-
Notifications
You must be signed in to change notification settings - Fork 2
/
transformers.py
531 lines (415 loc) · 20.4 KB
/
transformers.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
from sklearn.base import TransformerMixin
from sklearn import preprocessing
import pickle as pkl
import numpy as np
import warnings
#from .utils import closest_wavelength, ignore_warnings
# must redefine functions here as .utils depends on transformers
def find_wavelength(k, waves, validate=True, tol=5):
''' Index of closest wavelength '''
i = np.abs(np.array(waves) - k).argmin()
assert(not validate or (abs(k-waves[i]) <= tol)), f'Needed {k}nm, but closest was {waves[i]}nm in {waves}'
return i
def validate_wavelength(k, waves, validate=True, tol=5):
''' Index of closest wavelength '''
i = np.abs(np.array(waves) - k).argmin()
less_than_tol_bool = abs(k-waves[i]) <= tol
#if less_than_tol_bool:
#print('Available wavelength {} is within {} nm of desired wavelength {}'.format(waves[i],tol,k))
#assert(not validate or (abs(k-waves[i]) <= tol)), f'Needed {k}nm, but closest was {waves[i]}nm in {waves}'
return less_than_tol_bool
def closest_wavelength(k, waves, validate=True, tol=5):
''' Value of closest wavelength '''
return waves[find_wavelength(k, waves, validate, tol)]
class CustomTransformer(TransformerMixin):
''' Data transformer class which validates data shapes.
Child classes should override _fit, _transform, _inverse_transform '''
_input_shape = None
_output_shape = None
def fit(self, X, *args, **kwargs):
self._input_shape = X.shape[1]
return self._fit(X.copy(), *args, **kwargs)
def transform(self, X, *args, **kwargs):
# print('XSCALER SHAPES',self._input_shape,X.shape[1],X.shape[0])
if self._input_shape is not None:
assert(X.shape[1] == self._input_shape), f'Number of data features changed: {self._input_shape} vs {X.shape[1]}'
X = self._transform(X.copy(), *args, **kwargs)
if self._output_shape is not None:
assert(X.shape[1] == self._output_shape), f'Number of data features changed: {self._output_shape} vs {X.shape[1]}'
self._output_shape = X.shape[1]
return X
def inverse_transform(self, X, *args, **kwargs):
if self._output_shape is not None:
assert(X.shape[1] == self._output_shape), f'Number of data features changed: {self._output_shape} vs {X.shape[1]}'
X = self._inverse_transform(X.copy(), *args, **kwargs)
if self._input_shape is not None:
assert(X.shape[1] == self._input_shape), f'Number of data features changed: {self._input_shape} vs {X.shape[1]}'
self._input_shape = X.shape[1]
return X
def return_labels(self):
return self.return_labels()
def _fit(self, X, *args, **kwargs): return self
def _transform(self, X, *args, **kwargs): raise NotImplemented
def _inverse_transform(self, X, *args, **kwargs): raise NotImplemented
class IdentityTransformer(CustomTransformer):
def _transform(self, X, *args, **kwargs): return X
def _inverse_transform(self, X, *args, **kwargs): return X
class LogTransformer(CustomTransformer):
def _transform(self, X, *args, **kwargs): return np.log(X)
def _inverse_transform(self, X, *args, **kwargs): return np.exp(X)
class NegLogTransformer(CustomTransformer):
'''
Log-like transformation which allows negative values (Whittaker et al. 2005)
http://fmwww.bc.edu/repec/bocode/t/transint.html
'''
def _transform(self, X, *args, **kwargs): return np.sign(X) * np.log(np.abs(X) + 1)
