-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathutils.py
290 lines (246 loc) · 11.6 KB
/
utils.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
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
import random
from scipy.stats import multivariate_normal
import math
import pickle
def genMulNormal(mean, covar, n_samples): # genertaes samples based on multivariate normal distribution
return np.random.multivariate_normal(mean, covar, n_samples)
def genSyntData(mean, covar, n_samples, sens_at, cls_at): # generate samples with sensitive and class attribute
sample = genMulNormal(mean, covar, n_samples)
data = [(sample[i], sens_at, cls_at) for i in range(sample.shape[0])]
return data
def gen_gaussian(mean_in, cov_in, n_samples, class_label):
nv = multivariate_normal(mean=mean_in, cov=cov_in)
X = nv.rvs(n_samples)
y = np.ones(n_samples, dtype=float) * class_label
return nv, X, y
def gen_gaussian_diff_size(mean_in, cov_in, z_val, class_label, n):
nv = multivariate_normal(mean = mean_in, cov = cov_in)
X = nv.rvs(n)
y = np.ones(n, dtype=float) * class_label
z = np.ones(n, dtype=float) * z_val # all the points in this cluster get this value of the sensitive attribute
return nv, X, y, z
class Data:
def __init__(self, mean, covar, n_samples):
self.d_0_1 = genSyntData(mean[0], covar[0], int(n_samples//4), 0, 1) # sens = 0, cls = 1
self.d_1_1 = genSyntData(mean[1], covar[1], int(n_samples//4), 1, 1) # sens = 1, cls = 1
self.d_0_0 = genSyntData(mean[2], covar[2], int(n_samples//4), 0, 0) # sens = 0, cls = 0
self.d_1_0 = genSyntData(mean[3], covar[3], int(n_samples//4), 1, 0) # sens = 1, cls = 0
self.tr_dt = None
self.tr_s = None
self.tr_c = None
self.ts_dt = None
self.ts_s = None
self.ts_c = None
def traintestGen(self, train_size=0.2):
d_0_0_test, d_0_0_train = train_test_split(self.d_0_0, train_size=train_size)
d_0_1_test, d_0_1_train = train_test_split(self.d_0_1, train_size=train_size)
d_1_0_test, d_1_0_train = train_test_split(self.d_1_0, train_size=train_size)
d_1_1_test, d_1_1_train = train_test_split(self.d_1_1, train_size=train_size)
train_set = d_0_0_train + d_0_1_train + d_1_0_train + d_1_1_train
test_set = d_0_0_test + d_0_1_test + d_1_0_test + d_1_1_test
random.shuffle(train_set)
train_data, train_sens, train_cls = zip(*train_set)
test_data, test_sens, test_cls = zip(*test_set)
self.tr_dt = np.array(train_data)
self.tr_s = np.array(train_sens)
self.tr_c = np.array(train_cls)
self.ts_dt = np.array(test_data)
self.ts_s = np.array(test_sens)
self.ts_c = np.array(test_cls)
class DataDispImp:
def __init__(self, mean, covar, n_samples, disc_factor):
nv1, X1, y1 = gen_gaussian(mean[0], covar[0], n_samples, 1) # positive class
nv2, X2, y2 = gen_gaussian(mean[1], covar[1], n_samples, 0) # negative class
X = np.vstack((X1, X2))
y = np.hstack((y1, y2))
data_ = list(zip(X, y))
random.shuffle(data_)
rotation_mult = np.array(
[[math.cos(disc_factor), -math.sin(disc_factor)], [math.sin(disc_factor), math.cos(disc_factor)]])
self.data_aux = [(np.dot(r[0], rotation_mult), r[1]) for r in data_]
self.x_control = [] # this array holds the sensitive feature value
for i in range(0, len(self.data_aux)):
x = self.data_aux[i][0]
# probability for each cluster that the point belongs to it
p1 = nv1.pdf(x)
p2 = nv2.pdf(x)
# normalize the probabilities from 0 to 1
s = p1 + p2
