forked from pandas-dev/pandas
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathstata.py
2987 lines (2559 loc) · 106 KB
/
stata.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
"""
Module contains tools for processing Stata files into DataFrames
The StataReader below was originally written by Joe Presbrey as part of PyDTA.
It has been extended and improved by Skipper Seabold from the Statsmodels
project who also developed the StataWriter and was finally added to pandas in
a once again improved version.
You can find more information on http://presbrey.mit.edu/PyDTA and
http://www.statsmodels.org/devel/
"""
from collections import OrderedDict
import datetime
from io import BytesIO
import os
import struct
import sys
import warnings
from dateutil.relativedelta import relativedelta
import numpy as np
from pandas._libs.lib import infer_dtype
from pandas._libs.writers import max_len_string_array
from pandas.util._decorators import Appender, deprecate_kwarg
from pandas.core.dtypes.common import (
ensure_object, is_categorical_dtype, is_datetime64_dtype)
from pandas import (
Categorical, DatetimeIndex, NaT, Timestamp, concat, isna, to_datetime,
to_timedelta)
from pandas.core.base import StringMixin
from pandas.core.frame import DataFrame
from pandas.core.series import Series
from pandas.io.common import (
BaseIterator, _stringify_path, get_filepath_or_buffer)
_version_error = ("Version of given Stata file is not 104, 105, 108, "
"111 (Stata 7SE), 113 (Stata 8/9), 114 (Stata 10/11), "
"115 (Stata 12), 117 (Stata 13), or 118 (Stata 14)")
_statafile_processing_params1 = """\
convert_dates : boolean, defaults to True
Convert date variables to DataFrame time values.
convert_categoricals : boolean, defaults to True
Read value labels and convert columns to Categorical/Factor variables."""
_encoding_params = """\
encoding : string, None or encoding
Encoding used to parse the files. None defaults to latin-1."""
_statafile_processing_params2 = """\
index_col : string, optional, default: None
Column to set as index.
convert_missing : boolean, defaults to False
Flag indicating whether to convert missing values to their Stata
representations. If False, missing values are replaced with nan.
If True, columns containing missing values are returned with
object data types and missing values are represented by
StataMissingValue objects.
preserve_dtypes : boolean, defaults to True
Preserve Stata datatypes. If False, numeric data are upcast to pandas
default types for foreign data (float64 or int64).
columns : list or None
Columns to retain. Columns will be returned in the given order. None
returns all columns.
order_categoricals : boolean, defaults to True
Flag indicating whether converted categorical data are ordered."""
_chunksize_params = """\
chunksize : int, default None
Return StataReader object for iterations, returns chunks with
given number of lines."""
_iterator_params = """\
iterator : boolean, default False
Return StataReader object."""
_read_stata_doc = """
Read Stata file into DataFrame.
Parameters
----------
filepath_or_buffer : string or file-like object
Path to .dta file or object implementing a binary read() functions.
%s
%s
%s
%s
%s
Returns
-------
DataFrame or StataReader
See Also
--------
io.stata.StataReader : Low-level reader for Stata data files.
DataFrame.to_stata: Export Stata data files.
Examples
--------
Read a Stata dta file:
>>> df = pd.read_stata('filename.dta')
Read a Stata dta file in 10,000 line chunks:
>>> itr = pd.read_stata('filename.dta', chunksize=10000)
>>> for chunk in itr:
... do_something(chunk)
""" % (_statafile_processing_params1, _encoding_params,
_statafile_processing_params2, _chunksize_params,
_iterator_params)
_data_method_doc = """\
Read observations from Stata file, converting them into a dataframe
.. deprecated::
This is a legacy method. Use `read` in new code.
Parameters
----------
%s
%s
Returns
-------
DataFrame
""" % (_statafile_processing_params1, _statafile_processing_params2)
_read_method_doc = """\
Reads observations from Stata file, converting them into a dataframe
Parameters
----------
nrows : int
Number of lines to read from data file, if None read whole file.
