-
Notifications
You must be signed in to change notification settings - Fork 651
/
dataframe.py
2989 lines (2697 loc) · 109 KB
/
dataframe.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
# Licensed to Modin Development Team under one or more contributor license agreements.
# See the NOTICE file distributed with this work for additional information regarding
# copyright ownership. The Modin Development Team 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.
"""
Module contains class PandasDataframe.
PandasDataframe is a parent abstract class for any dataframe class
for pandas storage format.
"""
from collections import OrderedDict
import numpy as np
import pandas
import datetime
from pandas.core.indexes.api import ensure_index, Index, RangeIndex
from pandas.core.dtypes.common import is_numeric_dtype, is_list_like
from pandas._libs.lib import no_default
from typing import List, Hashable, Optional, Callable, Union, Dict
from modin.core.storage_formats.pandas.query_compiler import PandasQueryCompiler
from modin.error_message import ErrorMessage
from modin.core.storage_formats.pandas.parsers import (
find_common_type_cat as find_common_type,
)
from modin.core.dataframe.base.dataframe.dataframe import ModinDataframe
from modin.core.dataframe.base.dataframe.utils import (
Axis,
JoinType,
)
from modin.pandas.indexing import is_range_like
from modin.pandas.utils import is_full_grab_slice, check_both_not_none
from modin.logging import ClassLogger
def lazy_metadata_decorator(apply_axis=None, axis_arg=-1, transpose=False):
"""
Lazily propagate metadata for the ``PandasDataframe``.
This decorator first adds the minimum required reindexing operations
to each partition's queue of functions to be lazily applied for
each PandasDataframe in the arguments by applying the function
run_f_on_minimally_updated_metadata. The decorator also sets the
flags for deferred metadata synchronization on the function result
if necessary.
Parameters
----------
apply_axis : str, default: None
The axes on which to apply the reindexing operations to the `self._partitions` lazily.
Case None: No lazy metadata propagation.
Case "both": Add reindexing operations on both axes to partition queue.
Case "opposite": Add reindexing operations complementary to given axis.
Case "rows": Add reindexing operations on row axis to partition queue.
axis_arg : int, default: -1
The index or column axis.
transpose : bool, default: False
Boolean for if a transpose operation is being used.
Returns
-------
Wrapped Function.
"""
def decorator(f):
from functools import wraps
@wraps(f)
def run_f_on_minimally_updated_metadata(self, *args, **kwargs):
for obj in (
[self]
+ [o for o in args if isinstance(o, PandasDataframe)]
+ [v for v in kwargs.values() if isinstance(v, PandasDataframe)]
+ [
d
for o in args
if isinstance(o, list)
for d in o
if isinstance(d, PandasDataframe)
]
+ [
d
for _, o in kwargs.items()
if isinstance(o, list)
for d in o
if isinstance(d, PandasDataframe)
]
):
if apply_axis == "both":
if obj._deferred_index and obj._deferred_column:
obj._propagate_index_objs(axis=None)
elif obj._deferred_index:
obj._propagate_index_objs(axis=0)
elif obj._deferred_column:
obj._propagate_index_objs(axis=1)
elif apply_axis == "opposite":
if "axis" not in kwargs:
axis = args[axis_arg]
else:
axis = kwargs["axis"]
if axis == 0 and obj._deferred_column:
obj._propagate_index_objs(axis=1)
elif axis == 1 and obj._deferred_index:
obj._propagate_index_objs(axis=0)
elif apply_axis == "rows":
obj._propagate_index_objs(axis=0)
result = f(self, *args, **kwargs)
if apply_axis is None and not transpose:
result._deferred_index = self._deferred_index
result._deferred_column = self._deferred_column
elif apply_axis is None and transpose:
result._deferred_index = self._deferred_column
result._deferred_column = self._deferred_index
elif apply_axis == "opposite":
if axis == 0:
result._deferred_index = self._deferred_index
else:
result._deferred_column = self._deferred_column
elif apply_axis == "rows":
result._deferred_column = self._deferred_column
return result
return run_f_on_minimally_updated_metadata
return decorator
class PandasDataframe(ClassLogger):
"""
An abstract class that represents the parent class for any pandas storage format dataframe class.
