Repository navigation
Expand file tree
/
Copy pathuser_defined.py
More file actions
1406 lines (1156 loc) · 51.8 KB
/
Copy pathuser_defined.py
File metadata and controls
1406 lines (1156 loc) · 51.8 KB
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
# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements. See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership. The ASF licenses this file
# to you under the Apache License, Version 2.0 (the
# "License"); you may not use this file except in compliance
# with the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
# KIND, either express or implied. See the License for the
# specific language governing permissions and limitations
# under the License.
"""Provides the user-defined functions for evaluation of dataframes."""
from __future__ import annotations
import functools
import sys
from abc import ABCMeta, abstractmethod
from enum import Enum
from typing import TYPE_CHECKING, Any, Protocol, TypeGuard, TypeVar, cast, overload
import pyarrow as pa
import datafusion._internal as df_internal
from datafusion import SessionContext
from datafusion.expr import Expr
# Imported at runtime so ``typing.get_type_hints`` resolves the capsule
# overloads; typing_extensions is a runtime dependency below 3.13.
if sys.version_info >= (3, 13):
from types import CapsuleType as _PyCapsule
else:
from typing_extensions import CapsuleType as _PyCapsule
if TYPE_CHECKING:
_R = TypeVar("_R", bound=pa.Array)
from collections.abc import Callable, Sequence
class Volatility(Enum):
"""Defines how stable or volatile a function is.
When setting the volatility of a function, you can either pass this
enumeration or a ``str``. The ``str`` equivalent is the lower case value of the
name (`"immutable"`, `"stable"`, or `"volatile"`).
"""
Immutable = 1
"""An immutable function will always return the same output when given the
same input.
DataFusion will attempt to inline immutable functions during planning.
"""
Stable = 2
"""
Returns the same value for a given input within a single queries.
A stable function may return different values given the same input across
different queries but must return the same value for a given input within a
query. An example of this is the ``Now`` function. DataFusion will attempt to
inline ``Stable`` functions during planning, when possible. For query
``select col1, now() from t1``, it might take a while to execute but ``now()``
column will be the same for each output row, which is evaluated during
planning.
"""
Volatile = 3
"""A volatile function may change the return value from evaluation to
evaluation.
Multiple invocations of a volatile function may return different results
when used in the same query. An example of this is the random() function.
DataFusion can not evaluate such functions during planning. In the query
``select col1, random() from t1``, ``random()`` function will be evaluated
for each output row, resulting in a unique random value for each row.
"""
def __str__(self) -> str:
"""Returns the string equivalent."""
return self.name.lower()
def data_type_or_field_to_field(value: pa.DataType | pa.Field, name: str) -> pa.Field:
"""Helper function to return a Field from either a Field or DataType."""
if isinstance(value, pa.Field):
return value
return pa.field(name, type=value)
def data_types_or_fields_to_field_list(
inputs: Sequence[pa.Field | pa.DataType] | pa.Field | pa.DataType,
) -> list[pa.Field]:
"""Helper function to return a list of Fields."""
if isinstance(inputs, pa.DataType):
return [pa.field("value", type=inputs)]
if isinstance(inputs, pa.Field):
return [inputs]
return [
data_type_or_field_to_field(v, f"value_{idx}") for (idx, v) in enumerate(inputs)
]
class ScalarUDFExportable(Protocol):
"""Type hint for object that has __datafusion_scalar_udf__ PyCapsule."""
def __datafusion_scalar_udf__(self) -> object: ... # noqa: D105
def _is_pycapsule(value: object) -> TypeGuard[_PyCapsule]:
"""Return ``True`` when ``value`` is a CPython ``PyCapsule``."""
return value.__class__.__name__ == "PyCapsule"
class LogicalExtensionCodecExportable(Protocol):
"""Type hint for objects exposing ``__datafusion_logical_extension_codec__``.
