-
Notifications
You must be signed in to change notification settings - Fork 2
/
autood_utils.py
358 lines (292 loc) · 15.7 KB
/
autood_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
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
import pandas as pd
from scipy.io import arff
from sklearn.neighbors import LocalOutlierFactor
from sklearn.ensemble import IsolationForest
from sklearn.neighbors import NearestNeighbors
import multiprocessing
import numpy as np
import scipy as sp
from utils import time_this
from sklearn import metrics
from snorkel.labeling import MajorityLabelVoter
from snorkel.labeling import LabelModel
from SVMMP import SVMMP
from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC
from sklearn.preprocessing import RobustScaler
from sklearn.preprocessing import StandardScaler
def load_dataset(filename):
with open(filename, 'r') as f:
data, meta = arff.loadarff(f)
data = pd.DataFrame(data)
X = data.drop(columns=['id', 'outlier'])
# Map dataframe to encode values and put values into a numpy array
y = data["outlier"].map(lambda x: 1 if x == b'yes' else 0).values
return X, y
def run_knn(X, y, k=60):
neigh = NearestNeighbors(n_neighbors=k)
neigh.fit(X)
knn_dists = neigh.kneighbors(X)[0][:, -1]
return k, knn_dists
def run_lof(X, y, k=60):
clf = LocalOutlierFactor(n_neighbors=k)
clf.fit(X)
lof_scores = -clf.negative_outlier_factor_
return k, lof_scores
def run_isolation_forest(X, y, max_features=1.0):
# training the model
clf = IsolationForest(random_state=42, max_features=max_features)
clf.fit(X)
# predictions
sklearn_score_anomalies = clf.decision_function(X)
if_scores = [-1 * s + 0.5 for s in sklearn_score_anomalies]
return max_features, if_scores
def mahalanobis(x):
"""Compute the Mahalanobis Distance between each row of x and the data
"""
x_minus_mu = x - np.mean(x)
cov = np.cov(x.T)
inv_covmat = sp.linalg.inv(cov)
results = []
x_minus_mu = np.array(x_minus_mu)
for i in range(np.shape(x)[0]):
cur_data = x_minus_mu[i, :]
results.append(np.dot(np.dot(x_minus_mu[i, :], inv_covmat), x_minus_mu[i, :].T))
return np.array(results)
def run_mahalanobis(X, y):
# training the model
dist = mahalanobis(x=X)
return dist
def get_predictions_scores(scores, num_outliers=400, method_name='LOF'):
threshold = np.sort(scores)[::-1][num_outliers]
# threshold, max_f1 = get_best_f1_score(y, lof_scores)
predictions = np.array(scores > threshold)
predictions = np.array([int(i) for i in predictions])
# print('F1 for {} : {}'.format(method_name, metrics.f1_score(y, predictions)))
return predictions, scores
all_results = []
all_scores = []
@time_this
def majority_vote(L, y):
majority_model = MajorityLabelVoter()
preds_train = majority_model.predict_proba(L=L)
threshold = 0.5
predictions = np.full((len(y)), 0)
predictions[preds_train[:, 1] > threshold] = 1
# f1_score = metrics.f1_score(y, predictions)
# print(f"F1 for MV:{f1_score}")
# return f1_score
@time_this
def snorkel(L, y):
label_model = LabelModel(cardinality=2, verbose=True)
label_model.fit(L, seed=123, lr=0.001, n_epochs=1000)
probs_train = label_model.predict_proba(L=L)
threshold = 0.5
predictions = np.full((len(y)), 0)
predictions[probs_train[:, 1] > threshold] = 1
# f1_score = metrics.f1_score(y, predictions)
# print(f"F1 for MV:{f1_score}")
# return f1_score
@time_this
def autood_preprocessing(X, y, lof_krange, knn_krange, if_range, mahalanobis_N_range, if_N_range, N_range):