This class provides interfaces to run operations on dataframe partitions.
Parameters
----------
partitions : np.ndarray
A 2D NumPy array of partitions.
index : sequence
The index for the dataframe. Converted to a ``pandas.Index``.
columns : sequence
The columns object for the dataframe. Converted to a ``pandas.Index``.
row_lengths : list, optional
The length of each partition in the rows. The "height" of
each of the block partitions. Is computed if not provided.
column_widths : list, optional
The width of each partition in the columns. The "width" of
each of the block partitions. Is computed if not provided.
dtypes : pandas.Series, optional
The data types for the dataframe columns.
"""
_partition_mgr_cls = None
_query_compiler_cls = PandasQueryCompiler
# These properties flag whether or not we are deferring the metadata synchronization
_deferred_index = False
_deferred_column = False
@property
def __constructor__(self):
"""
Create a new instance of this object.
Returns
-------
PandasDataframe
"""
return type(self)
def __init__(
self,
partitions,
index,
columns,
row_lengths=None,
column_widths=None,
dtypes=None,
):
self._partitions = partitions
self._index_cache = ensure_index(index)
self._columns_cache = ensure_index(columns)
if row_lengths is not None and len(self.index) > 0:
# An empty frame can have 0 rows but a nonempty index. If the frame
# does have rows, the number of rows must equal the size of the
# index.
num_rows = sum(row_lengths)
if num_rows > 0:
ErrorMessage.catch_bugs_and_request_email(
num_rows != len(self._index_cache),
"Row lengths: {} != {}".format(num_rows, len(self._index_cache)),
)
ErrorMessage.catch_bugs_and_request_email(
any(val < 0 for val in row_lengths),
"Row lengths cannot be negative: {}".format(row_lengths),
)
self._row_lengths_cache = row_lengths
if column_widths is not None and len(self.columns) > 0:
# An empty frame can have 0 column but a nonempty column index. If
# the frame does have columns, the number of columns must equal the
# size of the columns.
num_columns = sum(column_widths)
if num_columns > 0:
ErrorMessage.catch_bugs_and_request_email(
num_columns != len(self._columns_cache),
"Column widths: {} != {}".format(
num_columns, len(self._columns_cache)
),
)
ErrorMessage.catch_bugs_and_request_email(
any(val < 0 for val in column_widths),
"Column widths cannot be negative: {}".format(column_widths),
)
self._column_widths_cache = column_widths
self._dtypes = dtypes
self._filter_empties()
@property
def _row_lengths(self):
"""
Compute the row partitions lengths if they are not cached.
Returns
-------
list
A list of row partitions lengths.
"""
if self._row_lengths_cache is None:
if len(self._partitions.T) > 0:
self._row_lengths_cache = [
obj.length() for obj in self._partitions.T[0]
]
else:
self._row_lengths_cache = []
return self._row_lengths_cache
@property
def _column_widths(self):
"""
Compute the column partitions widths if they are not cached.
Returns
-------
list
A list of column partitions widths.
"""
if self._column_widths_cache is None:
if len(self._partitions) > 0:
self._column_widths_cache = [obj.width() for obj in self._partitions[0]]
else:
self._column_widths_cache = []
return self._column_widths_cache
@property
def _axes_lengths(self):
"""
Get a pair of row partitions lengths and column partitions widths.
Returns
-------
list
The pair of row partitions lengths and column partitions widths.
"""
return [self._row_lengths, self._column_widths]
@property
def dtypes(self):
"""
Compute the data types if they are not cached.
Returns
-------
pandas.Series
A pandas Series containing the data types for this dataframe.
"""
if self._dtypes is None:
self._dtypes = self._compute_dtypes()
return self._dtypes
def _compute_dtypes(self):
"""
Compute the data types via TreeReduce pattern.
Returns
-------
pandas.Series
A pandas Series containing the data types for this dataframe.
