-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathdataset.py
258 lines (199 loc) · 13.6 KB
/
dataset.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
import numpy as np
import torch
import torchaudio
import librosa
import librosa.display
from keras.datasets import mnist, cifar10
from binreader import open_binary_file
from pathlib import Path
from typing import Dict
import matplotlib.pyplot as plt
from utils.utils import logical_and_arrays, logical_or_arrays
def import_dataset(name:str, split:float=0.2, shuffle=True, extra_args:Dict[str, bool]={}):
datasets = {"trend":import_data_TREND,
"noise_trend": import_noise_TREND,
"mnist": import_mnist,
"cifar10":import_cifar10,
"noisy_cifar10":import_noisy_cifar10,
}
if name in datasets:
return datasets[name](split, shuffle, extra_args)
else:
raise ValueError("This key is not associated with a dataset")
def add_impurity(labels:np.ndarray, nb_classes:int, impure_class:int, impurity_level:float):
if impurity_level == 0:
return labels
assert np.max(labels) <=1, "this is for biclassification"
indicies_impure = np.where((np.random.rand(labels.shape[0])<impurity_level) & (labels[:, 0]==impure_class))[0]
if nb_classes == 2:
labels[indicies_impure] = 1 - impure_class
else:
raise ValueError("Not implemented yet")
return labels
def import_mnist(split:float, shuffle:bool, extra_args:Dict[str, bool]):
print(Warning("Split is not supported for MNIST yet"))
(data_train, labels_train), (data_test, labels_test) = mnist.load_data()
max_classes = 10
if "max_classes" in extra_args:
max_classes = extra_args["max_classes"]
if max_classes is not None:
if type(max_classes) == int:
max_classes_size = max_classes
if max_classes <10:
indicies_train = np.where(labels_train<max_classes)[0]
data_train, labels_train = data_train[indicies_train], labels_train[indicies_train]
indicies_test = np.where(labels_test<max_classes)[0]
data_test, labels_test = data_test[indicies_test], labels_test[indicies_test]
elif(type(max_classes) == list):
max_classes_size = len(max_classes)
indicies_train = [labels_train[:, 0] == elem for elem in max_classes]
for incr in range(len(max_classes)):
labels_train[indicies_train[incr]] = incr
data_train, labels_train = data_train[logical_or_arrays(indicies_train)], labels_train[logical_or_arrays(indicies_train)]
indicies_test = [labels_test[:, 0] == elem for elem in max_classes]
for incr in range(len(max_classes)):
labels_test[indicies_test[incr]] = incr
data_test, labels_test = data_test[logical_or_arrays(indicies_test)], labels_test[logical_or_arrays(indicies_test)]
else:
raise TypeError("MaxClasses is not the good type")
#We add noise in the 1 class
if extra_args["impurity_level"]>0:
labels_train_impure = add_impurity(labels_train, max_classes_size, impure_class=extra_args["impure_class"], impurity_level=extra_args["impurity_level"])
labels_train_dict = {'clean': labels_train, 'noisy': labels_train_impure}
return (data_train, labels_train_dict), (data_test, labels_test)
def import_cifar10(split:float, shuffle:bool, extra_args:Dict[str, bool]):
print(Warning("Split is not supported for Cifar yet"))
(data_train, labels_train), (data_test, labels_test) = cifar10.load_data()
data_train, data_test = np.swapaxes(np.swapaxes(data_train, 2, 3), 1, 2)/255, np.swapaxes(np.swapaxes(data_test, 2, 3), 1, 2)/255
max_classes = 10
if "max_classes" in extra_args:
max_classes = extra_args["max_classes"]
if max_classes is not None:
if type(max_classes) == int:
max_classes_size = max_classes
if max_classes <10:
indicies_train = np.where(labels_train<max_classes)[0]
data_train, labels_train = data_train[indicies_train], labels_train[indicies_train]
indicies_test = np.where(labels_test<max_classes)[0]
data_test, labels_test = data_test[indicies_test], labels_test[indicies_test]
elif(type(max_classes) == list):
max_classes_size = len(max_classes)
indicies_train = [labels_train[:, 0] == elem for elem in max_classes]
for incr in range(len(max_classes)):
labels_train[indicies_train[incr]] = incr
data_train, labels_train = data_train[logical_or_arrays(indicies_train)], labels_train[logical_or_arrays(indicies_train)]