%s
%s
Returns
-------
DataFrame
""" % (_statafile_processing_params1, _statafile_processing_params2)
_stata_reader_doc = """\
Class for reading Stata dta files.
Parameters
----------
path_or_buf : path (string), buffer or path object
string, path object (pathlib.Path or py._path.local.LocalPath) or object
implementing a binary read() functions.
.. versionadded:: 0.23.0 support for pathlib, py.path.
%s
%s
%s
%s
""" % (_statafile_processing_params1, _statafile_processing_params2,
_encoding_params, _chunksize_params)
@Appender(_read_stata_doc)
@deprecate_kwarg(old_arg_name='encoding', new_arg_name=None)
@deprecate_kwarg(old_arg_name='index', new_arg_name='index_col')
def read_stata(filepath_or_buffer, convert_dates=True,
convert_categoricals=True, encoding=None, index_col=None,
convert_missing=False, preserve_dtypes=True, columns=None,
order_categoricals=True, chunksize=None, iterator=False):
reader = StataReader(filepath_or_buffer,
convert_dates=convert_dates,
convert_categoricals=convert_categoricals,
index_col=index_col, convert_missing=convert_missing,
preserve_dtypes=preserve_dtypes,
columns=columns,
order_categoricals=order_categoricals,
chunksize=chunksize)
if iterator or chunksize:
data = reader
else:
try:
data = reader.read()
finally:
reader.close()
return data
_date_formats = ["%tc", "%tC", "%td", "%d", "%tw", "%tm", "%tq", "%th", "%ty"]
stata_epoch = datetime.datetime(1960, 1, 1)
def _stata_elapsed_date_to_datetime_vec(dates, fmt):
"""
Convert from SIF to datetime. http://www.stata.com/help.cgi?datetime
Parameters
----------
dates : Series
The Stata Internal Format date to convert to datetime according to fmt
fmt : str
The format to convert to. Can be, tc, td, tw, tm, tq, th, ty
Returns
Returns
-------
converted : Series
The converted dates
Examples
--------
>>> dates = pd.Series([52])
>>> _stata_elapsed_date_to_datetime_vec(dates , "%tw")
0 1961-01-01
dtype: datetime64[ns]
Notes
-----
datetime/c - tc
milliseconds since 01jan1960 00:00:00.000, assuming 86,400 s/day
datetime/C - tC - NOT IMPLEMENTED
milliseconds since 01jan1960 00:00:00.000, adjusted for leap seconds
date - td
days since 01jan1960 (01jan1960 = 0)
weekly date - tw
weeks since 1960w1
This assumes 52 weeks in a year, then adds 7 * remainder of the weeks.
The datetime value is the start of the week in terms of days in the
year, not ISO calendar weeks.
monthly date - tm
months since 1960m1
quarterly date - tq
quarters since 1960q1
half-yearly date - th
half-years since 1960h1 yearly
date - ty
years since 0000
If you don't have pandas with datetime support, then you can't do
milliseconds accurately.
"""
MIN_YEAR, MAX_YEAR = Timestamp.min.year, Timestamp.max.year
MAX_DAY_DELTA = (Timestamp.max - datetime.datetime(1960, 1, 1)).days
MIN_DAY_DELTA = (Timestamp.min - datetime.datetime(1960, 1, 1)).days
MIN_MS_DELTA = MIN_DAY_DELTA * 24 * 3600 * 1000
MAX_MS_DELTA = MAX_DAY_DELTA * 24 * 3600 * 1000
def convert_year_month_safe(year, month):
"""
Convert year and month to datetimes, using pandas vectorized versions
when the date range falls within the range supported by pandas.
Otherwise it falls back to a slower but more robust method
using datetime.