all_results = []
all_scores = []
with multiprocessing.Pool(8) as pool:
temp_lof_results = dict()
unique_lof_ks = list(set(lof_krange))
processes = [pool.apply_async(run_lof, args=(X, y, k,)) for k in unique_lof_ks]
for p in processes:
k, lof_scores = p.get()
temp_lof_results[k] = lof_scores
for i in range(len(lof_krange)):
lof_predictions, lof_scores = get_predictions_scores(temp_lof_results[lof_krange[i]],
num_outliers=N_range[i], method_name='LOF')
all_results.append(lof_predictions)
all_scores.append(lof_scores)
temp_knn_results = dict()
unique_knn_ks = list(set(knn_krange))
processes = [pool.apply_async(run_knn, args=(X, y, k,)) for k in unique_knn_ks]
for p in processes:
k, knn_scores = p.get()
temp_knn_results[k] = knn_scores
for i in range(len(knn_krange)):
knn_predictions, knn_scores = get_predictions_scores(temp_knn_results[knn_krange[i]],
num_outliers=N_range[i], method_name='KNN')
all_results.append(knn_predictions)
all_scores.append(knn_scores)
temp_if_results = dict()
unique_if_features = list(set(if_range))
processes = [pool.apply_async(run_isolation_forest, args=(X, y, k,)) for k in unique_if_features]
for p in processes:
k, if_scores = p.get()
temp_if_results[k] = if_scores
for i in range(len(if_range)):
if_predictions, if_scores = get_predictions_scores(temp_if_results[if_range[i]], num_outliers=if_N_range[i],
method_name='IF')
all_results.append(if_predictions)
all_scores.append(if_scores)
mahalanobis_scores = run_mahalanobis(X, y)
for i in range(len(mahalanobis_N_range)):
mahalanobis_predictions, mahalanobis_scores = get_predictions_scores(mahalanobis_scores,
num_outliers=mahalanobis_N_range[i],
method_name='mahala')
all_results.append(mahalanobis_predictions)
all_scores.append(mahalanobis_scores)
L = np.stack(all_results).T
scores = np.stack(all_scores).T
return L, scores
@time_this
def autood(L, scores, X, y, max_iteration=10):
# prepare scores for training
prediction_result_list = []
classifier_result_list = []
prediction_list = []
cur_f1_scores = []
prediction_high_conf_outliers = np.array([])
prediction_high_conf_inliers = np.array([])
prediction_classifier_disagree = np.array([])
index_range = np.array([[0, 60], [60, 120], [120, 150], [150, 156]])
coef_index_range = np.array([[0, 10], [10, 20], [20, 25], [25, 26]])
coef_remain_index = range(156)
scores_for_training_indexes = []
for i in range(len(index_range)):
start = index_range[i][0]
temp_range = coef_index_range[i][1] - coef_index_range[i][0]
scores_for_training_indexes = scores_for_training_indexes + list(range(start, start + temp_range))
scores_for_training = scores[:, np.array(scores_for_training_indexes)]
# first iteration
high_confidence_threshold = 0.99
low_confidence_threshold = 0.01
max_iter = 200
remain_params_tracking = np.array(range(0, np.max(coef_index_range)))
training_data_F1 = []
two_prediction_corr = []
min_max_diff = []
N_size = 6
last_training_data_indexes = []
counter = 0
for i_range in range(0, max_iteration):
print("##################################################################")
print('Iteration = {}, L shape = {}'.format(i_range, np.shape(L)))
num_methods = np.shape(L)[1]
agree_outlier_indexes = np.sum(L, axis=1) == np.shape(L)[1]
# print('All agree, Number of outliers = {}'.format(sum(agree_outlier_indexes)))
agree_inlier_indexes = np.sum(L, axis=1) == 0