indicies_test = [labels_test[:, 0] == elem for elem in max_classes]
for incr in range(len(max_classes)):
labels_test[indicies_test[incr]] = incr
data_test, labels_test = data_test[logical_or_arrays(indicies_test)], labels_test[logical_or_arrays(indicies_test)]
else:
raise TypeError("max_classes has not the good type")
#We add noise in the 1 class
labels_train_impure = np.copy(labels_train)
if extra_args["impurity_level"]>0:
labels_train_impure = add_impurity(labels_train, max_classes_size, impure_class=extra_args["impure_class"], impurity_level=extra_args["impurity_level"])
labels_train_dict = {'clean': labels_train, 'noisy': labels_train_impure}
return (data_train, labels_train_dict), (data_test, labels_test)
def import_noisy_cifar10(split:float, shuffle:bool, extra_args:Dict[str, bool]):
noise_file = torch.load('./data/CIFAR-10_human.pt')
labels_rand = noise_file['worse_label']
labels_rand = np.expand_dims(labels_rand, axis=-1)
max_classes = 10
if "max_classes" in extra_args:
max_classes = extra_args["max_classes"]
(data_train, labels_train), (data_test, labels_test) = cifar10.load_data()
if max_classes is not None:
if type(max_classes) == int:
if max_classes <10:
indicies_train = np.where((labels_train<max_classes) & (labels_rand<max_classes))[0]
data_train, new_labels, labels_train = data_train[indicies_train], labels_rand[indicies_train], labels_train[indicies_train]
indicies_test = np.where(labels_test<max_classes)[0]
data_test, labels_test = data_test[indicies_test], labels_test[indicies_test]
elif(type(max_classes) == list):
indicies_train = [labels_train[:, 0] == elem for elem in max_classes]
for incr in range(len(max_classes)):
labels_train[indicies_train[incr]] = incr
indicies_rand = [labels_rand[:, 0] == elem for elem in max_classes]
for incr in range(len(max_classes)):
labels_rand[indicies_rand[incr]] = incr
indicies = logical_and_arrays([logical_or_arrays(indicies_train), logical_or_arrays(indicies_rand)])
data_train, new_labels, labels_train = data_train[indicies], labels_rand[indicies], labels_train[indicies]
indicies_test = [labels_test[:, 0] == elem for elem in max_classes]
for incr in range(len(max_classes)):
labels_test[indicies_test[incr]] = incr
data_test, labels_test = data_test[logical_or_arrays(indicies_test)], labels_test[logical_or_arrays(indicies_test)]
else:
raise TypeError("max_classes has not the good type")
data_train, data_test = np.swapaxes(np.swapaxes(data_train, 2, 3), 1, 2)/255, np.swapaxes(np.swapaxes(data_test, 2, 3), 1, 2)/255
labels_train_dict = {'clean': labels_train, 'noisy': new_labels}
return (data_train, labels_train_dict), (data_test, labels_test)
def import_noise_TREND(split:float, shuffle:bool, extra_args:Dict[str, bool]):
"""Import only a file with noise
"""
data_anthropique = open_binary_file(Path("./data/MLP6_transient_2.bin"))/255
data_anthropique = np.expand_dims(data_anthropique[:, 256:], axis=1)
data_anthropique = data_anthropique - np.expand_dims(np.mean(data_anthropique, axis=-1), axis=-1) #We normalize the input
labels = np.zeros((len(data_anthropique), 1))
return (None, None), (data_anthropique, labels)
def import_data_TREND(split:float, shuffle:bool, extra_args:Dict[str, bool]):
#Data for signal analysis
if "mode" in extra_args:
preprocessing_mode = extra_args["mode"]
else:
preprocessing_mode = None
data_selected = open_binary_file(Path("./data/MLP6_selected.bin"))/255
data_anthropique = open_binary_file(Path("./data/MLP6_transient.bin"))/255
if "import_more_noise" in extra_args and extra_args["import_more_noise"]:
data_anthropique2 = open_binary_file(Path("./data/MLP6_transient_2.bin"))/255
data_anthropique3 = open_binary_file(Path("./data/MLP6_transient_3.bin"))/255
data_anthropique4 = open_binary_file(Path("./data/MLP6_transient_4.bin"))/255
data_anthropique5 = open_binary_file(Path("./data/MLP6_transient_5.bin"))/255
data_anthropique = np.concatenate([data_anthropique, data_anthropique2, data_anthropique3, data_anthropique4, data_anthropique5], axis=0)
