/
lag.py
613 lines (515 loc) · 23.6 KB
/
lag.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
"""Lagging transformer."""
# copyright: sktime developers, BSD-3-Clause License (see LICENSE file)
__author__ = ["fkiraly"]
import numpy as np
import pandas as pd
from sktime.transformations.base import BaseTransformer
from sktime.utils.multiindex import flatten_multiindex
from sktime.utils.warnings import warn
# this function is needed since pandas DataFrame.shift
# seems to have problems with numpy int inside
def _coerce_to_int(obj):
"""Coerces numpy int or list of numpy int to python int."""
if isinstance(obj, np.integer):
return int(obj)
if isinstance(obj, list):
return [_coerce_to_int(x) for x in obj]
return obj
class Lag(BaseTransformer):
"""Lagging transformer. Lags time series by one or multiple lags.
Transforms a time series into a lagged version of itself.
Multiple lags can be provided, as a list.
Estimator-like wrapper of pandas.shift and integer index lagging.
Lags can be provided as a simple offset, ``lags``, or pair of (lag count,
frequency),
with lag count an int (``lags`` arg) and frequency a ``pandas`` frequency
descriptor.
When multiple lags are provided, multiple column concatenated copies of the lagged
time series will be created.
Names of columns are lagname__variablename, where lagname describes the lag/freq.
If data was provided in _fit or _update, Lag transformer memorizes those indices
and uses them for computing lagged values.
To use only data seen in transform, use the FitInTransform compositor.
Parameters
----------
lags : lag offset, or list of lag offsets, optional, default=0 (identity transform)
a "lag offset" can be one of the following:
int - number of periods to shift/lag
time-like: ``DateOffset``, ``tseries.offsets``, or ``timedelta``
time delta offset to shift/lag
requires time index of transformed data to be time-like (not int)
str - time rule from pandas.tseries module, e.g., "EOM"
freq : frequency descriptor of list of frequency descriptors, optional, default=None
if passed, must be scalar, or list of equal length to ``lags`` parameter
elements in ``freq`` correspond to elements in lags
if i-th element of ``freq`` is not None, i-th element of ``lags`` must be int
this is called the "corresponding lags element" below
"frequency descriptor" can be one of the following:
time-like: ``DateOffset``, ``tseries.offsets``, or ``timedelta``
multiplied to corresponding ``lags`` element when shifting
str - offset from pd.tseries module, e.g., "D", "M", or time rule, e.g., "EOM"
index_out : str, optional, one of "shift", "original", "extend", default="extend"
determines set of output indices in lagged time series
"shift" - only shifted indices are retained.
Will not create NA for single lag, but can create NA for multiple lags.
"original" - only original indices are retained. Will usually create NA.
"extend" - both original indices and shifted indices are retained.
Will usually create NA, possibly many, if shifted/original do not intersect.
flatten_transform_index : bool, optional (default=True)
if True, columns of return DataFrame are flat, by "lagname__variablename"
if False, columns are MultiIndex (lagname, variablename)
has no effect if return mtype is one without column names
keep_column_names : bool, optional (default=False)
has an effect only if ``lags`` contains only a single element
if True, ensures that column names of ``transform`` output are same as in input,
i.e., not ``lag_x__varname`` but ``varname``. Overrides
``flatten_transform_index``.
remember_data : bool, optional (default=True)
if True, memorizes data seen in ``fit``, ``update``, uses it in ``transform``
if False, only uses data seen in ``transform`` to produce lags
setting to False ensures faster runtime if only used via ``fit_transform``
Examples
--------
>>> from sktime.datasets import load_airline
>>> from sktime.transformations.series.lag import Lag
>>> X = load_airline()
Single lag will yield a time series with the same variables:
>>> t = Lag(2)
>>> Xt = t.fit_transform(X)
Multiple lags can be provided, this will result in multiple columns:
>>> t = Lag([2, 4, -1])
>>> Xt = t.fit_transform(X)
The default setting of index_out will extend indices either side.