"""
def dtype_builder(df):
return df.apply(lambda col: find_common_type(col.values), axis=0)
map_func = self._build_treereduce_func(0, lambda df: df.dtypes)
reduce_func = self._build_treereduce_func(0, dtype_builder)
# For now we will use a pandas Series for the dtypes.
if len(self.columns) > 0:
dtypes = self.tree_reduce(0, map_func, reduce_func).to_pandas().iloc[0]
else:
dtypes = pandas.Series([])
# reset name to None because we use "__reduced__" internally
dtypes.name = None
return dtypes
_index_cache = None
_columns_cache = None
def _validate_set_axis(self, new_labels, old_labels):
"""
Validate the possibility of replacement of old labels with the new labels.
Parameters
----------
new_labels : list-like
The labels to replace with.
old_labels : list-like
The labels to replace.
Returns
-------
list-like
The validated labels.
"""
new_labels = ensure_index(new_labels)
old_len = len(old_labels)
new_len = len(new_labels)
if old_len != new_len:
raise ValueError(
f"Length mismatch: Expected axis has {old_len} elements, "
+ f"new values have {new_len} elements"
)
return new_labels
def _get_index(self):
"""
Get the index from the cache object.
Returns
-------
pandas.Index
An index object containing the row labels.
"""
return self._index_cache
def _get_columns(self):
"""
Get the columns from the cache object.
Returns
-------
pandas.Index
An index object containing the column labels.
"""
return self._columns_cache
def _set_index(self, new_index):
"""
Replace the current row labels with new labels.
Parameters
----------
new_index : list-like
The new row labels.
"""
if self._index_cache is None:
self._index_cache = ensure_index(new_index)
else:
new_index = self._validate_set_axis(new_index, self._index_cache)
self._index_cache = new_index
self.synchronize_labels(axis=0)
def _set_columns(self, new_columns):
"""
Replace the current column labels with new labels.
Parameters
----------
new_columns : list-like
The new column labels.
"""
if self._columns_cache is None:
self._columns_cache = ensure_index(new_columns)
else:
new_columns = self._validate_set_axis(new_columns, self._columns_cache)
self._columns_cache = new_columns
if self._dtypes is not None:
self._dtypes.index = new_columns
self.synchronize_labels(axis=1)
columns = property(_get_columns, _set_columns)
index = property(_get_index, _set_index)
@property
def axes(self):
"""
Get index and columns that can be accessed with an `axis` integer.
Returns
-------
list
List with two values: index and columns.
"""
return [self.index, self.columns]
def _compute_axis_labels(self, axis: int, partitions=None):
"""
Compute the labels for specific `axis`.
Parameters
----------
axis : int
Axis to compute labels along.
partitions : np.ndarray, optional
A 2D NumPy array of partitions from which labels will be grabbed.
If not specified, partitions will be taken from `self._partitions`.
Returns
-------
pandas.Index
Labels for the specified `axis`.
"""
if partitions is None:
partitions = self._partitions
return self._partition_mgr_cls.get_indices(
axis, partitions, lambda df: df.axes[axis]
)
def _filter_empties(self):
"""Remove empty partitions from `self._partitions` to avoid triggering excess computation."""
if len(self.axes[0]) == 0 or len(self.axes[1]) == 0:
# This is the case for an empty frame. We don't want to completely remove
# all metadata and partitions so for the moment, we won't prune if the frame
# is empty.
# TODO: Handle empty dataframes better
return
self._partitions = np.array(
[
[
self._partitions[i][j]
for j in range(len(self._partitions[i]))
if j < len(self._column_widths) and self._column_widths[j] != 0
]
for i in range(len(self._partitions))
if i < len(self._row_lengths) and self._row_lengths[i] != 0
]
)
self._column_widths_cache = [w for w in self._column_widths if w != 0]
self._row_lengths_cache = [r for r in self._row_lengths if r != 0]
def synchronize_labels(self, axis=None):
"""
Set the deferred axes variables for the ``PandasDataframe``.
Parameters
----------
axis : int, default: None
The deferred axis.