"""
if year.max() < MAX_YEAR and year.min() > MIN_YEAR:
return to_datetime(100 * year + month, format='%Y%m')
else:
index = getattr(year, 'index', None)
return Series(
[datetime.datetime(y, m, 1) for y, m in zip(year, month)],
index=index)
def convert_year_days_safe(year, days):
"""
Converts year (e.g. 1999) and days since the start of the year to a
datetime or datetime64 Series
"""
if year.max() < (MAX_YEAR - 1) and year.min() > MIN_YEAR:
return (to_datetime(year, format='%Y') +
to_timedelta(days, unit='d'))
else:
index = getattr(year, 'index', None)
value = [datetime.datetime(y, 1, 1) + relativedelta(days=int(d))
for y, d in zip(year, days)]
return Series(value, index=index)
def convert_delta_safe(base, deltas, unit):
"""
Convert base dates and deltas to datetimes, using pandas vectorized
versions if the deltas satisfy restrictions required to be expressed
as dates in pandas.
"""
index = getattr(deltas, 'index', None)
if unit == 'd':
if deltas.max() > MAX_DAY_DELTA or deltas.min() < MIN_DAY_DELTA:
values = [base + relativedelta(days=int(d)) for d in deltas]
return Series(values, index=index)
elif unit == 'ms':
if deltas.max() > MAX_MS_DELTA or deltas.min() < MIN_MS_DELTA:
values = [base + relativedelta(microseconds=(int(d) * 1000))
for d in deltas]
return Series(values, index=index)
else:
raise ValueError('format not understood')
base = to_datetime(base)
deltas = to_timedelta(deltas, unit=unit)
return base + deltas
# TODO: If/when pandas supports more than datetime64[ns], this should be
# improved to use correct range, e.g. datetime[Y] for yearly
bad_locs = np.isnan(dates)
has_bad_values = False
if bad_locs.any():
has_bad_values = True
data_col = Series(dates)
data_col[bad_locs] = 1.0 # Replace with NaT
dates = dates.astype(np.int64)
if fmt.startswith(("%tc", "tc")): # Delta ms relative to base
base = stata_epoch
ms = dates
conv_dates = convert_delta_safe(base, ms, 'ms')
elif fmt.startswith(("%tC", "tC")):
warnings.warn("Encountered %tC format. Leaving in Stata "
"Internal Format.")
conv_dates = Series(dates, dtype=np.object)
if has_bad_values:
conv_dates[bad_locs] = NaT
return conv_dates
# Delta days relative to base
elif fmt.startswith(("%td", "td", "%d", "d")):
base = stata_epoch
days = dates
conv_dates = convert_delta_safe(base, days, 'd')
# does not count leap days - 7 days is a week.
# 52nd week may have more than 7 days
elif fmt.startswith(("%tw", "tw")):
year = stata_epoch.year + dates // 52
days = (dates % 52) * 7
conv_dates = convert_year_days_safe(year, days)
elif fmt.startswith(("%tm", "tm")): # Delta months relative to base
year = stata_epoch.year + dates // 12
month = (dates % 12) + 1
conv_dates = convert_year_month_safe(year, month)
elif fmt.startswith(("%tq", "tq")): # Delta quarters relative to base
year = stata_epoch.year + dates // 4
month = (dates % 4) * 3 + 1
conv_dates = convert_year_month_safe(year, month)
elif fmt.startswith(("%th", "th")): # Delta half-years relative to base
year = stata_epoch.year + dates // 2
month = (dates % 2) * 6 + 1
conv_dates = convert_year_month_safe(year, month)
elif fmt.startswith(("%ty", "ty")): # Years -- not delta
year = dates
month = np.ones_like(dates)
conv_dates = convert_year_month_safe(year, month)
else:
raise ValueError("Date fmt {fmt} not understood".format(fmt=fmt))