Implement the codec itself exactly as you would for a session that installs
only yours. A session may hold several codecs, but each payload records the
codec that wrote it and is only ever handed back to that codec, so there is
no need to recognise or reject another library's payloads.
A serialized plan records which codec wrote each payload, as a short id
taken from your class's module and qualified name. An optional
``__datafusion_codec_id__`` attribute pins that id instead. It is not part
of this protocol and is rarely needed: declare it when renaming your class
must not stop older plans from decoding, or when one library installs two
instances that own disjoint slices of the wire format.
Args:
session: A handle on the session this codec is being installed on. Take
the task context provider from it rather than building a session of
your own, so the decode callbacks resolve names against the session
that runs the query.
**Duck-type it; do not check its type.** This argument is the
reference for every capsule getter in the protocol, and across them
it is not always a session at all: the provider and catalog getters
receive the host's logical codec as a bare ``PyCapsule``. Even when
it *is* a session it is the PyO3 context from
``datafusion._internal``, not the
:py:class:`~datafusion.context.SessionContext` wrapper, so it
exposes every capsule getter and ``__datafusion_codec_id__`` —
everything the protocol asks of it — but
``isinstance(session, SessionContext)`` is ``False`` even though its
``repr`` reads ``datafusion.SessionContext``. Pass it to
``ffi_logical_codec_from_pycapsule``, which handles both.
The two bundle hooks are the exception:
:py:class:`~datafusion.extensions.SessionComponentsExportable` and
:py:class:`~datafusion.extensions.SessionPlannerExportable` are
dispatched from Python and receive the wrapper. See
:ref:`extension_getter_argument`.
Returns:
A ``datafusion_logical_extension_codec`` PyCapsule.
Examples:
A :py:class:`~datafusion.context.SessionContext` satisfies this
protocol, which is how you obtain a real capsule to test against:
>>> from datafusion import SessionContext
>>> ctx = SessionContext()
>>> type(ctx.__datafusion_logical_extension_codec__(ctx)).__name__
'PyCapsule'
"""
def __datafusion_logical_extension_codec__( # noqa: D105
self, session: Any
) -> object: ...
class PhysicalExtensionCodecExportable(Protocol):
"""Type hint for objects exposing ``__datafusion_physical_extension_codec__``.
See :py:class:`LogicalExtensionCodecExportable` for ``session``, for why a
codec need not recognise other libraries' payloads, and for
``__datafusion_codec_id__``.
"""
def __datafusion_physical_extension_codec__( # noqa: D105
self, session: Any
) -> object: ...
class ScalarUDF:
"""Class for performing scalar user-defined functions (UDF).
Scalar UDFs operate on a row by row basis. See also :py:class:`AggregateUDF` for
operating on a group of rows.
"""
@overload
def __init__(
self,
name: str,
func: Callable[..., _R],
input_fields: list[pa.Field],
return_field: pa.Field,
volatility: Volatility | str,
) -> None: ...
@overload
def __init__(
self,
name: str,
func: ScalarUDFExportable,
input_fields: None = ...,
return_field: None = ...,
volatility: None = ...,
) -> None: ...
def __init__(
self,
name: str,
func: Callable[..., _R] | ScalarUDFExportable,
input_fields: list[pa.Field] | None = None,
return_field: pa.Field | None = None,
volatility: Volatility | str | None = None,
) -> None:
"""Instantiate a scalar user-defined function (UDF).
See helper method :py:func:`udf` for argument details.
"""
if hasattr(func, "__datafusion_scalar_udf__"):
self._udf = df_internal.ScalarUDF.from_pycapsule(func)
return
if input_fields is None or return_field is None or volatility is None:
msg = (
"`input_fields`, `return_field`, and `volatility` "
"must be provided when `func` is callable."
)
raise TypeError(msg)
if isinstance(input_fields, pa.DataType):
input_fields = [input_fields]
self._udf = df_internal.ScalarUDF(
name, func, input_fields, return_field, str(volatility)
)
@classmethod
def _from_internal(cls, internal: df_internal.ScalarUDF) -> ScalarUDF:
"""Wrap an already-constructed internal ``ScalarUDF`` handle.