def _inverse_transform(self, X, *args, **kwargs): return np.sign(X) * (np.exp(np.abs(X)) - 1)
class ColumnTransformer(CustomTransformer):
''' Reduce columns to specified selections (feature selection) '''
def __init__(self, columns, *args, **kwargs): self._c = columns
def _transform(self, X, *args, **kwargs): return X[:, self._c]
class BaggingColumnTransformer(CustomTransformer):
''' Randomly select a percentage of columns to drop '''
percent = 0.75
def __init__(self, n_bands, *args, n_extra=0, **kwargs):
self.n_bands = n_bands
self.n_extra = n_extra
def _fit(self, X, *args, **kwargs):
# if X.shape[1] > 60:
# self.percent = 0.05
# n_bands_tmp = self.n_bands
# self.n_bands = 27
shp = X.shape[1] - self.n_bands
ncol = int(shp*self.percent)
cols = np.arange(shp-self.n_extra) + self.n_bands
np.random.shuffle(cols)
# if X.shape[1] > 60:
# shp2 = self.n_bands - n_bands_tmp
# ncol2 = int(shp2*0.75)
# cols2 = np.arange(shp2) + n_bands_tmp
# np.random.shuffle(cols2)
# self.cols = np.append(np.arange(n_bands_tmp), cols2)
# self.cols = np.append(self.cols, cols[:ncol])
# ncol += ncol2
# else:
if self.n_extra:
self.cols = np.append(np.arange(self.n_bands), list(cols[:ncol]) + list(X.shape[1]-(np.arange(self.n_extra)+1)), 0)
else:
self.cols = np.append(np.arange(self.n_bands), list(cols[:ncol]), 0)
# print(f'Reducing bands from {shp} ({X.shape[1]} total) to {ncol} ({len(self.cols)} total) ({self.cols})')
return self
def _transform(self, X, *args, **kwargs):
return X[:, self.cols.astype(int)]
class ExclusionTransformer(CustomTransformer):
'''
Exclude certain columns from being transformed by the given transformer.
The passed in transformer should be a transformer class, and exclude_slice can
be any object which, when used to slice a numpy array, will give the
appropriate columns which should be excluded. So, for example:
- slice(1)
- slice(-3, None)
- slice(1,None,2)
- np.array([True, False, False, True])
etc.
'''
def __init__(self, exclude_slice, transformer, transformer_args=[], transformer_kwargs={}):
self.excl = exclude_slice
self.transformer = transformer(*transformer_args, **transformer_kwargs)
def _fit(self, X):
cols = np.arange(X.shape[1])
cols = [c for c in cols if c not in cols[self.excl]]
self.transformer.fit(X[:, cols])
self.keep = cols
return self
def _transform(self, X, *args, **kwargs):
Z = np.zeros_like(X)
Z[:, self.keep] = self.transformer.transform(X[:, self.keep])
Z[:, self.excl] = X[:, self.excl]
return Z
def _inverse_transform(self, X, *args, **kwargs):
Z = np.zeros_like(X)
Z[:, self.keep] = self.transformer.inverse_transform(X[:, self.keep])
Z[:, self.excl] = X[:, self.excl]
return Z
class RatioTransformer(CustomTransformer):
''' Add ratio features '''
def __init__(self, wavelengths, only_ratio_bool=False,BRs=None,LHs=None,only_append_LH=None,*args, label='', **kwargs):
self.wavelengths = list(wavelengths)
self.label = label
self.only_ratio_bool = only_ratio_bool
self.BRs = BRs if BRs != None else None
self.LHs = LHs if LHs != None else None
self.only_append_LH = only_append_LH if only_append_LH != None else None
def _fit(self, X):
self.shape = X.shape[1]
return self
def _transform(self, X, *args, **kwargs):
'''
Simple feature engineering method. Add band
ratios as features. Does not add reciprocal
ratios or any other duplicate features;
adds a band sum ratio (based on three-band
Chl retrieval method).