p1 = p1 / s
p2 = p2 / s
r = np.random.uniform() # generate a random number from 0 to 1
if r < p1: # the first cluster is the positive class
self.x_control.append(1.0) # 1.0 means its male
else:
self.x_control.append(0.0) # 0.0 -> female
self.x_control = np.array(self.x_control)
d, cls = zip(*self.data_aux)
self.data = list(zip(d, cls, self.x_control))
def train_test_split(self, train_size=0.7, val=True):
random.shuffle(self.data)
if val:
tr_size = int(train_size*len(self.data))
val_size = int((train_size+0.15)*len(self.data))
train_data, train_cls, train_sens = zip(*self.data[:tr_size])
test_data, test_cls, test_sens = zip(*self.data[tr_size:val_size])
val_data, val_cls, val_sens = zip(*self.data[val_size:])
self.vl_dt = np.array(val_data)
self.vl_s = np.array(val_sens)
self.vl_c = np.array(val_cls)
else:
size = int(train_size*len(self.data))
train_data, train_cls, train_sens = zip(*self.data[:size])
test_data, test_cls, test_sens = zip(*self.data[size:])
self.tr_dt = np.array(train_data)
self.tr_s = np.array(train_sens)
self.tr_c = np.array(train_cls)
self.ts_dt = np.array(test_data)
self.ts_s = np.array(test_sens)
self.ts_c = np.array(test_cls)
def test_train_split_comp(self, tune=None):
'''
Splits the data into 80% train and 20% test sets
Tunes on a subset of the test set if None then on the entire test set
'''
random.shuffle(self.data)
size = int(0.8*(len(self.data)))
train_data_ = self.data[:size]
val_data_ = self.data[size:]
train_data, train_cls, train_sens = zip(*train_data_)
val_data, val_cls, val_sens = zip(*val_data_)
if tune is None:
test_data, test_cls, test_sens = zip(*train_data_)
else:
random.shuffle(train_data_)
size = int(tune*len(train_data_))
test_data, test_cls, test_sens = zip(*train_data_[:size])
self.vl_dt = np.array(val_data)
self.vl_s = np.array(val_sens)
self.vl_c = np.array(val_cls)
self.tr_dt = np.array(train_data)
self.tr_s = np.array(train_sens)
self.tr_c = np.array(train_cls)
self.ts_dt = np.array(test_data)
self.ts_s = np.array(test_sens)
self.ts_c = np.array(test_cls)
class DataDispMis:
def __init__(self, mean, covar, n_samples):
nv1, X1, y1, z1 = gen_gaussian_diff_size(mean[0], covar[0], 1, 1, int(n_samples * 1)) # z=1, +
nv2, X2, y2, z2 = gen_gaussian_diff_size(mean[1], covar[1], 0, 1, int(n_samples * 1)) # z=0, +
nv3, X3, y3, z3 = gen_gaussian_diff_size(mean[2], covar[2], 1, 0, int(n_samples * 1)) # z=1, -
nv4, X4, y4, z4 = gen_gaussian_diff_size(mean[3], covar[3], 0, 0, int(n_samples * 1)) # z=0, -
X = np.vstack((X1, X2, X3, X4))
y = np.hstack((y1, y2, y3, y4))
x_control = np.hstack((z1, z2, z3, z4))
self.data = list(zip(X, y, x_control))
def train_test_split(self, train_size=0.7, val=True):
random.shuffle(self.data)
if val:
tr_size = int(train_size*len(self.data))
val_size = int((train_size+0.15)*len(self.data))
train_data, train_cls, train_sens = zip(*self.data[:tr_size])
test_data, test_cls, test_sens = zip(*self.data[tr_size:val_size])
val_data, val_cls, val_sens = zip(*self.data[val_size:])
self.vl_dt = np.array(val_data)
self.vl_s = np.array(val_sens)
self.vl_c = np.array(val_cls)
else:
size = int(train_size*len(self.data))
train_data, train_cls, train_sens = zip(*self.data[:size])
test_data, test_cls, test_sens = zip(*self.data[size:])
self.tr_dt = np.array(train_data)
self.tr_s = np.array(train_sens)
self.tr_c = np.array(train_cls)