Used by :py:meth:`SessionContext.udf` and :py:meth:`from_pycapsule`
to wrap a handle from the session's function registry or an FFI
capsule without re-running :py:meth:`__init__`.
"""
wrapper = cls.__new__(cls)
wrapper._udf = internal
return wrapper
@property
def name(self) -> str:
"""Return the registered name of this UDF.
For UDFs imported via the FFI capsule protocol, this is the
name the capsule itself reports — not the ``name`` argument
passed to the constructor (which is ignored on the FFI path).
Examples:
>>> import pyarrow as pa
>>> from datafusion import udf
>>> double = udf(
... lambda arr: pa.array([(v.as_py() or 0) * 2 for v in arr]),
... [pa.int64()],
... pa.int64(),
... volatility="immutable",
... name="double",
... )
>>> double.name
'double'
"""
return self._udf.name
def __repr__(self) -> str:
"""Print a string representation of the Scalar UDF."""
return self._udf.__repr__()
def __call__(self, *args: Expr) -> Expr:
"""Execute the UDF.
This function is not typically called by an end user. These calls will
occur during the evaluation of the dataframe.
"""
args_raw = [arg.expr for arg in args]
return Expr(self._udf.__call__(*args_raw))
@overload
@staticmethod
def udf(
input_fields: Sequence[pa.DataType | pa.Field] | pa.DataType | pa.Field,
return_field: pa.DataType | pa.Field,
volatility: Volatility | str,
name: str | None = None,
) -> Callable[..., ScalarUDF]: ...
@overload
@staticmethod
def udf(
func: Callable[..., _R],
input_fields: Sequence[pa.DataType | pa.Field] | pa.DataType | pa.Field,
return_field: pa.DataType | pa.Field,
volatility: Volatility | str,
name: str | None = None,
) -> ScalarUDF: ...
@overload
@staticmethod
def udf(func: ScalarUDFExportable) -> ScalarUDF: ...
@overload
@staticmethod
def udf(func: _PyCapsule) -> ScalarUDF: ...
@staticmethod
def udf(*args: Any, **kwargs: Any): # noqa: D417
"""Create a new User-Defined Function (UDF).
This class can be used both as either a function or a decorator.
Usage:
- As a function: ``udf(func, input_fields, return_field,
volatility, name)``.
- As a decorator: ``@udf(input_fields, return_field, volatility, name)``.
When used a decorator, do **not** pass ``func`` explicitly.
In lieu of passing a PyArrow Field, you can pass a DataType for simplicity.
When you do so, it will be assumed that the nullability of the inputs and
output are True and that they have no metadata.
Args:
func (Callable, optional): Only needed when calling as a function.
Skip this argument when using `udf` as a decorator. If you have a Rust
backed ScalarUDF within a PyCapsule, you can pass this parameter
and ignore the rest. They will be determined directly from the
underlying function. See the online documentation for more information.
input_fields (list[pa.Field | pa.DataType]): The data types or Fields
of the arguments to ``func``. This list must be of the same length
as the number of arguments.
return_field (pa.DataType | pa.Field): The field of the return value
from the function.
volatility (Volatility | str): See `Volatility` for allowed values.
name (Optional[str]): A descriptive name for the function.
Returns:
A user-defined function that can be used in SQL expressions,
data aggregation, or window function calls.