Usage:
# one sample with three features, shaped [samples, features]
x = [[a, b, c]]
y = ratio(x)
# y -> [[a, b, c, b/a, b/(a+c), c/a, c/(a+b), c/b, a/(b+c)]
'''
def LH(L1, L2, L3, R1, R2, R3):
c = (L3 - L2) / (L3 - L1)
return R2 - c*R1 - (1-c)*R3
x = np.atleast_2d(X)
#If we only want to use the ratios, remove the Rrs
if self.only_ratio_bool:
x_new = []
else:
x_new = [v for v in x.T]
label = []
def wavelength_check(wavelength_list,wavelength,greater_bool):
if greater_bool:
return (any(x>wavelength for x in wavelength_list))
else:
return (any(x<wavelength for x in wavelength_list))
#default wrapper that checks that the wavelengths exist, adds specified label and formula if they do
def appendFormula(self,desired_wavelengths,X,x_new,formula,label): #args will be a tuple of the wavelengths, followed by the Rrs
found_wavelengths = []
Rrs = []
for wavelength_count, desired_wavelength in enumerate(desired_wavelengths):
if validate_wavelength(desired_wavelength,self.wavelengths):
found_wavelengths.append(closest_wavelength(desired_wavelength,self.wavelengths))
Rrs.append(x[:, self.wavelengths.index(found_wavelengths[wavelength_count])])
else:
#print('Wavelength {} Not Found'.format(desired_wavelength))
return False # returns false if a wavelength is not found
if self.wavelengths != [500, 507, 515, 523, 530,
538, 546, 554, 563, 571, 579, 588, 596, 605, 614, 623, 632, 641, 651, 660, 670, 679, 689,
699, 709, 719, ]:
if len(set(list(desired_wavelengths))) == len(list(desired_wavelengths)):
if len(set(found_wavelengths)) != len(found_wavelengths):
print('FOUND WAVELENGTHS ARE IDENTICAL',label)
return False
formula_result = formula(found_wavelengths,Rrs)
formula_result[np.isposinf(formula_result) == True] = 1e8
formula_result[np.isneginf(formula_result) == True] = -1e8
x_new.append(formula_result)
self.labels.append(label)
self.labels = []
wavelength_range = self.wavelengths
if self.only_append_LH:
print('NOT applying band ratios')
else:
if self.BRs == None:
for numerator in wavelength_range:
for denominator in wavelength_range:
label_txt = f'{numerator}|{denominator}'
appendFormula(self,[denominator,numerator],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]/Rrs[0],label=label_txt)
else:
for num_denominator in self.BRs:
#if numerator > denominator:
label_txt = f'{num_denominator[0]}|{num_denominator[1]}'
appendFormula(self,[num_denominator[1],num_denominator[0]],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]/Rrs[0],label=label_txt)
if self.LHs == None:
# appends on a wide range of line height algorithms following the standard setup
encircling_wavelengths = [5, 10,15,20]
highest_wavelength = max(wavelength_range)
lowest_wavelength = min(wavelength_range)
for center_wavelength in wavelength_range:
for wavelengths_above_wavelength_below in encircling_wavelengths:
lower_wavelength = center_wavelength - wavelengths_above_wavelength_below
upper_wavelength = center_wavelength + wavelengths_above_wavelength_below
if (lower_wavelength >= lowest_wavelength) and upper_wavelength<=highest_wavelength:
label_txt = f'{lower_wavelength}|{center_wavelength}|{upper_wavelength}'
appendFormula(self,[lower_wavelength,center_wavelength,upper_wavelength],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]-(Rrs[2]+((Rrs[0]-Rrs[2])*(wavelengths[2]-wavelengths[1])/(wavelengths[2]-wavelengths[0]))),label=label_txt)