if has_bad_values: # Restore NaT for bad values
conv_dates[bad_locs] = NaT
return conv_dates
def _datetime_to_stata_elapsed_vec(dates, fmt):
"""
Convert from datetime to SIF. http://www.stata.com/help.cgi?datetime
Parameters
----------
dates : Series
Series or array containing datetime.datetime or datetime64[ns] to
convert to the Stata Internal Format given by fmt
fmt : str
The format to convert to. Can be, tc, td, tw, tm, tq, th, ty
"""
index = dates.index
NS_PER_DAY = 24 * 3600 * 1000 * 1000 * 1000
US_PER_DAY = NS_PER_DAY / 1000
def parse_dates_safe(dates, delta=False, year=False, days=False):
d = {}
if is_datetime64_dtype(dates.values):
if delta:
delta = dates - stata_epoch
d['delta'] = delta.values.astype(
np.int64) // 1000 # microseconds
if days or year:
dates = DatetimeIndex(dates)
d['year'], d['month'] = dates.year, dates.month
if days:
days = (dates.astype(np.int64) -
to_datetime(d['year'], format='%Y').astype(np.int64))
d['days'] = days // NS_PER_DAY
elif infer_dtype(dates, skipna=False) == 'datetime':
if delta:
delta = dates.values - stata_epoch
f = lambda x: \
US_PER_DAY * x.days + 1000000 * x.seconds + x.microseconds
v = np.vectorize(f)
d['delta'] = v(delta)
if year:
year_month = dates.apply(lambda x: 100 * x.year + x.month)
d['year'] = year_month.values // 100
d['month'] = (year_month.values - d['year'] * 100)
if days:
f = lambda x: (x - datetime.datetime(x.year, 1, 1)).days
v = np.vectorize(f)
d['days'] = v(dates)
else:
raise ValueError('Columns containing dates must contain either '
'datetime64, datetime.datetime or null values.')
return DataFrame(d, index=index)
bad_loc = isna(dates)
index = dates.index
if bad_loc.any():
dates = Series(dates)
if is_datetime64_dtype(dates):
dates[bad_loc] = to_datetime(stata_epoch)
else:
dates[bad_loc] = stata_epoch
if fmt in ["%tc", "tc"]:
d = parse_dates_safe(dates, delta=True)
conv_dates = d.delta / 1000
elif fmt in ["%tC", "tC"]:
warnings.warn("Stata Internal Format tC not supported.")
conv_dates = dates
elif fmt in ["%td", "td"]:
d = parse_dates_safe(dates, delta=True)
conv_dates = d.delta // US_PER_DAY
elif fmt in ["%tw", "tw"]:
d = parse_dates_safe(dates, year=True, days=True)
conv_dates = (52 * (d.year - stata_epoch.year) + d.days // 7)
elif fmt in ["%tm", "tm"]:
d = parse_dates_safe(dates, year=True)
conv_dates = (12 * (d.year - stata_epoch.year) + d.month - 1)
elif fmt in ["%tq", "tq"]:
d = parse_dates_safe(dates, year=True)
conv_dates = 4 * (d.year - stata_epoch.year) + (d.month - 1) // 3
elif fmt in ["%th", "th"]:
d = parse_dates_safe(dates, year=True)
conv_dates = (2 * (d.year - stata_epoch.year) +
(d.month > 6).astype(np.int))
elif fmt in ["%ty", "ty"]:
d = parse_dates_safe(dates, year=True)
conv_dates = d.year
else:
raise ValueError(
"Format {fmt} is not a known Stata date format".format(fmt=fmt))
conv_dates = Series(conv_dates, dtype=np.float64)
missing_value = struct.unpack('<d', b'\x00\x00\x00\x00\x00\x00\xe0\x7f')[0]
conv_dates[bad_loc] = missing_value
return Series(conv_dates, index=index)
excessive_string_length_error = """
Fixed width strings in Stata .dta files are limited to 244 (or fewer)
characters. Column '%s' does not satisfy this restriction. Use the
'version=117' parameter to write the newer (Stata 13 and later) format.