Examples:
Using ``udf`` as a function:
>>> import pyarrow.compute as pc
>>> from datafusion.user_defined import ScalarUDF
>>> def double_func(x):
... return pc.multiply(x, 2)
>>> double_udf = ScalarUDF.udf(
... double_func, [pa.int64()], pa.int64(),
... "volatile", "double_it")
Using ``udf`` as a decorator:
>>> @ScalarUDF.udf([pa.int64()], pa.int64(), "volatile")
... def decorator_double_udf(x):
... return pc.multiply(x, 3)
Apply to a dataframe:
>>> ctx = dfn.SessionContext()
>>> df = ctx.from_pydict({"x": [1, 2, 3]})
>>> df.select(double_udf(col("x")).alias("result")).to_pydict()
{'result': [2, 4, 6]}
>>> df.select(decorator_double_udf(col("x")).alias("result")).to_pydict()
{'result': [3, 6, 9]}
"""
def _function(
func: Callable[..., _R],
input_fields: Sequence[pa.DataType | pa.Field] | pa.DataType | pa.Field,
return_field: pa.DataType | pa.Field,
volatility: Volatility | str,
name: str | None = None,
) -> ScalarUDF:
if not callable(func):
msg = "`func` argument must be callable"
raise TypeError(msg)
if name is None:
name = getattr(func, "__qualname__", func.__class__.__name__).lower()
input_fields = data_types_or_fields_to_field_list(input_fields)
return_field = data_type_or_field_to_field(return_field, "value")
return ScalarUDF(
name=name,
func=func,
input_fields=input_fields,
return_field=return_field,
volatility=volatility,
)
def _decorator(
input_fields: Sequence[pa.DataType | pa.Field] | pa.DataType | pa.Field,
return_field: pa.DataType | pa.Field,
volatility: Volatility | str,
name: str | None = None,
) -> Callable:
def decorator(func: Callable) -> Callable:
udf_caller = ScalarUDF.udf(
func, input_fields, return_field, volatility, name
)
@functools.wraps(func)
def wrapper(*args: Any, **kwargs: Any) -> Expr:
return udf_caller(*args, **kwargs)
return wrapper
return decorator
if not args and "func" in kwargs:
args = (kwargs.pop("func"),)
if args and (
hasattr(args[0], "__datafusion_scalar_udf__") or _is_pycapsule(args[0])
):
return ScalarUDF.from_pycapsule(args[0])
if args and callable(args[0]):
# Case 1: Used as a function, require the first parameter to be callable
return _function(*args, **kwargs)
# Case 2: Used as a decorator with parameters
return _decorator(*args, **kwargs)
@staticmethod
def from_pycapsule(func: ScalarUDFExportable | _PyCapsule) -> ScalarUDF:
"""Create a Scalar UDF from ScalarUDF PyCapsule object.
This function will instantiate a Scalar UDF that uses a DataFusion
ScalarUDF that is exported via the FFI bindings.
"""
if _is_pycapsule(func):
return ScalarUDF._from_internal(df_internal.ScalarUDF.from_pycapsule(func))
func = cast("ScalarUDFExportable", func)
name = str(func.__class__)
return ScalarUDF(
name=name,
func=func,
input_fields=None,
return_field=None,
volatility=None,
)
class Accumulator(metaclass=ABCMeta):
"""Defines how an :py:class:`AggregateUDF` accumulates values."""
@abstractmethod
def state(self) -> list[pa.Scalar]:
"""Return the current state.
While this function template expects PyArrow Scalar values return type,
you can return any value that can be converted into a Scalar. This
includes basic Python data types such as integers and strings. In
addition to primitive types, we currently support PyArrow, nanoarrow,
and arro3 objects in addition to primitive data types. Other objects
that support the Arrow FFI standard will be given a "best attempt" at
conversion to scalar objects.
"""
@abstractmethod
def update(self, *values: pa.Array) -> None:
"""Evaluate an array of values and update state."""
@abstractmethod
def merge(self, states: list[pa.Array]) -> None:
"""Merge a set of states."""
@abstractmethod
def evaluate(self) -> pa.Scalar:
"""Return the resultant value.
While this function template expects a PyArrow Scalar value return type,
you can return any value that can be converted into a Scalar. This
includes basic Python data types such as integers and strings. In
addition to primitive types, we currently support PyArrow, nanoarrow,
and arro3 objects in addition to primitive data types. Other objects
that support the Arrow FFI standard will be given a "best attempt" at
conversion to scalar objects.