else:
for center_bandwidth in self.LHs:
lower_wavelength = center_bandwidth[0] - center_bandwidth[1]
upper_wavelength = center_bandwidth[0] + center_bandwidth[1]
center_wavelength = center_bandwidth[0]
label_txt = f'{lower_wavelength}|{center_wavelength}|{upper_wavelength}'
appendFormula(self,[lower_wavelength,center_wavelength,upper_wavelength],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]-(Rrs[2]+((Rrs[0]-Rrs[2])*(wavelengths[2]-wavelengths[1])/(wavelengths[2]-wavelengths[0]))),label=label_txt)
# SLH algorithm from Kudela et Al. RSE
appendFormula(self,[654,714,754],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]-(Rrs[0] +(Rrs[2]-Rrs[0])*((wavelengths[1]-wavelengths[0])/(wavelengths[2]-wavelengths[0]))),label='SLH')
# MCI L1 665, L2 709, L3 754 ,#Multi-Algorithm Indices and Look-Up Table for Chlorophyll-a Retrieval in Highly Turbid WaterBodies Using Multispectral Data
appendFormula(self,[665,709,754],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]-Rrs[0]*(((wavelengths[1]-wavelengths[0])/(wavelengths[2]-wavelengths[0]))*Rrs[2]-Rrs[0]),label='MCI_665')
appendFormula(self,[680,709,754],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]-Rrs[0]*(((wavelengths[1]-wavelengths[0])/(wavelengths[2]-wavelengths[0]))*Rrs[2]-Rrs[0]),label='MCI_680')
# CI, from OCx (https://oceancolor.gsfc.nasa.gov/atbd/chlor_a/)
appendFormula(self,[443,555,670],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]-(Rrs[0]+(wavelengths[1]-wavelengths[0])/(wavelengths[2]-wavelengths[0])*(Rrs[2]-Rrs[0])),label='Color Index (chl)')
# NDCI, Multi-Algorithm Indices and Look-Up Table forChlorophyll-a Retrieval in Highly Turbid WaterBodies Using Multispectral Data
appendFormula(self,[665,709],x,x_new, formula=lambda wavelengths,Rrs: (Rrs[1] - Rrs[0])/(Rrs[1] + Rrs[0]),label='NDCI (chl)')
#Mishra PC
appendFormula(self,[600,700],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]/Rrs[0],label='mishra (600,700)')
# directly calculate Green and NIR max wavelength, add to input
# Schalles et al. found that these locations varied with PC and chl concentration
green_wavelengths = np.asarray(self.wavelengths)
NIR_wavelengths = np.asarray(self.wavelengths)
green_wavelengths = green_wavelengths[np.logical_and(green_wavelengths>550 , green_wavelengths<600)]
NIR_wavelengths = NIR_wavelengths[np.logical_and(NIR_wavelengths>694 , NIR_wavelengths<716)]
green_wavelengths = np.ndarray.tolist(green_wavelengths)
NIR_wavelengths = np.ndarray.tolist(NIR_wavelengths)
appendFormula(self,green_wavelengths,x,x_new, formula=lambda wavelengths,Rrs: np.argmax(np.asarray(Rrs),axis=0),label='Max green location')
appendFormula(self,NIR_wavelengths,x,x_new, formula=lambda wavelengths,Rrs: np.argmax(np.asarray(Rrs),axis=0),label='Max NIR location')
#Hunter PC
appendFormula(self,[600,615,725],x,x_new, formula=lambda wavelengths,Rrs: Rrs[2]*(1/Rrs[1]-1/Rrs[0]),label='hunter (600,615,725)')
#Schalles BR
appendFormula(self,[625,650],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]/Rrs[0],label='Schalles 650/625')
#Decker 1993, Using Rrs instead of R(0-)
appendFormula(self,[600,624,648],x,x_new, formula=lambda wavelengths,Rrs: 0.5*(Rrs[0]+Rrs[2])-Rrs[1],label='Decker 0.5*(R(600)+R(648))-R(624)')