"""
class PossiblePrecisionLoss(Warning):
pass
precision_loss_doc = """
Column converted from %s to %s, and some data are outside of the lossless
conversion range. This may result in a loss of precision in the saved data.
"""
class ValueLabelTypeMismatch(Warning):
pass
value_label_mismatch_doc = """
Stata value labels (pandas categories) must be strings. Column {0} contains
non-string labels which will be converted to strings. Please check that the
Stata data file created has not lost information due to duplicate labels.
"""
class InvalidColumnName(Warning):
pass
invalid_name_doc = """
Not all pandas column names were valid Stata variable names.
The following replacements have been made:
{0}
If this is not what you expect, please make sure you have Stata-compliant
column names in your DataFrame (strings only, max 32 characters, only
alphanumerics and underscores, no Stata reserved words)
"""
def _cast_to_stata_types(data):
"""Checks the dtypes of the columns of a pandas DataFrame for
compatibility with the data types and ranges supported by Stata, and
converts if necessary.
Parameters
----------
data : DataFrame
The DataFrame to check and convert
Notes
-----
Numeric columns in Stata must be one of int8, int16, int32, float32 or
float64, with some additional value restrictions. int8 and int16 columns
are checked for violations of the value restrictions and upcast if needed.
int64 data is not usable in Stata, and so it is downcast to int32 whenever
the value are in the int32 range, and sidecast to float64 when larger than
this range. If the int64 values are outside of the range of those
perfectly representable as float64 values, a warning is raised.
bool columns are cast to int8. uint columns are converted to int of the
same size if there is no loss in precision, otherwise are upcast to a
larger type. uint64 is currently not supported since it is concerted to
object in a DataFrame.
"""
ws = ''
# original, if small, if large
conversion_data = ((np.bool, np.int8, np.int8),
(np.uint8, np.int8, np.int16),
(np.uint16, np.int16, np.int32),
(np.uint32, np.int32, np.int64))
float32_max = struct.unpack('<f', b'\xff\xff\xff\x7e')[0]
float64_max = struct.unpack('<d', b'\xff\xff\xff\xff\xff\xff\xdf\x7f')[0]
for col in data:
dtype = data[col].dtype
# Cast from unsupported types to supported types
for c_data in conversion_data:
if dtype == c_data[0]:
if data[col].max() <= np.iinfo(c_data[1]).max:
dtype = c_data[1]
else:
dtype = c_data[2]
if c_data[2] == np.float64: # Warn if necessary
if data[col].max() >= 2 ** 53:
ws = precision_loss_doc % ('uint64', 'float64')
data[col] = data[col].astype(dtype)
# Check values and upcast if necessary
if dtype == np.int8:
if data[col].max() > 100 or data[col].min() < -127:
data[col] = data[col].astype(np.int16)
elif dtype == np.int16:
if data[col].max() > 32740 or data[col].min() < -32767:
data[col] = data[col].astype(np.int32)
elif dtype == np.int64:
if (data[col].max() <= 2147483620 and
data[col].min() >= -2147483647):
data[col] = data[col].astype(np.int32)
else:
data[col] = data[col].astype(np.float64)
if data[col].max() >= 2 ** 53 or data[col].min() <= -2 ** 53:
ws = precision_loss_doc % ('int64', 'float64')
elif dtype in (np.float32, np.float64):
value = data[col].max()
if np.isinf(value):
raise ValueError('Column {col} has a maximum value of '
'infinity which is outside the range '
'supported by Stata.'.format(col=col))
if dtype == np.float32 and value > float32_max:
data[col] = data[col].astype(np.float64)
elif dtype == np.float64:
if value > float64_max:
raise ValueError('Column {col} has a maximum value '
'({val}) outside the range supported by '
'Stata ({float64_max})'
.format(col=col, val=value,
float64_max=float64_max))
if ws:
warnings.warn(ws, PossiblePrecisionLoss)
return data
class StataValueLabel:
"""
Parse a categorical column and prepare formatted output
Parameters
----------
value : int8, int16, int32, float32 or float64
The Stata missing value code
Attributes
----------
string : string
String representation of the Stata missing value
value : int8, int16, int32, float32 or float64
The original encoded missing value
Methods
-------
generate_value_label
"""
def __init__(self, catarray):
self.labname = catarray.name
categories = catarray.cat.categories
self.value_labels = list(zip(np.arange(len(categories)), categories))
self.value_labels.sort(key=lambda x: x[0])
self.text_len = np.int32(0)
self.off = []
self.val = []
self.txt = []
self.n = 0
# Compute lengths and setup lists of offsets and labels
for vl in self.value_labels:
category = vl[1]
if not isinstance(category, str):
category = str(category)
warnings.warn(value_label_mismatch_doc.format(catarray.name),
ValueLabelTypeMismatch)
self.off.append(self.text_len)
self.text_len += len(category) + 1 # +1 for the padding
self.val.append(vl[0])
self.txt.append(category)
self.n += 1
if self.text_len > 32000:
raise ValueError('Stata value labels for a single variable must '
'have a combined length less than 32,000 '
'characters.')