"""
class AggregateUDFExportable(Protocol):
"""Type hint for object that has __datafusion_aggregate_udf__ PyCapsule."""
def __datafusion_aggregate_udf__(self) -> object: ... # noqa: D105
class AggregateUDF:
"""Class for performing scalar user-defined functions (UDF).
Aggregate UDFs operate on a group of rows and return a single value. See
also :py:class:`ScalarUDF` for operating on a row by row basis.
"""
@overload
def __init__(
self,
name: str,
accumulator: Callable[[], Accumulator],
input_types: list[pa.DataType],
return_type: pa.DataType,
state_type: list[pa.DataType],
volatility: Volatility | str,
) -> None: ...
@overload
def __init__(
self,
name: str,
accumulator: AggregateUDFExportable,
input_types: None = ...,
return_type: None = ...,
state_type: None = ...,
volatility: None = ...,
) -> None: ...
def __init__(
self,
name: str,
accumulator: Callable[[], Accumulator] | AggregateUDFExportable,
input_types: list[pa.DataType] | None = None,
return_type: pa.DataType | None = None,
state_type: list[pa.DataType] | None = None,
volatility: Volatility | str | None = None,
) -> None:
"""Instantiate a user-defined aggregate function (UDAF).
See :py:func:`udaf` for a convenience function and argument
descriptions.
"""
if hasattr(accumulator, "__datafusion_aggregate_udf__"):
self._udaf = df_internal.AggregateUDF.from_pycapsule(accumulator)
return
if (
input_types is None
or return_type is None
or state_type is None
or volatility is None
):
msg = (
"`input_types`, `return_type`, `state_type`, and `volatility` "
"must be provided when `accumulator` is callable."
)
raise TypeError(msg)
self._udaf = df_internal.AggregateUDF(
name,
accumulator,
input_types,
return_type,
state_type,
str(volatility),
)
@classmethod
def _from_internal(cls, internal: df_internal.AggregateUDF) -> AggregateUDF:
"""Wrap an already-constructed internal ``AggregateUDF`` handle.
Used by :py:meth:`SessionContext.udaf` and :py:meth:`from_pycapsule`
to wrap a handle from the session's function registry or an FFI
capsule without re-running :py:meth:`__init__`.
"""
wrapper = cls.__new__(cls)
wrapper._udaf = internal
return wrapper
@property
def name(self) -> str:
"""Return the registered name of this UDAF.
For UDAFs imported via the FFI capsule protocol, this is the
name the capsule itself reports — not the ``name`` argument
passed to the constructor (which is ignored on the FFI path).
"""
return self._udaf.name
def __repr__(self) -> str:
"""Print a string representation of the Aggregate UDF."""
return self._udaf.__repr__()
def __call__(self, *args: Expr) -> Expr:
"""Execute the UDAF.
This function is not typically called by an end user. These calls will
occur during the evaluation of the dataframe.
"""
args_raw = [arg.expr for arg in args]
return Expr(self._udaf.__call__(*args_raw))
@overload
@staticmethod
def udaf(
input_types: pa.DataType | list[pa.DataType],
return_type: pa.DataType,
state_type: list[pa.DataType],
volatility: Volatility | str,
name: str | None = None,
) -> Callable[..., AggregateUDF]: ...
@overload
@staticmethod
def udaf(
accum: Callable[[], Accumulator],
input_types: pa.DataType | list[pa.DataType],
return_type: pa.DataType,
state_type: list[pa.DataType],
volatility: Volatility | str,
name: str | None = None,
) -> AggregateUDF: ...
@overload
@staticmethod
def udaf(accum: AggregateUDFExportable) -> AggregateUDF: ...
@overload
@staticmethod
def udaf(accum: _PyCapsule) -> AggregateUDF: ...
