-
Notifications
You must be signed in to change notification settings - Fork 89
/
_redcomets.py
617 lines (511 loc) · 20.8 KB
/
_redcomets.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
# copyright: aeon developers, BSD-3-Clause License (see LICENSE file)
"""Random EnhanceD Co-eye for Multivariate Time Series (RED CoMETS).
Ensemble of symbolically represented time series using random forests as the base
classifier.
"""
__author__ = ["zy18811"]
__all__ = ["REDCOMETS"]
from collections import Counter
import numpy as np
from joblib import Parallel, delayed
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import cross_val_score
from sklearn.neighbors import NearestNeighbors
from sklearn.utils import check_random_state
from aeon.classification.base import BaseClassifier
from aeon.transformations.collection.dictionary_based import SAX, SFA
from aeon.utils.validation._dependencies import _check_soft_dependencies
class REDCOMETS(BaseClassifier):
"""
Random EnhanceD Co-eye for Multivariate Time Series (RED CoMETS).
Ensemble of symbolically represented time series using random forests as the base
classifier as described in [1]_. Based on Co-eye [2]_.
Parameters
----------
variant : int, default=3
RED CoMETS variant to use from {1, 2, 3, 4, 5, 6, 7, 8, 9} to use as per [1]_.
Defaults to RED CoMETS-3. Variants 4-9 only support multivariate problems.
perc_length : int or float, default=5
Percentage of time series length used to determinne number of lenses during
pair selection.
n_trees : int, default=100
Number of trees used by each random forest sub-classifier.
random_state : int, RandomState instance or None, default=None
If ``int``, random_state is the seed used by the random number generator;
If ``RandomState`` instance, ``random_state`` is the random number generator;
If ``None``, the random number generator is the ``RandomState`` instance used
by ``np.random``.
n_jobs : int, default=1
The number of jobs to run in parallel for both `fit` and `predict`.
``-1`` means using all processors.
parallel_backend : str, ParallelBackendBase instance or None, default=None
Specify the parallelisation backend implementation in joblib for Catch22,
if ``None`` a 'prefer' value of "threads" is used by default.
Valid options are "loky", "multiprocessing", "threading" or a custom backend.
See the joblib Parallel documentation for more details.
Attributes
----------
n_classes_ : int
The number of classes.
classes_ : list
The unique class labels.
See Also
--------
SAX, SFA
Notes
-----
Adapted from the implementation at https://github.com/zy18811/RED-CoMETS
References
----------
.. [1] Luca A. Bennett and Zahraa S. Abdallah, "RED CoMETS: An ensemble classifier
for symbolically represented multivariate time series."
Preprint, https://arxiv.org/abs/2307.13679
.. [2] Zahraa S. Abdallah and Mohamed Medhat Gaber, "Co-eye: a multi-resolution
ensemble classifier for symbolically approximated time series."
Machine Learning (2020).
Examples
--------
>>> from aeon.classification.dictionary_based import REDCOMETS
>>> from aeon.datasets import load_unit_test
>>> X_train, y_train = load_unit_test(split="train", return_X_y=True)
>>> X_test, y_test = load_unit_test(split="test", return_X_y=True)
>>> clf = REDCOMETS() # doctest: +SKIP
>>> clf.fit(X_train, y_train) # doctest: +SKIP
REDCOMETS(...)
>>> y_pred = clf.predict(X_test) # doctest: +SKIP
"""
_tags = {
"python_dependencies": "imblearn",
"capability:multivariate": True,
"capability:multithreading": True,
"algorithm_type": "dictionary",
}
def __init__(
self,
variant=3,
perc_length=5,
n_trees=100,
random_state=None,
n_jobs=1,
parallel_backend=None,
):
assert variant in [1, 2, 3, 4, 5, 6, 7, 8, 9]
self.variant = variant
assert 0 < perc_length <= 100
self.perc_length = perc_length
self.n_trees = n_trees
self.random_state = random_state
self.n_jobs = n_jobs
self.parallel_backend = parallel_backend
self._n_channels = 1
self.sfa_clfs = []
self.sfa_transforms = []
self.sax_clfs = []
self.sax_transforms = []
super(REDCOMETS, self).__init__()
def _fit(self, X, y):
"""Build a REDCOMETS classifier from the training set (X, y).
Parameters
----------
X : 3D np.ndarray, shape = [n_instances, n_channels, n_timepoints]
The training data.
y : 1D np.ndarray, shape = [n_instances]
The class labels.
Returns
-------
self :
Reference to self.