To ensure that the index remains the same after transform,
use index_out="original"
>>> t = Lag([2, 4, -1], index_out="original")
>>> Xt = t.fit_transform(X)
The lag transformer may (and usually will) create NAs.
(except when index_out="shift" and there is only a single lag, or in
trivial cases). This may need to be handled, e.g., if a subsequent
pipeline step does not accept NA. To deal with the NAs,
pipeline with the Imputer:
>>> from sktime.datasets import load_airline
>>> from sktime.transformations.series.impute import Imputer
>>> from sktime.transformations.series.lag import Lag
>>> X = load_airline()
>>>
>>> t = Lag([2, 4, -1]) * Imputer("nearest")
>>> Xt = t.fit_transform(X)
"""
_tags = {
"authors": ["fkiraly"],
"scitype:transform-input": "Series",
# what is the scitype of X: Series, or Panel
"scitype:transform-output": "Series",
# what scitype is returned: Primitives, Series, Panel
"scitype:instancewise": True, # is this an instance-wise transform?
"capability:inverse_transform": False, # can the transformer inverse transform?
"univariate-only": False, # can the transformer handle multivariate X?
"X_inner_mtype": "pd.DataFrame", # which mtypes do _fit/_predict support for X?
"y_inner_mtype": "None", # which mtypes do _fit/_predict support for y?
"fit_is_empty": False, # is fit empty and can be skipped? Yes = True
"transform-returns-same-time-index": False,
# does transform return have the same time index as input X
"skip-inverse-transform": True, # is inverse-transform skipped when called?
"capability:unequal_length": True,
"capability:unequal_length:removes": False,
"handles-missing-data": True, # can estimator handle missing data?
"capability:missing_values:removes": False,
"remember_data": True, # remember all data seen as _X
}
# todo: add any hyper-parameters and components to constructor
def __init__(
self,
lags=0,
freq=None,
index_out="extend",
flatten_transform_index=True,
keep_column_names=False,
remember_data=True,
):
self.lags = lags
self.freq = freq
self.index_out = index_out
self.flatten_transform_index = flatten_transform_index
self.keep_column_names = keep_column_names
self.remember_data = remember_data
if index_out not in ["shift", "extend", "original"]:
raise ValueError(
'index_out must be one of the strings "shift", "extend", "original"'
f'but found "{index_out}"'
)
# _lags and _freq are list-coerced variants of lags, freq
if not isinstance(lags, list):
self._lags = [lags]
else:
self._lags = lags
if not isinstance(freq, list):
# if freq is a single value, expand it to length of lags
self._freq = [freq] * len(self._lags)
else:
self._freq = freq
msg = "freq must be a list of equal length to lags, or a scalar."
assert len(self._lags) == len(self._freq), msg
super().__init__()
if index_out == "original":
self.set_tags(**{"transform-returns-same-time-index": True})
if not remember_data:
self.set_tags(**{"remember_data": False, "fit_is_empty": True})
def _yield_shift_params(self):
"""Yield (periods, freq) pairs to pass to pandas.DataFrame.shift."""
# we need to coerce lags, or shift will break with numpy
coerced_lags = _coerce_to_int(self._lags)
for lag, freq in zip(coerced_lags, self._freq):
if not isinstance(lag, int):
yield 1, lag
elif lag is None:
yield 1, freq
else:
yield lag, freq
def _yield_shift_param_names(self):
"""Yield string representation of (periods, freq) pairs."""
for lag, freq in self._yield_shift_params():
if freq is None:
name = str(lag)
elif lag is None:
name = str(freq)
else:
name = f"{lag}{freq}"
name = "lag_" + name
yield name
def _transform(self, X, y=None):
"""Transform X and return a transformed version.