@staticmethod
def udaf(*args: Any, **kwargs: Any): # noqa: D417, C901
"""Create a new User-Defined Aggregate Function (UDAF).
This class allows you to define an aggregate function that can be used in
data aggregation or window function calls.
Usage:
- As a function: ``udaf(accum, input_types, return_type, state_type, volatility, name)``.
- As a decorator: ``@udaf(input_types, return_type, state_type, volatility, name)``.
When using ``udaf`` as a decorator, do not pass ``accum`` explicitly.
If your :py:class:`Accumulator` can be instantiated with no arguments, you
can simply pass its type as ``accum``. If you need to pass additional
arguments to its constructor, you can define a lambda or a factory method.
During runtime the :py:class:`Accumulator` will be constructed for every
instance in which this UDAF is used.
Examples:
>>> import pyarrow.compute as pc
>>> from datafusion.user_defined import AggregateUDF, Accumulator, udaf
>>> class Summarize(Accumulator):
... def __init__(self, bias: float = 0.0):
... self._sum = pa.scalar(bias)
... def state(self):
... return [self._sum]
... def update(self, values):
... self._sum = pa.scalar(
... self._sum.as_py() + pc.sum(values).as_py())
... def merge(self, states):
... self._sum = pa.scalar(
... self._sum.as_py() + pc.sum(states[0]).as_py())
... def evaluate(self):
... return self._sum
Using ``udaf`` as a function:
>>> udaf1 = AggregateUDF.udaf(
... Summarize, pa.float64(), pa.float64(),
... [pa.float64()], "immutable")
Wrapping ``udaf`` with a function:
>>> def sum_bias_10() -> Summarize:
... return Summarize(10.0)
>>> udaf2 = udaf(sum_bias_10, pa.float64(), pa.float64(), [pa.float64()],
... "immutable")
Using ``udaf`` with lambda:
>>> udaf3 = udaf(lambda: Summarize(20.0), pa.float64(), pa.float64(),
... [pa.float64()], "immutable")
Using ``udaf`` as a decorator:
>>> @AggregateUDF.udaf(
... pa.float64(), pa.float64(),
... [pa.float64()], "immutable")
... def udaf4():
... return Summarize(10.0)
Apply to a dataframe:
>>> ctx = dfn.SessionContext()
>>> df = ctx.from_pydict({"a": [1.0, 2.0, 3.0]})
>>> df.aggregate([], [udaf1(col("a")).alias("total")]).collect_column(
... "total")[0].as_py()
6.0
>>> df.aggregate([], [udaf2(col("a")).alias("total")]).collect_column(
... "total")[0].as_py()
16.0
>>> df.aggregate([], [udaf3(col("a")).alias("total")]).collect_column(
... "total")[0].as_py()
26.0
>>> df.aggregate([], [udaf4(col("a")).alias("total")]).collect_column(
... "total")[0].as_py()
16.0
Args:
accum: The accumulator python function. Only needed when calling as a
function. Skip this argument when using ``udaf`` as a decorator.
If you have a Rust backed AggregateUDF within a PyCapsule, you can
pass this parameter and ignore the rest. They will be determined
directly from the underlying function. See the online documentation
for more information.
input_types: The data types of the arguments to ``accum``.
return_type: The data type of the return value.
state_type: The data types of the intermediate accumulation.
volatility: See :py:class:`Volatility` for allowed values.
name: A descriptive name for the function.
Returns:
A user-defined aggregate function, which can be used in either data
aggregation or window function calls.