0 for the index, 1 for the columns.
"""
if axis is None:
self._deferred_index = True
self._deferred_column = True
elif axis == 0:
self._deferred_index = True
else:
self._deferred_column = True
def _propagate_index_objs(self, axis=None):
"""
Synchronize labels by applying the index object for specific `axis` to the `self._partitions` lazily.
Adds `set_axis` function to call-queue of each partition from `self._partitions`
to apply new axis.
Parameters
----------
axis : int, default: None
The axis to apply to. If it's None applies to both axes.
"""
self._filter_empties()
if axis is None or axis == 0:
cum_row_lengths = np.cumsum([0] + self._row_lengths)
if axis is None or axis == 1:
cum_col_widths = np.cumsum([0] + self._column_widths)
if axis is None:
def apply_idx_objs(df, idx, cols):
return df.set_axis(idx, axis="index", inplace=False).set_axis(
cols, axis="columns", inplace=False
)
self._partitions = np.array(
[
[
self._partitions[i][j].add_to_apply_calls(
apply_idx_objs,
idx=self.index[
slice(cum_row_lengths[i], cum_row_lengths[i + 1])
],
cols=self.columns[
slice(cum_col_widths[j], cum_col_widths[j + 1])
],
)
for j in range(len(self._partitions[i]))
]
for i in range(len(self._partitions))
]
)
self._deferred_index = False
self._deferred_column = False
elif axis == 0:
def apply_idx_objs(df, idx):
return df.set_axis(idx, axis="index", inplace=False)
self._partitions = np.array(
[
[
self._partitions[i][j].add_to_apply_calls(
apply_idx_objs,
idx=self.index[
slice(cum_row_lengths[i], cum_row_lengths[i + 1])
],
)
for j in range(len(self._partitions[i]))
]
for i in range(len(self._partitions))
]
)
self._deferred_index = False
elif axis == 1:
def apply_idx_objs(df, cols):
return df.set_axis(cols, axis="columns", inplace=False)
self._partitions = np.array(
[
[
self._partitions[i][j].add_to_apply_calls(
apply_idx_objs,
cols=self.columns[
slice(cum_col_widths[j], cum_col_widths[j + 1])
],
)
for j in range(len(self._partitions[i]))
]
for i in range(len(self._partitions))
]
)
self._deferred_column = False
else:
ErrorMessage.catch_bugs_and_request_email(
axis is not None and axis not in [0, 1]
)
@lazy_metadata_decorator(apply_axis=None)
def mask(
self,
row_labels: Optional[List[Hashable]] = None,
row_positions: Optional[List[int]] = None,
col_labels: Optional[List[Hashable]] = None,
col_positions: Optional[List[int]] = None,
) -> "PandasDataframe":
"""
Lazily select columns or rows from given indices.
Parameters
----------
row_labels : list of hashable, optional
The row labels to extract.
row_positions : list-like of ints, optional
The row positions to extract.
col_labels : list of hashable, optional
The column labels to extract.
col_positions : list-like of ints, optional
The column positions to extract.
Returns
-------
PandasDataframe
A new PandasDataframe from the mask provided.
Notes
-----
If both `row_labels` and `row_positions` are provided, a ValueError is raised.
The same rule applies for `col_labels` and `col_positions`.
"""
if check_both_not_none(row_labels, row_positions):
raise ValueError(
"Both row_labels and row_positions were provided - please provide only one of row_labels and row_positions."
)
if check_both_not_none(col_labels, col_positions):
raise ValueError(
"Both col_labels and col_positions were provided - please provide only one of col_labels and col_positions."