# Ensure int32
self.off = np.array(self.off, dtype=np.int32)
self.val = np.array(self.val, dtype=np.int32)
# Total length
self.len = 4 + 4 + 4 * self.n + 4 * self.n + self.text_len
def _encode(self, s):
"""
Python 3 compatibility shim
"""
return s.encode(self._encoding)
def generate_value_label(self, byteorder, encoding):
"""
Parameters
----------
byteorder : str
Byte order of the output
encoding : str
File encoding
Returns
-------
value_label : bytes
Bytes containing the formatted value label
"""
self._encoding = encoding
bio = BytesIO()
null_string = '\x00'
null_byte = b'\x00'
# len
bio.write(struct.pack(byteorder + 'i', self.len))
# labname
labname = self._encode(_pad_bytes(self.labname[:32], 33))
bio.write(labname)
# padding - 3 bytes
for i in range(3):
bio.write(struct.pack('c', null_byte))
# value_label_table
# n - int32
bio.write(struct.pack(byteorder + 'i', self.n))
# textlen - int32
bio.write(struct.pack(byteorder + 'i', self.text_len))
# off - int32 array (n elements)
for offset in self.off:
bio.write(struct.pack(byteorder + 'i', offset))
# val - int32 array (n elements)
for value in self.val:
bio.write(struct.pack(byteorder + 'i', value))
# txt - Text labels, null terminated
for text in self.txt:
bio.write(self._encode(text + null_string))
bio.seek(0)
return bio.read()
class StataMissingValue(StringMixin):
"""
An observation's missing value.
Parameters
----------
value : int8, int16, int32, float32 or float64
The Stata missing value code
Attributes
----------
string : string
String representation of the Stata missing value
value : int8, int16, int32, float32 or float64
The original encoded missing value
Notes
-----
More information: <http://www.stata.com/help.cgi?missing>
Integer missing values make the code '.', '.a', ..., '.z' to the ranges
101 ... 127 (for int8), 32741 ... 32767 (for int16) and 2147483621 ...
2147483647 (for int32). Missing values for floating point data types are
more complex but the pattern is simple to discern from the following table.
np.float32 missing values (float in Stata)
0000007f .
0008007f .a
0010007f .b
...
00c0007f .x
00c8007f .y
00d0007f .z
np.float64 missing values (double in Stata)
000000000000e07f .
000000000001e07f .a
000000000002e07f .b
...