""" # noqa: E501 W505
def _function(
accum: Callable[[], Accumulator],
input_types: pa.DataType | list[pa.DataType],
return_type: pa.DataType,
state_type: list[pa.DataType],
volatility: Volatility | str,
name: str | None = None,
) -> AggregateUDF:
if not callable(accum):
msg = "`func` must be callable."
raise TypeError(msg)
if not isinstance(accum(), Accumulator):
msg = "Accumulator must implement the abstract base class Accumulator"
raise TypeError(msg)
if name is None:
name = accum().__class__.__qualname__.lower()
if isinstance(input_types, pa.DataType):
input_types = [input_types]
return AggregateUDF(
name=name,
accumulator=accum,
input_types=input_types,
return_type=return_type,
state_type=state_type,
volatility=volatility,
)
def _decorator(
input_types: pa.DataType | list[pa.DataType],
return_type: pa.DataType,
state_type: list[pa.DataType],
volatility: Volatility | str,
name: str | None = None,
) -> Callable[..., Callable[..., Expr]]:
def decorator(accum: Callable[[], Accumulator]) -> Callable[..., Expr]:
udaf_caller = AggregateUDF.udaf(
accum, input_types, return_type, state_type, volatility, name
)
@functools.wraps(accum)
def wrapper(*args: Any, **kwargs: Any) -> Expr:
return udaf_caller(*args, **kwargs)
return wrapper
return decorator
if not args and "accum" in kwargs:
args = (kwargs.pop("accum"),)
if args and (
hasattr(args[0], "__datafusion_aggregate_udf__") or _is_pycapsule(args[0])
):
return AggregateUDF.from_pycapsule(args[0])
if args and callable(args[0]):
# Case 1: Used as a function, require the first parameter to be callable
return _function(*args, **kwargs)
# Case 2: Used as a decorator with parameters
return _decorator(*args, **kwargs)
@staticmethod
def from_pycapsule(func: AggregateUDFExportable | _PyCapsule) -> AggregateUDF:
"""Create an Aggregate UDF from AggregateUDF PyCapsule object.
This function will instantiate a Aggregate UDF that uses a DataFusion
AggregateUDF that is exported via the FFI bindings.
"""
if _is_pycapsule(func):
return AggregateUDF._from_internal(
df_internal.AggregateUDF.from_pycapsule(func)
)
capsule = cast("AggregateUDFExportable", func)
name = str(capsule.__class__)
return AggregateUDF(
name=name,
accumulator=capsule,
input_types=None,
return_type=None,
state_type=None,
volatility=None,
)
class WindowEvaluator:
"""Evaluator class for user-defined window functions (UDWF).
It is up to the user to decide which evaluate function is appropriate.
+------------------------+--------------------------------+------------------+---------------------------+
| ``uses_window_frame`` | ``supports_bounded_execution`` | ``include_rank`` | function_to_implement |
+========================+================================+==================+===========================+
| False (default) | False (default) | False (default) | ``evaluate_all`` |
+------------------------+--------------------------------+------------------+---------------------------+
| False | True | False | ``evaluate`` |
+------------------------+--------------------------------+------------------+---------------------------+
| False | True/False | True | ``evaluate_all_with_rank``|
+------------------------+--------------------------------+------------------+---------------------------+
| True | True/False | True/False | ``evaluate`` |
+------------------------+--------------------------------+------------------+---------------------------+
""" # noqa: W505, E501
def memoize(self) -> None:
"""Perform a memoize operation to improve performance.
When the window frame has a fixed beginning (e.g UNBOUNDED
PRECEDING), some functions such as FIRST_VALUE and
NTH_VALUE do not need the (unbounded) input once they have
seen a certain amount of input.
`memoize` is called after each input batch is processed, and
such functions can save whatever they need
"""
def get_range(self, idx: int, num_rows: int) -> tuple[int, int]: # noqa: ARG002
"""Return the range for the window function.
If `uses_window_frame` flag is `false`. This method is used to
calculate required range for the window function during
stateful execution.
Generally there is no required range, hence by default this
returns smallest range(current row). e.g seeing current row is
enough to calculate window result (such as row_number, rank,
etc)
Args:
idx:: Current index
num_rows: Number of rows.