"""
if (n_channels := X.shape[1]) == 1: # Univariate
assert self.variant in [1, 2, 3]
(
self.sfa_transforms,
self.sfa_clfs,
self.sax_transforms,
self.sax_clfs,
) = self._build_univariate_ensemble(np.squeeze(X), y)
else: # Multivariate
self._n_channels = n_channels
if self.variant in [1, 2, 3]: # Concatenate
X_concat = X.reshape(*X.shape[:-2], -1)
(
self.sfa_transforms,
self.sfa_clfs,
self.sax_transforms,
self.sax_clfs,
) = self._build_univariate_ensemble(X_concat, y)
elif self.variant in [4, 5, 6, 7, 8, 9]: # Ensemble
(
self.sfa_transforms,
self.sfa_clfs,
self.sax_transforms,
self.sax_clfs,
) = self._build_dimension_ensemble(X, y)
def _build_univariate_ensemble(self, X, y):
"""Build RED CoMETS ensemble from the univariate training set (X, y).
Parameters
----------
X : 2D np.ndarray, shape = [n_instances, n_timepoints]
The training data.
y : 1D np.ndarray, shape = [n_instances]
The class labels.
Returns
-------
sfa_transforms :
List of ``SFA()`` instances with random word length and alpabet size
sfa_clfs :
List of ``(RandomForestClassifier(), weight)`` tuples fitted on `SFA`
transformed training data
sax_transforms :
List of ``SAX()`` instances with random word length and alpabet size
sax_clfs :
List of ``(RandomForestClassifier(), weight)`` tuples fitted on `SAX`
transformed training data
"""
_check_soft_dependencies(
"imbalanced-learn",
package_import_alias={"imbalanced-learn": "imblearn"},
severity="error",
obj=self,
)
from imblearn.over_sampling import SMOTE, RandomOverSampler
if self.variant in [1, 2, 3]:
perc_length = self.perc_length / self._n_channels
else:
perc_length = self.perc_length
n_lenses = max(2 * int(perc_length * X.shape[1] // 100), 2)
min_neighbours = min(Counter(y).items(), key=lambda k: k[1])[1]
max_neighbours = max(Counter(y).items(), key=lambda k: k[1])[1]
if min_neighbours == max_neighbours:
X_smote = X
y_smote = y
else:
if min_neighbours > 5:
min_neighbours = 6
try:
X_smote, y_smote = SMOTE(
sampling_strategy="all",
k_neighbors=NearestNeighbors(
n_neighbors=min_neighbours - 1, n_jobs=self.n_jobs
),
random_state=self.random_state,
).fit_resample(X, y)
except ValueError:
X_smote, y_smote = RandomOverSampler(
sampling_strategy="all", random_state=self.random_state
).fit_resample(X, y)
lenses = self._get_random_lenses(X_smote, n_lenses)
sax_lenses = lenses[: n_lenses // 2]
sfa_lenses = lenses[n_lenses // 2 :]
cv = np.min([5, len(y_smote) // len(list(set(y_smote)))])
sfa_transforms = [
SFA(
word_length=w,
alphabet_size=a,
window_size=X_smote.shape[1],
binning_method="equi-width",
save_words=True,
n_jobs=self.n_jobs,
random_state=self.random_state,
)
for w, a in sfa_lenses
]
sfa_clfs = []
for sfa in sfa_transforms:
sfa.fit_transform(X_smote, y_smote)
X_sfa = np.array(
[sfa.word_list(word) for word in np.array(sfa.words).ravel()]
)
rf = RandomForestClassifier(
n_estimators=self.n_trees,
random_state=self.random_state,
n_jobs=self.n_jobs,
)
rf.fit(X_sfa, y_smote)
if self.variant == 1:
weight = 1
elif self.variant == 3:
weight = cross_val_score(
rf, X_sfa, y_smote, cv=cv, n_jobs=self.n_jobs
).mean()
else:
weight = None
sfa_clfs.append((rf, weight))
sax_transforms = [
SAX(n_segments=w, alphabet_size=a, znormalized=False) for w, a in sax_lenses
]
sax_clfs = []
for X_sax in self._parallel_sax(sax_transforms, X_smote):
rf = RandomForestClassifier(
n_estimators=self.n_trees,
random_state=self.random_state,
n_jobs=self.n_jobs,
)
rf.fit(X_sax, y_smote)
if self.variant == 1:
weight = 1
elif self.variant == 3:
weight = cross_val_score(
rf, X_sax, y_smote, cv=cv, n_jobs=self.n_jobs
).mean()
else:
weight = None
sax_clfs.append((rf, weight))
return sfa_transforms, sfa_clfs, sax_transforms, sax_clfs
def _build_dimension_ensemble(self, X, y):
"""Build an ensemble of univariate RED CoMETS ensembles over dimensions.
