-
Notifications
You must be signed in to change notification settings - Fork 2
/
utils.py
184 lines (148 loc) · 5.96 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
import os
import sys
import numpy as np
import pandas as pd
from tqdm import tqdm
from sklearn.preprocessing import MinMaxScaler
from sklearn import metrics
from glob import glob
import datetime
def read_data(file, normalization='z-score', seed=42):
if file.endswith('.npz'):
data = np.load(file, allow_pickle=True)
x, y = data['X'], data['y']
y = np.array(y, dtype=int)
else:
if file.endswith('pkl'):
func = pd.read_pickle
elif file.endswith('csv'):
func = pd.read_csv
else:
raise NotImplementedError('')
df = func(file)
df.replace([np.inf, -np.inf], np.nan, inplace=True)
df.fillna(method='ffill', inplace=True)
x = df.values[:, :-1]
y = np.array(df.values[:, -1], dtype=int)
# train-test splitting
rng = np.random.RandomState(seed)
idx = rng.permutation(np.arange(len(x)))
x, y = x[idx], y[idx]
norm_idx = np.where(y==0)[0]
anom_idx = np.where(y==1)[0]
split = int(0.5 * len(norm_idx))
train_norm_idx, test_norm_idx = norm_idx[:split], norm_idx[split:]
x_train = x[train_norm_idx]
y_train = y[train_norm_idx]
x_test = x[np.hstack([test_norm_idx, anom_idx])]
y_test = y[np.hstack([test_norm_idx, anom_idx])]
print(f'Original size: [{x.shape}], Normal/Anomaly: [{len(norm_idx)}/{len(anom_idx)}] \n'
f'After splitting: training/testing [{len(x_train)}/{len(x_test)}]')
# normalization
if normalization == 'min-max':
minmax_scaler = MinMaxScaler()
minmax_scaler.fit(x_train)
x_train = minmax_scaler.transform(x_train)
x_test = minmax_scaler.transform(x_test)
elif normalization == 'z-score':
mus = np.mean(x_train, axis=0)
sds = np.std(x_train, axis=0)
sds[sds == 0] = 1
x_train = np.array([(xx - mus) / sds for xx in x_train])
x_test = np.array([(xx - mus) / sds for xx in x_test])
elif normalization == 'scale':
x_train = x_train / 255
x_test = x_test / 255
return x_train, y_train, x_test, y_test
def min_max_normalize(x):
filter_lst = []
for k in range(x.shape[1]):
s = np.unique(x[:, k])
if len(s) <= 1:
filter_lst.append(k)
if len(filter_lst) > 0:
print('remove features', filter_lst)
x = np.delete(x, filter_lst, 1)
scaler = MinMaxScaler()
scaler.fit(x)
x = scaler.transform(x)
return x
def evaluate(y_true, scores):
"""calculate evaluation metrics"""
roc_auc = metrics.roc_auc_score(y_true, scores)
ap = metrics.average_precision_score(y_true, scores)
# F1@k, using real percentage to calculate F1-score
ratio = 100.0 * len(np.where(y_true==0)[0]) / len(y_true)
thresh = np.percentile(scores, ratio)
y_pred = (scores >= thresh).astype(int)
y_true = y_true.astype(int)
precision, recall, f_score, support = metrics.precision_recall_fscore_support(y_true, y_pred, average='binary')
return roc_auc, ap, f_score
def get_data_lst(dataset_dir, dataset):
if dataset == 'FULL':
print(os.path.join(dataset_dir, '*.*'))
data_lst = glob(os.path.join(dataset_dir, '*.*'))
else:
name_lst = dataset.split(',')
data_lst = []
for d in name_lst:
data_lst.extend(glob(os.path.join(dataset_dir, d + '.*')))
data_lst = sorted(data_lst)
if 'fmnist' in dataset_dir:
data_lst = data_lst[::-1]
return data_lst
def adjust_contamination(x_train, y_train, x_test, y_test,
contamination_r, swap_ratio=0.05, random_state=42):
"""
used only for 50%normal-setting
add/remove anomalies in training data to replicate anomaly contaminated data sets.
randomly swap 5% features of two anomalies to avoid duplicate contaminated anomalies.
"""
rng = np.random.RandomState(random_state)
test_anomalies = x_test[np.where(y_test == 1)[0]]
test_inliers = x_test[np.where(y_test == 0)[0]]
a = np.arange(len(test_anomalies))
rng.shuffle(a)
test_anomalies = test_anomalies[a]
anomalies = test_anomalies[:int(0.5 * len(test_anomalies))]
rest_anomalies = test_anomalies[int(0.5 * len(test_anomalies)):]
x_test_new = np.vstack([test_inliers, rest_anomalies])
y_test_new = np.hstack([np.zeros(len(test_inliers)), np.ones(len(rest_anomalies))])
# anomalies = test_anomalies
# # anomalies = test_anomalies[:int(0.5 * len(test_anomalies))]
# x_test_new = x_test
# y_test_new = y_test
# else:
# anomalies = x_train[np.where(y_train==1)[0]]
# x_test_new = x_test
# y_test_new = y_test
n_add_anom = int(len(x_train) * contamination_r / (1. - contamination_r))
n_inj_noise = n_add_anom - len(anomalies)
print(f'Control Contamination Rate: \n'
f'Contain : [{n_add_anom}] Anomalies, '
f'injecting: [{n_inj_noise}] Noisy samples, \n'
f'testing : {len(np.where(y_test_new==1)[0])}/{len(np.where(y_test_new==0)[0])}')
# use all anomalies and inject new anomalies
if n_inj_noise > 0:
n_sample, dim = anomalies.shape
n_swap_feat = int(swap_ratio * dim)
inj_noise = np.empty((n_inj_noise, dim))
for i in np.arange(n_inj_noise):
idx = rng.choice(n_sample, 2, replace=False)
o1 = anomalies[idx[0]]
o2 = anomalies[idx[1]]
swap_feats = rng.choice(dim, n_swap_feat, replace=False)
inj_noise[i] = o1.copy()
inj_noise[i][swap_feats] = o2[swap_feats]
x = np.vstack([x_train, anomalies])
y = np.hstack([y_train, np.ones(n_add_anom)])
x = np.vstack([x, inj_noise])
y = np.hstack([y, np.ones(n_inj_noise)])
# use original anomalies
else:
n_sample, dim = anomalies.shape
idx = rng.choice(n_sample, n_add_anom, replace=False)
x = np.append(x_train, anomalies[idx], axis=0)
y = np.append(y_train, np.ones(n_add_anom))
print(x.shape)
return x, y, x_test_new, y_test_new