data_selected = data_selected[:, 256:] #We remove the beginning where there is nothing
data_anthropique = data_anthropique[:, 256:]
data_size_signal = len(data_selected)
data_size_noise = len(data_anthropique)
print(data_selected.shape)
print(data_anthropique.shape)
indicies = np.arange(data_size_signal)
np.random.shuffle(indicies)
data_selected = data_selected[indicies]
indicies = np.arange(data_size_noise)
np.random.shuffle(indicies)
data_anthropique = data_anthropique[indicies]
if "import_more_noise" in extra_args and extra_args["import_more_noise"]: #We put it here because if we shuffle the dataset after concatenating the train and test data will be identical
data_selected = np.repeat(data_selected, 5, axis=0)
data_size_signal = len(data_selected)
if preprocessing_mode == "fft":
data_selected_fft = np.fft.fft(data_selected)
data_anthropique_fft = np.fft.fft(data_anthropique)
data_selected_fft_r = data_selected_fft.real/np.expand_dims(np.maximum(np.max(data_selected_fft.real, axis=-1), -np.min(data_selected_fft.real, axis=-1)), axis=-1)
data_selected_fft_i = data_selected_fft.imag/np.expand_dims(np.maximum(np.max(data_selected_fft.imag, axis=-1), -np.min(data_selected_fft.imag, axis=-1)), axis=-1)
data_selected = np.stack([data_selected, data_selected_fft_r, data_selected_fft_i], axis=1)
data_anthropique_fft_r = data_anthropique_fft.real/np.expand_dims(np.maximum(np.max(data_anthropique_fft.real, axis=-1), -np.min(data_anthropique_fft.real, axis=-1)), axis=-1)
data_anthropique_fft_i = data_anthropique_fft.imag/np.expand_dims(np.maximum(np.max(data_anthropique_fft.imag, axis=-1), -np.min(data_anthropique_fft.imag, axis=-1)), axis=-1)
data_anthropique = np.stack([data_anthropique, data_anthropique_fft_r, data_anthropique_fft_i], axis=1)
data_train = np.concatenate([data_selected[:int(data_size_signal*(1-split))], data_anthropique[:int(data_size_noise*(1-split))]], axis=0)
data_test = np.concatenate([data_selected[int(data_size_signal*(1-split)):], data_anthropique[int(data_size_noise*(1-split)):]], axis=0)
elif preprocessing_mode == "spectrogram":
data = np.concatenate([data_selected, data_anthropique], axis=0)
sgram = librosa.amplitude_to_db(abs(librosa.stft(data, n_fft=80)))
sgram = np.pad(sgram, [(0,0), (0, 1), (1, 2)])
sgram = sgram - np.mean(sgram, axis=(-2, -1), keepdims=True)
sgram = sgram/np.maximum(np.max(sgram, axis=(-2, -1), keepdims=True), -np.min(sgram, axis=(-2, -1), keepdims=True))
data_signal = sgram[:data_size_signal]
data_noise = sgram[data_size_signal:]
data_train = np.expand_dims(np.concatenate([data_signal[:int(data_size_signal*(1-split))], data_noise[:int(data_size_noise*(1-split))]], axis=0), axis=1)
data_test = np.expand_dims(np.concatenate([data_signal[int(data_size_signal*(1-split)):], data_noise[int(data_size_noise*(1-split)):]], axis=0), axis=1)
labels_train = np.expand_dims(np.concatenate([np.ones((int(data_size_signal*(1-split)),)), np.zeros((int(data_size_noise*(1-split)),))]), axis=1)
labels_test = np.expand_dims(np.concatenate([np.ones((int(data_size_signal*(split)),)), np.zeros((int(data_size_noise*(split)),))]), axis=1)
return (data_train, labels_train), (data_test, labels_test)
else:
data_train = np.expand_dims(np.concatenate([data_selected[:int(data_size_signal*(1-split))], data_anthropique[:int(data_size_noise*(1-split))]]), axis=1)
data_test = np.expand_dims(np.concatenate([data_selected[int(data_size_signal*(1-split)):], data_anthropique[int(data_size_noise*(1-split)):]]), axis=1)
data_train = data_train - np.expand_dims(np.mean(data_train, axis=-1), axis=-1) #We normalize the input
data_test = data_test - np.expand_dims(np.mean(data_test, axis=-1), axis=-1)
labels_train = np.expand_dims(np.concatenate([np.ones((int(data_size_signal*(1-split)),)), np.zeros((int(data_size_noise*(1-split)),))]), axis=1)
labels_test = np.expand_dims(np.concatenate([np.ones((int(data_size_signal*(split)),)), np.zeros((int(data_size_noise*(split)),))]), axis=1)
labels_train_dict = {'clean': labels_train, 'noisy': labels_train}
return (data_train, labels_train_dict), (data_test, labels_test)