private _transform containing core logic, called from transform
Parameters
----------
X : pd.DataFrame
Data to be transformed
y : ignored, passed for interface compatibility
Returns
-------
pd.DataFrame, transformed version of X
"""
index_out = self.index_out
remember_data = self.remember_data
X_orig_idx = X.index
X_orig_cols = X.columns
if remember_data:
X = X.combine_first(self._X).copy()
shift_params = list(self._yield_shift_params())
Xt_list = []
for lag, freq in shift_params:
# need to deal separately with RangeIndex
# because shift always cuts off the end values
if isinstance(lag, int) and pd.api.types.is_integer_dtype(X.index):
Xt = X.copy()
Xt.index = X.index + lag
X_orig_idx_shifted = X_orig_idx + lag
else:
if hasattr(X.index, "freq") and X.index.freq is None and freq is None:
freq = pd.infer_freq(X.index)
X_orig_idx_shifted = X_orig_idx.shift(periods=lag, freq=freq)
if isinstance(lag, int) and freq is None:
freq = "infer"
Xt = X.copy().shift(periods=lag, freq=freq)
# sub-set to original plus shifted, if "extend"
# this is necessary, because we added indices from _X above
if index_out == "extend":
X_orig_idx_extended = X_orig_idx_shifted.union(X_orig_idx)
Xt = Xt.reindex(X_orig_idx_extended)
# sub-set to original, if "original"
if index_out == "original":
Xt = Xt.reindex(X_orig_idx)
# sub-set to shifted index, if "shifted"
# this is necessary if we added indices from _X above
if index_out == "shift" and remember_data:
Xt = Xt.loc[X_orig_idx_shifted]
Xt_list.append(Xt)
lag_names = self._yield_shift_param_names()
Xt = pd.concat(Xt_list, axis=1, keys=lag_names, names=["lag", "variable"])
if self.flatten_transform_index:
Xt.columns = flatten_multiindex(Xt.columns)
if len(shift_params) == 1 and self.keep_column_names:
Xt.columns = X_orig_cols
# some pandas versions do not sort index automatically after concat
# so removing will break specific pandas versions
Xt = Xt.sort_index()
return Xt
# todo: consider implementing this, optional
# if not implementing, delete the _inverse_transform method
# inverse transform exists only if transform does not change scitype
# i.e., Series transformed to Series
def _inverse_transform(self, X, y=None):
"""Inverse transform, inverse operation to transform.
private _inverse_transform containing core logic, called from inverse_transform
Parameters
----------
X : pd.DataFrame
Data to be inverse transformed
y : ignored, passed for interface compatibility
Returns
-------
pd.DataFrame, inverse transformed version of X
"""
# implement here
# IMPORTANT: avoid side effects to X, y
#
# type conventions are exactly those in _transform, reversed
#
# for example: if transform-output is "Series":
# return should be of same mtype as input, X_inner_mtype
# if multiple X_inner_mtype are supported, ensure same input/output
#
# todo: add the return mtype/scitype to the docstring, e.g.,
# Returns
# -------
# X_inv_transformed : Series of mtype pd.DataFrame
# inverse transformed version of X
def _update(self, X, y=None):
"""Update transformer with X and y.
private _update containing the core logic, called from update
Parameters
----------
X : pd.DataFrame
Data to update transformer with
y : ignored, passed for interface compatibility
Returns
-------
self: reference to self
"""
return self
# todo: return default parameters, so that a test instance can be created
# required for automated unit and integration testing of estimator
@classmethod
def get_test_params(cls, parameter_set="default"):
"""Return testing parameter settings for the estimator.
Parameters
----------
parameter_set : str, default="default"
Name of the set of test parameters to return, for use in tests. If no
special parameters are defined for a value, will return ``"default"`` set.
There are currently no reserved values for transformers.
Returns
-------
params : dict or list of dict, default = {}
Parameters to create testing instances of the class
Each dict are parameters to construct an "interesting" test instance, i.e.,
``MyClass(**params)`` or ``MyClass(**params[i])`` creates a valid test
instance.