)
indexers = []
for axis, indexer in enumerate((row_positions, col_positions)):
if is_range_like(indexer):
if indexer.step == 1 and len(indexer) == len(self.axes[axis]):
# By this function semantics, `None` indexer is a full-axis access
indexer = None
elif indexer is not None and not isinstance(indexer, pandas.RangeIndex):
# Pure python's range is not fully compatible with a list of ints,
# converting it to ``pandas.RangeIndex``` that is compatible.
indexer = pandas.RangeIndex(
indexer.start, indexer.stop, indexer.step
)
else:
ErrorMessage.catch_bugs_and_request_email(
failure_condition=not (indexer is None or is_list_like(indexer)),
extra_log=f"Mask takes only list-like numeric indexers, received: {type(indexer)}",
)
indexers.append(indexer)
row_positions, col_positions = indexers
if (
col_labels is None
and col_positions is None
and row_labels is None
and row_positions is None
):
return self.copy()
# Get numpy array of positions of values from `row_labels`
if row_labels is not None:
row_positions = self.index.get_indexer_for(row_labels)
if row_positions is not None:
sorted_row_positions = (
row_positions
if (
(is_range_like(row_positions) and row_positions.step > 0)
# `np.sort` of empty list returns an array with `float` dtype,
# which doesn't work well as an indexer
or len(row_positions) == 0
)
else np.sort(row_positions)
)
# Get dict of row_parts as {row_index: row_internal_indices}
# TODO: Rename `row_partitions_list`->`row_partitions_dict`
row_partitions_list = self._get_dict_of_block_index(
0, sorted_row_positions, are_indices_sorted=True
)
new_row_lengths = [
len(
# Row lengths for slice are calculated as the length of the slice
# on the partition. Often this will be the same length as the current
# length, but sometimes it is different, thus the extra calculation.
range(*part_indexer.indices(self._row_lengths[part_idx]))
if isinstance(part_indexer, slice)
else part_indexer
)
for part_idx, part_indexer in row_partitions_list.items()
]
new_index = self.index[
# pandas Index is more likely to preserve its metadata if the indexer is slice
slice(row_positions.start, row_positions.stop, row_positions.step)
# TODO: Fast range processing of non-positive-step ranges is not yet supported
if is_range_like(row_positions) and row_positions.step > 0
else sorted_row_positions
]
else:
row_partitions_list = {
i: slice(None) for i in range(len(self._row_lengths))
}
new_row_lengths = self._row_lengths
new_index = self.index
# Get numpy array of positions of values from `col_labels`
if col_labels is not None:
col_positions = self.columns.get_indexer_for(col_labels)
if col_positions is not None:
sorted_col_positions = (
col_positions
if (
(is_range_like(col_positions) and col_positions.step > 0)
# `np.sort` of empty list returns an array with `float` dtype,
# which doesn't work well as an indexer
or len(col_positions) == 0
)
else np.sort(col_positions)
)
# Get dict of col_parts as {col_index: col_internal_indices}
col_partitions_list = self._get_dict_of_block_index(
1, sorted_col_positions, are_indices_sorted=True
)
new_col_widths = [
len(
# Column widths for slice are calculated as the length of the slice
# on the partition. Often this will be the same length as the current
# length, but sometimes it is different, thus the extra calculation.
range(*part_indexer.indices(self._column_widths[part_idx]))
if isinstance(part_indexer, slice)
else part_indexer
)
for part_idx, part_indexer in col_partitions_list.items()
]
# Use the slice to calculate the new columns
# TODO: Support fast processing of negative-step ranges
if is_range_like(col_positions) and col_positions.step > 0:
# pandas Index is more likely to preserve its metadata if the indexer is slice
monotonic_col_idx = slice(
col_positions.start, col_positions.stop, col_positions.step
)
else:
monotonic_col_idx = sorted_col_positions
new_columns = self.columns[monotonic_col_idx]
ErrorMessage.catch_bugs_and_request_email(
failure_condition=sum(new_col_widths) != len(new_columns),
extra_log=f"{sum(new_col_widths)} != {len(new_columns)}.\n{col_positions}\n{self._column_widths}\n{col_partitions_list}",
)
if self._dtypes is not None:
new_dtypes = self.dtypes.iloc[monotonic_col_idx]
else:
new_dtypes = None
else:
col_partitions_list = {
i: slice(None) for i in range(len(self._column_widths))
}
new_col_widths = self._column_widths
new_columns = self.columns
if self._dtypes is not None:
new_dtypes = self.dtypes
else:
new_dtypes = None
new_partitions = np.array(
[
[
self._partitions[row_idx][col_idx].mask(
row_internal_indices, col_internal_indices
)
for col_idx, col_internal_indices in col_partitions_list.items()
if isinstance(col_internal_indices, slice)
or len(col_internal_indices) > 0
]
for row_idx, row_internal_indices in row_partitions_list.items()
if isinstance(row_internal_indices, slice)
or len(row_internal_indices) > 0
]
)
intermediate = self.__constructor__(
new_partitions,
new_index,
new_columns,
new_row_lengths,
new_col_widths,
new_dtypes,
)