000000000018e07f .x
000000000019e07f .y
00000000001ae07f .z
"""
# Construct a dictionary of missing values
MISSING_VALUES = {}
bases = (101, 32741, 2147483621)
for b in bases:
# Conversion to long to avoid hash issues on 32 bit platforms #8968
MISSING_VALUES[b] = '.'
for i in range(1, 27):
MISSING_VALUES[i + b] = '.' + chr(96 + i)
float32_base = b'\x00\x00\x00\x7f'
increment = struct.unpack('<i', b'\x00\x08\x00\x00')[0]
for i in range(27):
value = struct.unpack('<f', float32_base)[0]
MISSING_VALUES[value] = '.'
if i > 0:
MISSING_VALUES[value] += chr(96 + i)
int_value = struct.unpack('<i',
struct.pack('<f', value))[0] + increment
float32_base = struct.pack('<i', int_value)
float64_base = b'\x00\x00\x00\x00\x00\x00\xe0\x7f'
increment = struct.unpack('q', b'\x00\x00\x00\x00\x00\x01\x00\x00')[0]
for i in range(27):
value = struct.unpack('<d', float64_base)[0]
MISSING_VALUES[value] = '.'
if i > 0:
MISSING_VALUES[value] += chr(96 + i)
int_value = struct.unpack('q', struct.pack('<d', value))[0] + increment
float64_base = struct.pack('q', int_value)
BASE_MISSING_VALUES = {'int8': 101,
'int16': 32741,
'int32': 2147483621,
'float32': struct.unpack('<f', float32_base)[0],
'float64': struct.unpack('<d', float64_base)[0]}
def __init__(self, value):
self._value = value
# Conversion to int to avoid hash issues on 32 bit platforms #8968
value = int(value) if value < 2147483648 else float(value)
self._str = self.MISSING_VALUES[value]
string = property(lambda self: self._str,
doc="The Stata representation of the missing value: "
"'.', '.a'..'.z'")
value = property(lambda self: self._value,
doc='The binary representation of the missing value.')
def __str__(self):
return self.string
def __repr__(self):
# not perfect :-/
return "{cls}({obj})".format(cls=self.__class__, obj=self)
def __eq__(self, other):
return (isinstance(other, self.__class__) and
self.string == other.string and self.value == other.value)
@classmethod
def get_base_missing_value(cls, dtype):
if dtype == np.int8:
value = cls.BASE_MISSING_VALUES['int8']
elif dtype == np.int16:
value = cls.BASE_MISSING_VALUES['int16']
elif dtype == np.int32:
value = cls.BASE_MISSING_VALUES['int32']
elif dtype == np.float32:
value = cls.BASE_MISSING_VALUES['float32']
elif dtype == np.float64:
value = cls.BASE_MISSING_VALUES['float64']
else:
raise ValueError('Unsupported dtype')
return value
class StataParser:
def __init__(self):
# type code.
# --------------------
# str1 1 = 0x01
# str2 2 = 0x02
# ...
# str244 244 = 0xf4
# byte 251 = 0xfb (sic)
# int 252 = 0xfc
# long 253 = 0xfd
# float 254 = 0xfe
# double 255 = 0xff
# --------------------
# NOTE: the byte type seems to be reserved for categorical variables
# with a label, but the underlying variable is -127 to 100
# we're going to drop the label and cast to int
self.DTYPE_MAP = dict(
list(zip(range(1, 245), ['a' + str(i) for i in range(1, 245)])) +
[
(251, np.int8),
(252, np.int16),
(253, np.int32),
(254, np.float32),
(255, np.float64)
]
)
self.DTYPE_MAP_XML = \
dict(
[
(32768, np.uint8), # Keys to GSO
(65526, np.float64),
(65527, np.float32),
(65528, np.int32),
(65529, np.int16),
(65530, np.int8)
]
)
self.TYPE_MAP = list(range(251)) + list('bhlfd')
self.TYPE_MAP_XML = \
dict(
[
# Not really a Q, unclear how to handle byteswap
(32768, 'Q'),
(65526, 'd'),
(65527, 'f'),
(65528, 'l'),