# print('All agree, Number of inliers = {}'.format(sum(agree_inlier_indexes)))
disagree_indexes = np.where(np.logical_or(np.sum(L, axis=1) == 0, np.sum(L, axis=1) == num_methods) == 0)[0]
all_inlier_indexes = np.setdiff1d(np.where(agree_inlier_indexes)[0], prediction_high_conf_outliers)
if len(prediction_high_conf_inliers) > 0:
all_inlier_indexes = np.intersect1d(
np.setdiff1d(np.where(agree_inlier_indexes)[0], prediction_high_conf_outliers),
prediction_high_conf_inliers)
# print('num of inliers = {}'.format(np.shape(all_inlier_indexes)[0]))
all_outlier_indexes = np.union1d(np.where(agree_outlier_indexes)[0], prediction_high_conf_outliers)
# print('num of outliers = {}'.format(np.shape(all_outlier_indexes)[0]))
all_inlier_indexes = np.setdiff1d(all_inlier_indexes, prediction_classifier_disagree)
self_agree_index_list = []
if ((len(all_outlier_indexes) == 0) or (len(all_inlier_indexes) / len(all_outlier_indexes) > 1000)):
for i in range(0, len(index_range)):
if (index_range[i, 1] - index_range[i, 0] <= 6):
continue
temp_index = disagree_indexes[np.where(
np.sum(L[disagree_indexes][:, index_range[i, 0]: index_range[i, 1]], axis=1) == (
index_range[i, 1] - index_range[i, 0]))[0]]
self_agree_index_list = np.union1d(self_agree_index_list, temp_index)
self_agree_index_list = [int(i) for i in self_agree_index_list]
# self_agree_index_list = np.random.RandomState(1).permutation(self_agree_index_list)[:500]
all_outlier_indexes = np.union1d(all_outlier_indexes, self_agree_index_list)
all_outlier_indexes = np.setdiff1d(all_outlier_indexes, prediction_classifier_disagree)
# print('num of outliers = {}'.format(np.shape(all_outlier_indexes)[0]))
data_indexes = np.concatenate((all_inlier_indexes, all_outlier_indexes), axis=0)
data_indexes = np.array([int(i) for i in data_indexes])
labels = np.concatenate((np.zeros(len(all_inlier_indexes)), np.ones(len(all_outlier_indexes))), axis=0)
transformer = RobustScaler().fit(scores_for_training)
scores_transformed = transformer.transform(scores_for_training)
training_data = scores_transformed[data_indexes]
# print('Training data shape: ', np.shape(training_data))
# training_data_F1.append(metrics.f1_score(y[data_indexes], labels))
# print('Training data F-1', metrics.f1_score(y[data_indexes], labels))
clf = LogisticRegression(random_state=0, penalty='l2', max_iter=max_iter).fit(training_data, labels)
clf_predictions = clf.predict(scores_transformed)
clf_predict_proba = clf.predict_proba(scores_transformed)[:, 1]
# print("F-1 score from LR:",metrics.f1_score(y, np.array([int(i) for i in clf_predict_proba > 0.5])))
transformer = RobustScaler().fit(X)
X_transformed = transformer.transform(X)
X_training_data = X_transformed[data_indexes]
clf_X = SVC(gamma='auto', probability=True, random_state=0)
clf_X.fit(X_training_data, labels)
clf_predictions_X = clf_X.predict(X_transformed)
clf_predict_proba_X = clf_X.predict_proba(X_transformed)[:,1]
# clf_X = SVMMP(n_partition=4)
# clf_X.fit(X_training_data, labels.astype(int), set_partition=False)
# clf_predictions_X = clf_X.predict(X, 'Mean Proba')
# clf_predict_proba_X = clf_X.predict_proba(X_transformed)
#
# print("F-1 score from SVM:",metrics.f1_score(y, np.array([int(i) for i in clf_predict_proba_X > 0.5])))
agreed_outlier_indexes = np.where(np.sum(L, axis=1) == np.shape(L)[1])[0]
agreed_inlier_indexes = np.where(np.sum(L, axis=1) == 0)[0]
prediction_result_list.append(clf_predict_proba)