``create_test_instance`` uses the first (or only) dictionary in ``params``
"""
params1 = {"lags": 2, "index_out": "original"}
params2 = {"lags": [-1, 4]}
params3 = {"lags": [0, 1, -1], "index_out": "shift"}
return [params1, params2, params3]
class ReducerTransform(BaseTransformer):
"""Transformer for forecasting reduction. Prepares tabular X/y via lag and trafos.
Parameters
----------
window_length : int, optional, default=0
window length used in the reduction algorithm
lags : lag offset, or list of lag offsets, optional, default=0 (identity transform)
a "lag offset" can be one of the following:
int - number of periods to shift/lag
time-like: ``DateOffset``, ``tseries.offsets``, or ``timedelta``
time delta offset to shift/lag
requires time index of transformed data to be time-like (not int)
str - time rule from pandas.tseries module, e.g., "EOM"
freq : frequency descriptor of list of frequency descriptors, optional, default=None
if passed, must be scalar, or list of equal length to ``lags`` parameter
elements in ``freq`` correspond to elements in lags
if i-th element of ``freq`` is not None, i-th element of ``lags`` must be int
this is called the "corresponding lags element" below
"frequency descriptor" can be one of the following:
time-like: ``DateOffset``, ``tseries.offsets``, or ``timedelta``
multiplied to corresponding ``lags`` element when shifting
str - offset from pd.tseries module, e.g., "D", "M", or time rule, e.g., "EOM"
shifted_vars : None
shifted_vars_lag : 0
shifted_vars_freq :
transformers : sktime series-to-series transformer, or list thereof
impute_method : str or None, optional, method string passed to Imputer
default="bfill", admissible strings are of Imputer.method parameter, see there
if None, no imputation is done when applying Lag transformer to obtain inner X
Examples
--------
>>> from sktime.datasets import load_airline
>>> from sktime.transformations.series.lag import Lag
>>> X = load_airline()
Single lag will yield a time series with the same variables:
>>> t = Lag(2)
>>> Xt = t.fit_transform(X)
Multiple lags can be provided, this will result in multiple columns:
>>> t = Lag([2, 4, -1])
>>> Xt = t.fit_transform(X)
The default setting of index_out will extend indices either side.
To ensure that the index remains the same after transform, use index_out="original"
>>> t = Lag([2, 4, -1], index_out="original")
>>> Xt = t.fit_transform(X)
The lag transformer may (and usually will) create NAs.
(except when index_out="shift" and there is only a single lag, or in trivial cases)
This may need to be handled, e.g., if a subsequent pipeline step does not accept NA.
To deal with the NAs, pipeline with the Imputer:
>>> from sktime.datasets import load_airline
>>> from sktime.transformations.series.impute import Imputer
>>> from sktime.transformations.series.lag import Lag
>>> X = load_airline()
>>>
>>> t = Lag([2, 4, -1]) * Imputer("nearest")
>>> Xt = t.fit_transform(X)
"""
_tags = {
"scitype:transform-input": "Series",
# what is the scitype of X: Series, or Panel
"scitype:transform-output": "Series",
# what scitype is returned: Primitives, Series, Panel
"scitype:instancewise": True, # is this an instance-wise transform?
"capability:inverse_transform": False, # can the transformer inverse transform?
"univariate-only": False, # can the transformer handle multivariate X?
"X_inner_mtype": "pd.DataFrame", # which mtypes do _fit/_predict support for X?
"y_inner_mtype": "pd.DataFrame", # which mtypes do _fit/_predict support for y?
"fit_is_empty": False, # is fit empty and can be skipped? Yes = True
"transform-returns-same-time-index": False,
# does transform return have the same time index as input X
"skip-inverse-transform": True, # is inverse-transform skipped when called?
"capability:unequal_length": True,
"capability:unequal_length:removes": False,
"handles-missing-data": True, # can estimator handle missing data?