# Mishra 2014
appendFormula(self,[629,659,724],x,x_new, formula=lambda wavelengths,Rrs: (1/Rrs[0]-1/Rrs[1])*Rrs[2],label='Mishra 2014 (1/629-1/659) * 724')
# Simis BR
appendFormula(self,[620,709],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]/Rrs[0],label='Simis 709/620')
appendFormula(self,[665,709],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]/Rrs[0],label='Simis 709/665')
appendFormula(self,[560,620,665],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]-(Rrs[2]+((Rrs[0]-Rrs[2])*(wavelengths[2]-wavelengths[1])/(wavelengths[2]-wavelengths[0]))),label='Nima LH 560,620,665')
appendFormula(self,[665,673,681],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]-(Rrs[2]+((Rrs[0]-Rrs[2])*(wavelengths[2]-wavelengths[1])/(wavelengths[2]-wavelengths[0]))),label='Nima LH 665,673,681')
appendFormula(self,[690,709,720],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]-(Rrs[2]+((Rrs[0]-Rrs[2])*(wavelengths[2]-wavelengths[1])/(wavelengths[2]-wavelengths[0]))),label='Nima LH 690,709,720')
appendFormula(self,[620,650,670],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]-(Rrs[2]+((Rrs[0]-Rrs[2])*(wavelengths[2]-wavelengths[1])/(wavelengths[2]-wavelengths[0]))),label='Nima LH 620,650,670')
appendFormula(self,[640,650,660],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]-(Rrs[2]+((Rrs[0]-Rrs[2])*(wavelengths[2]-wavelengths[1])/(wavelengths[2]-wavelengths[0]))),label='Nima LH 640,650,660')
appendFormula(self,[613,620,627],x,x_new, formula=lambda wavelengths,Rrs: Rrs[1]-(Rrs[2]+((Rrs[0]-Rrs[2])*(wavelengths[2]-wavelengths[1])/(wavelengths[2]-wavelengths[0]))),label='Nima LH 613,620,627')
#Wynne 2010 Characterizing a cyanobacterial bloom in western Lake Erie using satellite imagery and meteorological data
# 665,681,709
appendFormula(self,[665,681,709],x,x_new, formula=lambda wavelengths,Rrs: -1 * (np.pi*Rrs[1]-np.pi*Rrs[0]-(np.pi*Rrs[2]-np.pi*Rrs[0])*(wavelengths[1]-wavelengths[0])/(wavelengths[2]-wavelengths[0])),label='Cyanobacteria Index 665,681,709')
self.n_features = len(self.labels)
return np.hstack([v[:,None] for v in x_new])
def transform2(self, X):
x = np.atleast_2d(X)
x_new = [v for v in x.T]
label = []
# Band ratios
for i, L1 in enumerate(self.wavelengths):
for j, L2 in enumerate(self.wavelengths):
if L1 < L2:
R1 = x[:, i]
R2 = x[:, j]
x_new.append(R2 / R1)
label.append(f'{self.label}{L2}/{L1}')
for k, L3 in enumerate(self.wavelengths):
R3 = x[:, k]
if L3 not in [L1, L2]:
if L1 < L3:
x_new.append(R2 * (1/R1 - 1/R3))
label.append(f'{self.label}{L2}*(1/{L1}-1/{L3})')
else:
x_new.append(R3 * (1/R1 - 1/R2))
label.append(f'{self.label}{L3}*(1/{L1}-1/{L2})')
# Line height variations, examining height of center between two shoulder bands
for i, L1 in enumerate(self.wavelengths):
for j, L2 in enumerate(self.wavelengths):
for k, L3 in enumerate(self.wavelengths):
if (L3 > L2) and (L2 > L1):
c = (L3 - L2) / (L3 - L1)
R1 = x[:, i]
R2 = x[:, j]
R3 = x[:, k]
x_new.append(R2 - c*R1 - (1-c)*R3)
label.append(f'{self.label}({L2}-a{L1}-b{L3})')
self.labels = label
return np.hstack([v[:,None] for v in x_new])
def _inverse_transform(self, X, *args, **kwargs):
return np.array(X)[:, :self.shape]
def return_labels(self):
available_labels = self.labels
return available_labels
class TanhTransformer(CustomTransformer):
''' tanh-estimator (Hampel et al. 1986; Latha & Thangasamy, 2011) '''
scale = 0.01
def _fit(self, X, *args, **kwargs):