# Check if monotonically increasing, return if it is. Fast track code path for
# common case to keep it fast.
if (
row_positions is None
# Fast range processing of non-positive-step ranges is not yet supported
or (is_range_like(row_positions) and row_positions.step > 0)
or len(row_positions) == 1
or np.all(row_positions[1:] >= row_positions[:-1])
) and (
col_positions is None
# Fast range processing of non-positive-step ranges is not yet supported
or (is_range_like(col_positions) and col_positions.step > 0)
or len(col_positions) == 1
or np.all(col_positions[1:] >= col_positions[:-1])
):
return intermediate
# The new labels are often smaller than the old labels, so we can't reuse the
# original order values because those were mapped to the original data. We have
# to reorder here based on the expected order from within the data.
# We create a dictionary mapping the position of the numeric index with respect
# to all others, then recreate that order by mapping the new order values from
# the old. This information is sent to `_reorder_labels`.
if row_positions is not None:
row_order_mapping = dict(
zip(sorted_row_positions, range(len(row_positions)))
)
new_row_order = [row_order_mapping[idx] for idx in row_positions]
else:
new_row_order = None
if col_positions is not None:
col_order_mapping = dict(
zip(sorted_col_positions, range(len(col_positions)))
)
new_col_order = [col_order_mapping[idx] for idx in col_positions]
else:
new_col_order = None
return intermediate._reorder_labels(
row_positions=new_row_order, col_positions=new_col_order
)
@lazy_metadata_decorator(apply_axis="rows")
def from_labels(self) -> "PandasDataframe":
"""
Convert the row labels to a column of data, inserted at the first position.
Gives result by similar way as `pandas.DataFrame.reset_index`. Each level
of `self.index` will be added as separate column of data.
Returns
-------
PandasDataframe
A PandasDataframe with new columns from index labels.
"""
new_row_labels = pandas.RangeIndex(len(self.index))
if self.index.nlevels > 1:
level_names = [
self.index.names[i]
if self.index.names[i] is not None
else "level_{}".format(i)
for i in range(self.index.nlevels)
]
else:
level_names = [
self.index.names[0]
if self.index.names[0] is not None
else "index"
if "index" not in self.columns
else "level_{}".format(0)
]
# We will also use the `new_column_names` in the calculation of the internal metadata, so this is a
# lightweight way of ensuring the metadata matches.
if self.columns.nlevels > 1:
# Column labels are different for multilevel index.
new_column_names = pandas.MultiIndex.from_tuples(
# Set level names on the 1st columns level and fill up empty level names with empty string.
# Expand tuples in level names. This is how reset_index works when col_level col_fill are not specified.
[
tuple(
list(level) + [""] * (self.columns.nlevels - len(level))
if isinstance(level, tuple)
else [level] + [""] * (self.columns.nlevels - 1)
)
for level in level_names
],
names=self.columns.names,
)
else:
new_column_names = pandas.Index(level_names, tupleize_cols=False)
new_columns = new_column_names.append(self.columns)
def from_labels_executor(df, **kwargs):