self.ts_dt = np.array(test_data)
self.ts_s = np.array(test_sens)
self.ts_c = np.array(test_cls)
def test_train_split_comp(self, tune=None):
'''
Splits the data into 80% train and 20% test sets
Tunes on a subset of the test set if None then on the entire test set
'''
random.shuffle(self.data)
size = int(0.8*(len(self.data)))
train_data_ = self.data[:size]
val_data_ = self.data[size:]
train_data, train_cls, train_sens = zip(*train_data_)
val_data, val_cls, val_sens = zip(*val_data_)
if tune is None:
test_data, test_cls, test_sens = zip(*train_data_)
else:
random.shuffle(train_data_)
size = int(tune*len(train_data_))
test_data, test_cls, test_sens = zip(*train_data_[:size])
self.vl_dt = np.array(val_data)
self.vl_s = np.array(val_sens)
self.vl_c = np.array(val_cls)
self.tr_dt = np.array(train_data)
self.tr_s = np.array(train_sens)
self.tr_c = np.array(train_cls)
self.ts_dt = np.array(test_data)
self.ts_s = np.array(test_sens)
self.ts_c = np.array(test_cls)
class Dataset: # class for real-world datasets
def __init__(self, fname, scale=True):
with open(fname, 'rb') as fs:
data_ = pickle.load(fs)
if scale:
scaler = StandardScaler()
scaler.fit(data_['data'])
scaled_data = scaler.transform(data_['data'])
self.data = list(zip(scaled_data, data_['s_attr'], data_['class']))
else:
self.data = list(zip(data_['data'], data_['s_attr'], data_['class']))
def train_test_split(self, train_size=0.7, val=True):
random.shuffle(self.data)
if val:
tr_size = int(len(self.data)*train_size)
val_size = int((train_size + 0.15) * len(self.data))
train_data, train_sens, train_cls = zip(*self.data[:tr_size])
test_data, test_sens, test_cls = zip(*self.data[tr_size:val_size])
val_data, val_sens, val_cls = zip(*self.data[val_size:])
self.vl_dt = np.array(val_data)
self.vl_s = np.array(val_sens)
self.vl_c = np.array(val_cls)
else:
size = int(len(self.data)*train_size)
train_data, train_sens, train_cls = zip(*self.data[:size])
test_data, test_sens, test_cls = zip(*self.data[size:])
self.tr_dt = np.array(train_data)
self.tr_s = np.array(train_sens)
self.tr_c = np.array(train_cls)
self.ts_dt = np.array(test_data)
self.ts_s = np.array(test_sens)
self.ts_c = np.array(test_cls)
def test_train_split_comp(self, tune=None):
'''
Splits the data into 80% train and 20% test sets
Tunes on a subset of the test set if None then on the entire test set
'''
random.shuffle(self.data)
tr_size = int(len(self.data)*0.8)
train_data_ = self.data[:tr_size]
self.tr_dt, self.tr_s, self.tr_c = zip(*train_data_)
self.vl_dt, self.vl_s, self.vl_c = zip(*self.data[tr_size:])
if tune is None:
self.ts_dt, self.ts_s, self.ts_c = zip(*train_data_)
else:
size = int(tune*len(train_data_))
self.ts_dt, self.ts_s, self.ts_c = zip(*train_data_[:size])
if __name__ == "__main__":
means = [[2, 2], [2, 2], [-2, -2], [-1, 0]]
covars = [[[3, 1], [1, 3]], [[3, 1], [1, 3]], [[3, 1], [1, 3]], [[3, 1], [1, 3]]]
n_samples = 2000
disc_factor = math.pi / 4.0
data = DataDispImp(means, covars, n_samples, disc_factor)
print(data.data_aux[100])
print(data.x_control[100])
print(data.data[100])
#data = Dataset('Datasets/compas_data.p')
data.train_test_split(train_size=0.8, val=False)
print(len(data.tr_dt), len(data.tr_c), len(data.tr_s))
print(len(data.ts_dt), len(data.ts_c), len(data.ts_s))
#print(len(data.vl_dt), len(data.vl_c), len(data.vl_s))