(65529, 'h'),
(65530, 'b')
]
)
# NOTE: technically, some of these are wrong. there are more numbers
# that can be represented. it's the 27 ABOVE and BELOW the max listed
# numeric data type in [U] 12.2.2 of the 11.2 manual
float32_min = b'\xff\xff\xff\xfe'
float32_max = b'\xff\xff\xff\x7e'
float64_min = b'\xff\xff\xff\xff\xff\xff\xef\xff'
float64_max = b'\xff\xff\xff\xff\xff\xff\xdf\x7f'
self.VALID_RANGE = {
'b': (-127, 100),
'h': (-32767, 32740),
'l': (-2147483647, 2147483620),
'f': (np.float32(struct.unpack('<f', float32_min)[0]),
np.float32(struct.unpack('<f', float32_max)[0])),
'd': (np.float64(struct.unpack('<d', float64_min)[0]),
np.float64(struct.unpack('<d', float64_max)[0]))
}
self.OLD_TYPE_MAPPING = {
98: 251, # byte
105: 252, # int
108: 253, # long
102: 254 # float
# don't know old code for double
}
# These missing values are the generic '.' in Stata, and are used
# to replace nans
self.MISSING_VALUES = {
'b': 101,
'h': 32741,
'l': 2147483621,
'f': np.float32(struct.unpack('<f', b'\x00\x00\x00\x7f')[0]),
'd': np.float64(
struct.unpack('<d', b'\x00\x00\x00\x00\x00\x00\xe0\x7f')[0])
}
self.NUMPY_TYPE_MAP = {
'b': 'i1',
'h': 'i2',
'l': 'i4',
'f': 'f4',
'd': 'f8',
'Q': 'u8'
}
# Reserved words cannot be used as variable names
self.RESERVED_WORDS = ('aggregate', 'array', 'boolean', 'break',
'byte', 'case', 'catch', 'class', 'colvector',
'complex', 'const', 'continue', 'default',
'delegate', 'delete', 'do', 'double', 'else',
'eltypedef', 'end', 'enum', 'explicit',
'export', 'external', 'float', 'for', 'friend',
'function', 'global', 'goto', 'if', 'inline',
'int', 'local', 'long', 'NULL', 'pragma',
'protected', 'quad', 'rowvector', 'short',
'typedef', 'typename', 'virtual')
class StataReader(StataParser, BaseIterator):
__doc__ = _stata_reader_doc
@deprecate_kwarg(old_arg_name='encoding', new_arg_name=None)
@deprecate_kwarg(old_arg_name='index', new_arg_name='index_col')
def __init__(self, path_or_buf, convert_dates=True,
convert_categoricals=True, index_col=None,
convert_missing=False, preserve_dtypes=True,
columns=None, order_categoricals=True,
encoding=None, chunksize=None):
super().__init__()
self.col_sizes = ()
# Arguments to the reader (can be temporarily overridden in
# calls to read).
self._convert_dates = convert_dates
self._convert_categoricals = convert_categoricals
self._index_col = index_col
self._convert_missing = convert_missing
self._preserve_dtypes = preserve_dtypes
self._columns = columns
self._order_categoricals = order_categoricals
self._encoding = None
self._chunksize = chunksize
# State variables for the file
self._has_string_data = False
self._missing_values = False
self._can_read_value_labels = False
self._column_selector_set = False
self._value_labels_read = False
self._data_read = False
self._dtype = None
self._lines_read = 0
self._native_byteorder = _set_endianness(sys.byteorder)
path_or_buf = _stringify_path(path_or_buf)
if isinstance(path_or_buf, str):
path_or_buf, encoding, _, should_close = get_filepath_or_buffer(
path_or_buf)
if isinstance(path_or_buf, (str, bytes)):
self.path_or_buf = open(path_or_buf, 'rb')
else:
# Copy to BytesIO, and ensure no encoding
contents = path_or_buf.read()
self.path_or_buf = BytesIO(contents)
self._read_header()
self._setup_dtype()
def __enter__(self):
""" enter context manager """
return self