m = np.median(X, 0)
d = np.abs(X - m)
a = np.percentile(d, 70, 0)
b = np.percentile(d, 85, 0)
c = np.percentile(d, 95, 0)
Xab = np.abs(X)
Xsi = np.sign(X)
phi = np.zeros(X.shape)
idx = np.logical_and(0 <= Xab, Xab < a)
phi[idx] = X[idx]
idx = np.logical_and(a <= Xab, Xab < b)
phi[idx] = (a * Xsi)[idx]
idx = np.logical_and(b <= Xab, Xab < c)
phi[idx] = (a * Xsi * ((c - Xab) / (c - b)))[idx]
self.mu_gh = np.mean(phi, 0)
self.sig_gh = np.std(phi, 0)
return self
def _transform(self, X, *args, **kwargs):
return 0.5 * (np.tanh(self.scale * ((X - self.mu_gh)/self.sig_gh)) + 1)
def _inverse_transform(self, X, *args, **kwargs):
return ((np.tan(X * 2 - 1) / self.scale) * self.sig_gh) + self.mu_gh
class TransformerPipeline(CustomTransformer):
''' Apply multiple transformers seamlessly '''
def __init__(self, scalers=None):
if scalers is None or len(scalers) == 0:
self.scalers = [
LogTransformer(),
preprocessing.RobustScaler(),
preprocessing.MinMaxScaler((-1, 1)),
]
else:
self.scalers = scalers
def _fit(self, X, *args, **kwargs):
for scaler in self.scalers:
X = scaler.fit_transform(X, *args, **kwargs)
return self
def _transform(self, X, *args, **kwargs):
for scaler in self.scalers:
X = scaler.transform(X, *args, **kwargs)
return X
def _inverse_transform(self, X, *args, **kwargs):
for scaler in self.scalers[::-1]:
X = scaler.inverse_transform(X, *args, **kwargs)
return X
def fit_transform(self, X, *args, **kwargs):
# Manually apply a fit_transform to avoid transforming twice
for scaler in self.scalers:
X = scaler.fit_transform(X, *args, **kwargs)
return X
class TransformerPipeline_ratio(CustomTransformer):
''' Apply multiple transformers seamlessly '''
def __init__(self, scalers=None):
if scalers is None or len(scalers) == 0:
self.scalers = [
LogTransformer(),
preprocessing.RobustScaler(),
preprocessing.MinMaxScaler((-1, 1)),
]
else:
self.scalers = scalers
def _fit(self, X, *args, **kwargs):
for scaler in self.scalers:
X = scaler.fit_transform(X, *args, **kwargs)
return self
def _transform(self, X, *args, **kwargs):
for scaler in self.scalers:
X = scaler.transform(X, *args, **kwargs)
return X
def _inverse_transform(self, X, *args, **kwargs):
for scaler in self.scalers[::-1]:
X = scaler.inverse_transform(X, *args, **kwargs)
return X
def fit_transform(self, X, *args, **kwargs):
# Manually apply a fit_transform to avoid transforming twice
for scaler in self.scalers:
X = scaler.fit_transform(X, *args, **kwargs)
return X
def return_labels(self):
for scaler in self.scalers:
labels = scaler.return_labels()
return labels
class CustomUnpickler(pkl.Unpickler):
''' Ensure the classes are found, without requiring an import '''
_warned = False
def find_class(self, module, name):
if name in globals():
return globals()[name]
return super().find_class(module, name)
def load(self, *args, **kwargs):
with warnings.catch_warnings(record=True) as w:
pickled_object = super().load(*args, **kwargs)
# For whatever reason, warnings does not respect the 'once' action for
# sklearn's "UserWarning: trying to unpickle [...] from version [...] when
# using version [...]". So instead, we catch it ourselves, and set the
# 'once' tracker via the unpickler itself.
if len(w) and not CustomUnpickler._warned:
warnings.warn(w[0].message, w[0].category)
CustomUnpickler._warned = True
return pickled_object