumpyArray):
if array.format.upper().startswith("M"):
return (
[
x
for x in ak.nplike.of(array)
.asarray(array.view_int64)
.view(array.format)
]
# FIXME: .tolist() returns
# [[1567416600000000000], [1568367000000000000], [1569096000000000000]]
# instead of [numpy.datetime64('2019-09-02T09:30:00'), numpy.datetime64('2019-09-13T09:30:00'), numpy.datetime64('2019-09-21T20:00:00')]
# see test_from_pandas() test
)
else:
return ak.nplike.of(array).asarray(array).tolist()
elif isinstance(array, (ak.layout.Content, ak.partition.PartitionedArray)):
return [to_list(x) for x in array]
elif isinstance(array, ak._v2.contents.Content):
import awkward._v2.tmp_for_testing
return to_list(awkward._v2.tmp_for_testing.v2_to_v1(array))
elif isinstance(array, dict):
return dict((n, to_list(x)) for n, x in array.items())
elif isinstance(array, Iterable):
return [to_list(x) for x in array]
else:
raise TypeError(
"unrecognized array type: {0}".format(type(array))
+ ak._util.exception_suffix(__file__)
)
_maybe_json_str = re.compile(r"^\s*(\[|\{|\"|[0-9]|true|false|null)")
_maybe_json_bytes = re.compile(br"^\s*(\[|\{|\"|[0-9]|true|false|null)")
def from_json(
source,
nan_string=None,
infinity_string=None,