"capability:missing_values:removes": False,
}
# todo: add any hyper-parameters and components to constructor
def __init__(
self,
lags=0,
freq=None,
shifted_vars=None,
shifted_vars_lag=0,
shifted_vars_freq=None,
transformers=None,
impute_method="bfill",
):
self.lags = lags
self.freq = freq
self.shifted_vars = shifted_vars
self.shifted_vars_lag = shifted_vars_lag
self.shifted_vars_freq = shifted_vars_freq
self.transformers = transformers
self.impute_method = impute_method
# _lags and _freq are list-coerced variants of lags, freq
if isinstance(lags, int):
self._lags = list(range(lags))
else:
self._lags = lags
super().__init__()
def _fit(self, X, y=None):
"""Fit transformer to X and y.
private _fit containing the core logic, called from fit
Parameters
----------
X : pd.DataFrame
Data to fit transform to
y : ignored, passed for interface compatibility
Returns
-------
self: reference to self
"""
from sktime.transformations.compose import FeatureUnion, YtoX
from sktime.transformations.series.impute import Imputer
impute_method = self.impute_method
lags = self._lags
freq = self.freq
# edge case of no reduction features - prepare to return all-zeros matrix
if len(lags) == 0 and y is None:
warn(
"no lags specified and no exogeneous data present, "
"empty reduction X. Returning all-zeros X.",
obj=self,
)
self.trafo_ = 0
return self
transformers = []
if len(lags) > 0:
t = Lag(lags=lags, freq=freq, index_out="original", keep_column_names=True)
transformers += [("Lag", t)]
if y is not None:
exog_t = YtoX()
if self.shifted_vars_lag != 0:
lag = self.shifted_vars_lag
freq = self.shifted_vars_freq
exog_t = exog_t * Lag(
lags=lag, freq=freq, index_out="original", keep_column_names=True
)
transformers += [("exog", exog_t)]
if self.transformers is not None:
transformers += self.transformers
t = FeatureUnion(transformers, flatten_transform_index=False)
if impute_method is not None:
t = t * Imputer(method=impute_method)
self.trafo_ = t.fit(X=X, y=y)
return self
def _transform(self, X, y=None):
"""Transform X and return a transformed version.
private _transform containing core logic, called from transform
Parameters
----------
X : pd.DataFrame
Data to be transformed
y : ignored, passed for interface compatibility
Returns
-------
pd.DataFrame, transformed version of X
"""
# treat edge case of empty reduction matrix: return all-zeros
if self.trafo_ == 0:
return pd.DataFrame(0, index=X.index, columns=["zero"])
Xt = self.trafo_.transform(X=X, y=y)
varnames = Xt.columns.get_level_values(1)
if varnames.is_unique:
Xt.columns = Xt.columns.droplevel(0)
Xt.columns.name = None
else:
duplicates = list(varnames[varnames.duplicated()])
warn(
f"duplicate variable names found in ReducerTransform: {duplicates}, "
"returning variables with transformer name prefix",
obj=self,
)
Xt.columns = flatten_multiindex(Xt.columns)
Xt = Xt.loc[Xt.index.intersection(X.index)]
return Xt
def _update(self, X, y=None):
"""Update transformer with X and y.
private _update containing the core logic, called from update
Parameters
----------
X : pd.DataFrame
Data to update transformer with
y : ignored, passed for interface compatibility
Returns
-------
self: reference to self
"""
self.trafo_.update(X=X, y=y)
return self
# todo: return default parameters, so that a test instance can be created
# required for automated unit and integration testing of estimator
@classmethod
def get_test_params(cls, parameter_set="default"):
"""Return testing parameter settings for the estimator.
Parameters
----------
parameter_set : str, default="default"
Name of the set of test parameters to return, for use in tests. If no
special parameters are defined for a value, will return ``"default"`` set.
There are currently no reserved values for transformers.
Returns
-------
params : dict or list of dict, default = {}
Parameters to create testing instances of the class
Each dict are parameters to construct an "interesting" test instance, i.e.,
``MyClass(**params)`` or ``MyClass(**params[i])`` creates a valid test
instance.
``create_test_instance`` uses the first (or only) dictionary in ``params``
"""
params1 = {"lags": 2}
return [params1]