classifier_result_list.append(clf_predict_proba_X)
prediction_list.append(np.array([int(i) for i in clf_predictions]))
prediction_high_conf_outliers = np.intersect1d(
np.where(prediction_result_list[-1] > high_confidence_threshold)[0],
np.where(classifier_result_list[-1] > high_confidence_threshold)[0])
# print('length of prediction_high_conf_outliers:' , len(prediction_high_conf_outliers))
prediction_high_conf_inliers = np.intersect1d(
np.where(prediction_result_list[-1] < low_confidence_threshold)[0],
np.where(classifier_result_list[-1] < low_confidence_threshold)[0])
# print('length of prediction high conf inliers: ', len(prediction_high_conf_inliers))
temp_prediction = np.array([int(i) for i in prediction_result_list[-1] > 0.5])
temp_classifier = np.array([int(i) for i in classifier_result_list[-1] > 0.5])
prediction_classifier_disagree = np.where(temp_prediction != temp_classifier)[0]
two_prediction_corr.append(np.corrcoef(clf_predict_proba, clf_predict_proba_X)[0, 1])
if np.max(coef_index_range) >= 2:
if (len(prediction_high_conf_outliers) > 0 and len(prediction_high_conf_inliers) > 0):
new_data_indexes = np.concatenate((prediction_high_conf_outliers, prediction_high_conf_inliers), axis=0)
new_data_indexes = np.array([int(i) for i in new_data_indexes])
new_labels = np.concatenate(
(np.ones(len(prediction_high_conf_outliers)), np.zeros(len(prediction_high_conf_inliers))), axis=0)
clf_prune_2 = LogisticRegression(random_state=0, penalty='l2', max_iter=max_iter).fit(
scores_transformed[new_data_indexes], new_labels)
combined_coef = clf_prune_2.coef_[0]
else:
combined_coef = clf.coef_[0]
if (np.max(coef_index_range) >= 2):
if (len(set(combined_coef)) > 1):
cur_clf_coef = combined_coef
cutoff = max(max(0, np.mean(combined_coef) - np.std(combined_coef)), min(combined_coef))
remain_indexes_after_cond = (
cur_clf_coef > cutoff) # np.logical_and(cur_clf_coef > cutoff, abs(cur_clf_coef) > 0.01) # #
remain_params_tracking = remain_params_tracking[remain_indexes_after_cond]
remain_indexes_after_cond_expanded = []
for i in range(0, len(coef_index_range)): #
s_e_range = coef_index_range[i, 1] - coef_index_range[i, 0]
s1, e1 = coef_index_range[i, 0], coef_index_range[i, 1]
s2, e2 = index_range[i, 0], index_range[i, 1]
saved_indexes = np.where(cur_clf_coef[s1:e1] > cutoff)[0]
for j in range(N_size):
remain_indexes_after_cond_expanded.extend(np.array(saved_indexes) + j * s_e_range + s2)
new_coef_index_range_seq = []
for i in range(0, len(coef_index_range)): #
s, e = coef_index_range[i, 0], coef_index_range[i, 1]
new_coef_index_range_seq.append(sum((remain_indexes_after_cond)[s:e]))
coef_index_range = []
index_range = []
cur_sum = 0
for i in range(0, len(new_coef_index_range_seq)):
coef_index_range.append([cur_sum, cur_sum + new_coef_index_range_seq[i]])
index_range.append([cur_sum * 6, 6 * (cur_sum + new_coef_index_range_seq[i])])
cur_sum += new_coef_index_range_seq[i]
coef_index_range = np.array(coef_index_range)
index_range = np.array(index_range)
L = L[:, remain_indexes_after_cond_expanded]
scores_for_training = scores_for_training[:, remain_indexes_after_cond]
if ((len(last_training_data_indexes) == len(data_indexes)) and
(sum(last_training_data_indexes == data_indexes) == len(data_indexes)) and
(np.max(coef_index_range) < 2)):
counter = counter + 1
else:
counter = 0
if (counter > 3):
break
last_training_data_indexes = data_indexes