Parameters
----------
X : 3D np.ndarray, shape = [n_instances, n_channels, n_timepoints]
The training data.
``n_channels > 1``
y : 1D np.ndarray, shape = [n_instances]
The class labels.
Returns
-------
sfa_transforms : list
List of lists of ``SFA()`` instances with random word length and alpabet
size
sfa_clfs : list
List of lists of ``(RandomForestClassifier(), weight)`` tuples fitted on
`SFA` transformed training data
sax_transforms : list
List of lists of ``SAX()`` instances with random word length and alpabet
size
sax_clfs : list
List of lists ``(RandomForestClassifier(), weight)`` tuples fitted on `SAX`
transformed training data
"""
sfa_transforms = []
sfa_clfs = []
sax_transforms = []
sax_clfs = []
for d in range(self._n_channels):
X_d = X[:, d, :]
(
sfa_trans_d,
sfa_clfs_d,
sax_trans_d,
sax_clfs_d,
) = self._build_univariate_ensemble(X_d, y)
sfa_transforms.append(sfa_trans_d)
sfa_clfs.append(sfa_clfs_d)
sax_transforms.append(sax_trans_d)
sax_clfs.append(sax_clfs_d)
return sfa_transforms, sfa_clfs, sax_transforms, sax_clfs
def _predict(self, X) -> np.ndarray:
"""Predicts labels for sequences in X.
Parameters
----------
X : 3D np.ndarray, shape = [n_instances, n_channels, series_length]
The data to make predictions for.
Returns
-------
y : 1D np.ndarray, shape = [n_instances]
Predicted class labels.
"""
return np.array(
[self.classes_[i] for i in self._predict_proba(X).argmax(axis=1)]
)
def _predict_proba(self, X) -> np.ndarray:
"""Predicts labels probabilities for sequences in X.
Parameters
----------
X : 3D np.ndarray, shape = [n_instances, n_channels, series_length]
The data to make predict probabilities for.
Returns
-------
y : 1D np.ndarray, shape = [n_instances, n_classes_]
Predicted probabilities using the ordering in ``classes_``.
"""
if X.shape[1] == 1: # Univariate
return self._predict_proba_unvivariate(np.squeeze(X))
else: # Multivariate
if self.variant in [1, 2, 3]: # Concatenate
X_concat = X.reshape(*X.shape[:-2], -1)
return self._predict_proba_unvivariate(X_concat)
elif self.variant in [4, 5, 6, 7, 8, 9]:
return self._predict_proba_dimension_ensemble(X) # Ensemble
def _predict_proba_unvivariate(self, X) -> np.ndarray:
"""Predicts labels probabilities for sequences in univariate X.
Parameters
----------
X : 3D np.array of shape = [n_instances, n_channels, series_length]
The data to make predict probabilities for.
Returns
-------
y : array-like, shape = [n_instances, n_classes_]
Predicted probabilities using the ordering in ``classes_``.
"""
pred_mat = np.zeros((X.shape[0], self.n_classes_))
placeholder_y = np.zeros(X.shape[0])
for sfa, (rf, weight) in zip(self.sfa_transforms, self.sfa_clfs):
sfa.fit_transform(X, placeholder_y)
X_sfa = np.array(
[sfa.word_list(word) for word in np.array(sfa.words).ravel()]
)
rf_pred_mat = rf.predict_proba(X_sfa)
if self.variant == 2:
weight = np.mean(rf_pred_mat.max(axis=1))
pred_mat += rf_pred_mat * weight
for X_sax, (rf, weight) in zip(
self._parallel_sax(self.sax_transforms, X), self.sax_clfs
):
rf_pred_mat = rf.predict_proba(X_sax)
if self.variant == 2:
weight = np.mean(rf_pred_mat.max(axis=1))
pred_mat += rf_pred_mat * weight
pred_mat /= np.sum(pred_mat, axis=1).reshape(-1, 1) # Rescales rows to sum to 1
return pred_mat
def _predict_proba_dimension_ensemble(self, X) -> np.ndarray:
"""Predicts labels probabilities using ensemble over the dimensions.
Parameters
----------
X : 3D np.array of shape = [n_instances, n_channels, series_length]
The data to make predict probabilities for.
``n_channels > 1``
Returns
-------
y : array-like, shape = [n_instances, n_classes_]
Predicted probabilities using the ordering in ``classes_``.