"""
return (idx, idx + 1)
def is_causal(self) -> bool:
"""Get whether evaluator needs future data for its result."""
return False
def evaluate_all(self, values: list[pa.Array], num_rows: int) -> pa.Array:
"""Evaluate a window function on an entire input partition.
This function is called once per input *partition* for window functions that
*do not use* values from the window frame, such as
:py:func:`~datafusion.functions.row_number`,
:py:func:`~datafusion.functions.rank`,
:py:func:`~datafusion.functions.dense_rank`,
:py:func:`~datafusion.functions.percent_rank`,
:py:func:`~datafusion.functions.cume_dist`,
:py:func:`~datafusion.functions.lead`,
and :py:func:`~datafusion.functions.lag`.
It produces the result of all rows in a single pass. It
expects to receive the entire partition as the ``value`` and
must produce an output column with one output row for every
input row.
``num_rows`` is required to correctly compute the output in case
``len(values) == 0``
Implementing this function is an optimization. Certain window
functions are not affected by the window frame definition or
the query doesn't have a frame, and ``evaluate`` skips the
(costly) window frame boundary calculation and the overhead of
calling ``evaluate`` for each output row.
For example, the `LAG` built in window function does not use
the values of its window frame (it can be computed in one shot
on the entire partition with ``Self::evaluate_all`` regardless of the
window defined in the ``OVER`` clause)
.. code-block:: text
lag(x, 1) OVER (ORDER BY z ROWS BETWEEN 2 PRECEDING AND 3 FOLLOWING)
However, ``avg()`` computes the average in the window and thus
does use its window frame.
.. code-block:: text
avg(x) OVER (PARTITION BY y ORDER BY z ROWS BETWEEN 2 PRECEDING AND 3 FOLLOWING)
""" # noqa: W505, E501
def evaluate(
self, values: list[pa.Array], eval_range: tuple[int, int]
) -> pa.Scalar:
"""Evaluate window function on a range of rows in an input partition.
This is the simplest and most general function to implement
but also the least performant as it creates output one row at
a time. It is typically much faster to implement stateful
evaluation using one of the other specialized methods on this
trait.
Returns a [`ScalarValue`] that is the value of the window
function within `range` for the entire partition. Argument
`values` contains the evaluation result of function arguments
and evaluation results of ORDER BY expressions. If function has a
single argument, `values[1..]` will contain ORDER BY expression results.
"""
def evaluate_all_with_rank(
self, num_rows: int, ranks_in_partition: list[tuple[int, int]]
) -> pa.Array:
"""Called for window functions that only need the rank of a row.
Evaluate the partition evaluator against the partition using
the row ranks. For example, ``rank(col("a"))`` produces
.. code-block:: text
a | rank
- + ----
A | 1
A | 1
C | 3
D | 4
D | 4
For this case, `num_rows` would be `5` and the
`ranks_in_partition` would be called with
.. code-block:: text
[
(0,1),
(2,2),
(3,4),
]
The user must implement this method if ``include_rank`` returns True.
"""
def supports_bounded_execution(self) -> bool:
"""Can the window function be incrementally computed using bounded memory?"""
return False
def uses_window_frame(self) -> bool:
"""Does the window function use the values from the window frame?"""
return False
def include_rank(self) -> bool:
"""Can this function be evaluated with (only) rank?"""
return False
class WindowUDFExportable(Protocol):
"""Type hint for object that has __datafusion_window_udf__ PyCapsule."""
def __datafusion_window_udf__(self) -> object: ... # noqa: D105
class WindowUDF:
"""Class for performing window user-defined functions (UDF).
Window UDFs operate on a partition of rows. See
also :py:class:`ScalarUDF` for operating on a row by row basis.
"""
@overload
def __init__(
self,
name: str,
func: Callable[[], WindowEvaluator],
input_types: list[pa.DataType],
return_type: pa.DataType,
volatility: Volatility | str,
) -> None: ...
@overload
def __init__(
self,
name: str,
func: WindowUDFExportable,
input_types: None = ...,