# Setting the names here ensures that external and internal metadata always match.
df.index.names = new_column_names
# Handling of a case when columns have the same name as one of index levels names.
# In this case `df.reset_index` provides errors related to columns duplication.
# This case is possible because columns metadata updating is deferred. To workaround
# `df.reset_index` error we allow columns duplication in "if" branch via `concat`.
if any(name_level in df.columns for name_level in df.index.names):
columns_to_add = df.index.to_frame()
columns_to_add.reset_index(drop=True, inplace=True)
df = df.reset_index(drop=True)
result = pandas.concat([columns_to_add, df], axis=1, copy=False)
else:
result = df.reset_index()
# Put the index back to the original due to GH#4394
result.index = df.index
return result
new_parts = self._partition_mgr_cls.apply_func_to_select_indices(
0,
self._partitions,
from_labels_executor,
[0],
keep_remaining=True,
)
new_column_widths = [
self.index.nlevels + self._column_widths[0]
] + self._column_widths[1:]
result = self.__constructor__(
new_parts,
new_row_labels,
new_columns,
row_lengths=self._row_lengths_cache,
column_widths=new_column_widths,
)
# Set flag for propagating deferred row labels across dataframe partitions
result.synchronize_labels(axis=0)
return result
def to_labels(self, column_list: List[Hashable]) -> "PandasDataframe":
"""
Move one or more columns into the row labels. Previous labels are dropped.
Parameters
----------
column_list : list of hashable
The list of column names to place as the new row labels.
Returns
-------
PandasDataframe
A new PandasDataframe that has the updated labels.
"""
extracted_columns = self.mask(col_labels=column_list).to_pandas()
if len(column_list) == 1:
new_labels = pandas.Index(extracted_columns.squeeze(axis=1))
else:
new_labels = pandas.MultiIndex.from_frame(extracted_columns)
result = self.mask(col_labels=[i for i in self.columns if i not in column_list])
result.index = new_labels
return result
@lazy_metadata_decorator(apply_axis="both")
def _reorder_labels(self, row_positions=None, col_positions=None):
"""
Reorder the column and or rows in this DataFrame.
Parameters
----------
row_positions : list of int, optional
The ordered list of new row orders such that each position within the list
indicates the new position.
col_positions : list of int, optional
The ordered list of new column orders such that each position within the
list indicates the new position.
Returns
-------
PandasDataframe
A new PandasDataframe with reordered columns and/or rows.
"""
if row_positions is not None:
ordered_rows = self._partition_mgr_cls.map_axis_partitions(
0, self._partitions, lambda df: df.iloc[row_positions]
)
row_idx = self.index[row_positions]
else:
ordered_rows = self._partitions
row_idx = self.index
if col_positions is not None:
ordered_cols = self._partition_mgr_cls.map_axis_partitions(
1, ordered_rows, lambda df: df.iloc[:, col_positions]
)
col_idx = self.columns[col_positions]
else:
ordered_cols = ordered_rows
col_idx = self.columns
return self.__constructor__(ordered_cols, row_idx, col_idx)
@lazy_metadata_decorator(apply_axis=None)
def copy(self):
"""
Copy this object.
Returns
-------
PandasDataframe
A copied version of this object.
"""
return self.__constructor__(
self._partitions,
self.index.copy(),
self.columns.copy(),
self._row_lengths,
self._column_widths,
self._dtypes,
)
@classmethod
def combine_dtypes(cls, list_of_dtypes, column_names):
"""
Describe how data types should be combined when they do not match.
Parameters
----------
list_of_dtypes : list
A list of pandas Series with the data types.
column_names : list
The names of the columns that the data types map to.
Returns
-------
pandas.Series
A pandas Series containing the finalized data types.
"""
# Compute dtypes by getting collecting and combining all of the partitions. The
# reported dtypes from differing rows can be different based on the inference in
# the limited data seen by each worker. We use pandas to compute the exact dtype
# over the whole column for each column.
dtypes = (
pandas.concat(list_of_dtypes, axis=1)
.apply(lambda row: find_common_type(row.values), axis=1)
.squeeze(axis=0)
)
dtypes.index = column_names
return dtypes
@lazy_metadata_decorator(apply_axis="both")
def astype(self, col_dtypes):
"""
Convert the columns dtypes to given dtypes.
Parameters
----------
col_dtypes : dictionary of {col: dtype,...}