"""
ensemble_pred_mats = None
placeholder_y = np.zeros(X.shape[0])
for d in range(self._n_channels):
sfa_transforms = self.sfa_transforms[d]
sfa_clfs = self.sfa_clfs[d]
sax_transforms = self.sax_transforms[d]
sax_clfs = self.sax_clfs[d]
X_d = X[:, d, :]
if self.variant in [6, 7, 8, 9]:
dimension_pred_mats = None
for sfa, (rf, _) in zip(sfa_transforms, sfa_clfs):
sfa.fit_transform(X_d, placeholder_y)
X_sfa = np.array(
[sfa.word_list(word) for word in np.array(sfa.words).ravel()]
)
rf_pred_mat = rf.predict_proba(X_sfa)
if self.variant in [4, 5]:
if ensemble_pred_mats is None:
ensemble_pred_mats = [rf_pred_mat]
else:
ensemble_pred_mats = np.concatenate(
(ensemble_pred_mats, [rf_pred_mat])
)
elif self.variant in [6, 7, 8, 9]:
if dimension_pred_mats is None:
dimension_pred_mats = [rf_pred_mat]
else:
dimension_pred_mats = np.concatenate(
(dimension_pred_mats, [rf_pred_mat])
)
for X_sax, (rf, _) in zip(
self._parallel_sax(sax_transforms, X_d), sax_clfs
):
rf_pred_mat = rf.predict_proba(X_sax)
if self.variant in [4, 5]:
if ensemble_pred_mats is None:
ensemble_pred_mats = [rf_pred_mat]
else:
ensemble_pred_mats = np.concatenate(
(ensemble_pred_mats, [rf_pred_mat])
)
elif self.variant in [6, 7, 8, 9]:
if dimension_pred_mats is None:
dimension_pred_mats = [rf_pred_mat]
else:
dimension_pred_mats = np.concatenate(
(dimension_pred_mats, [rf_pred_mat])
)
if self.variant in [6, 7, 8, 9]:
if self.variant in [6, 7]:
fused_dimension_pred_mat = np.sum(dimension_pred_mats, axis=0)
elif self.variant in [8, 9]:
weights = np.array(
[np.mean(mat.max(axis=1)) for mat in dimension_pred_mats]
).reshape(-1, 1)
fused_dimension_pred_mat = np.sum(
dimension_pred_mats * weights[:, np.newaxis], axis=0
)
if ensemble_pred_mats is None:
ensemble_pred_mats = [fused_dimension_pred_mat]
else:
ensemble_pred_mats = np.concatenate(
(ensemble_pred_mats, [fused_dimension_pred_mat])
)
if self.variant in [4, 6, 7]:
pred_mat = np.sum(np.array(ensemble_pred_mats), axis=0)
elif self.variant in [5, 8, 9]:
weights = np.array(
[np.mean(mat.max(axis=1)) for mat in ensemble_pred_mats]
).reshape(-1, 1)
pred_mat = np.sum(ensemble_pred_mats * weights[:, np.newaxis], axis=0)
pred_mat /= np.sum(pred_mat, axis=1).reshape(-1, 1) # Rescales rows to sum to 1
return pred_mat
def _get_random_lenses(self, X, n_lenses):
"""Randomly select <word length, alphabet size> pairs.
Parameters
----------
X : 3D np.ndarray, shape = [n_instances, n_channels, n_timepoints]
The training data.
n_lenses : int
Number of lenses to select.
Returns
-------
lenses : list of list
Randomly selected lenses.
"""
maxCoof = 130
if X.shape[1] < maxCoof:
maxCoof = X.shape[1] - 1
if X.shape[1] < 100:
n_segments = range(5, maxCoof, 5)
else:
n_segments = range(10, maxCoof, 10)
maxBin = 26
if X.shape[1] < maxBin:
maxBin = X.shape[1] - 2
if X.shape[0] < maxBin:
maxBin = X.shape[0] - 2
alphas = range(3, maxBin)
rng = check_random_state(self.random_state)
lenses = np.transpose(
[rng.choice(n_segments, size=n_lenses), rng.choice(alphas, size=n_lenses)]
).tolist()
return lenses
def _parallel_sax(self, sax_transforms, X):
"""Apply multiple SAX transforms to X in parallel.
Parameters
----------
sax_transforms : list
List of ``SAX()`` instances
X : 2D np.ndarray, shape = [n_instances, n_timepoint]
The data to transform.
"""
def _sax_wrapper(sax):
return np.squeeze(sax.fit_transform(X))
sax_parallel_res = Parallel(n_jobs=self.n_jobs, backend=self.parallel_backend)(
delayed(_sax_wrapper)(sax) for sax in sax_transforms
)
return sax_parallel_res
@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.
Returns
-------
dict
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``.
"""
return {
"variant": 1,
"n_trees": 1,
}