-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_experiment.py
227 lines (189 loc) · 9.95 KB
/
run_experiment.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
import os
import time
from scipy.io import loadmat
import h5py
import pandas as pd
import numpy as np
from tqdm import tqdm
from modAL.models import ActiveLearner
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.cluster import KMeans
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import confusion_matrix
def unisel(data_X, n_instances, seed):
ss = StandardScaler()
kmeans = KMeans(n_clusters=n_instances, n_init=1, random_state=seed)
labels = kmeans.fit_predict(ss.fit_transform(data_X))
unique_labels = list(set(labels))
sample_idx = []
for label in unique_labels:
instance_idx = np.where(labels == label)[0]
dist = np.sqrt(np.sum((ss.transform(data_X[instance_idx]) -
kmeans.cluster_centers_[label])**2, axis=-1))
sample_idx.append(instance_idx[np.argmin(dist)])
nonsample_idx = list(set(np.arange(data_X.shape[0])) -
set(sample_idx))
return sample_idx, nonsample_idx
def active_learning(model, data_X, data_y, sample_idx, n_iter):
nonsample_idx = np.array(list(set(np.arange(data_X.shape[0])) - set(sample_idx)))
learner = ActiveLearner(model, X_training=data_X[sample_idx], y_training=data_y[sample_idx])
for al_idx in range(n_iter):
# query for labels
query_idx, _ = learner.query(data_X[nonsample_idx])
# supply label for queried instance
learner.teach(data_X[nonsample_idx][query_idx], data_y[nonsample_idx][query_idx])
sample_idx.append(nonsample_idx[query_idx][0])
nonsample_idx = np.array(list(set(np.arange(data_X.shape[0])) - set(sample_idx)))
return sample_idx, nonsample_idx
if __name__ == '__main__':
datasets = ['cover.mat',
'http.mat',
'mammography.mat',
'musk.mat',
'optdigits.mat',
'pendigits.mat',
'satimage-2.mat',
'smtp.mat',
'thyroid.mat',
'vowels.mat']
sample_sizes = [10, 50, 100, 500, 1000]
n_repeats = 10
seed = 0
np.random.seed(seed)
seeds = np.random.randint(0, np.iinfo(np.int32).max, size=n_repeats)
for dataset in datasets:
# load data
data_file = os.path.join('data', dataset)
try:
data = loadmat(data_file)
except:
with h5py.File(data_file, 'r') as f:
data = {'X':None, 'y':None}
data['X'] = f['X'][:].T
data['y'] = f['y'][:].T
# create empty list to store results
overall_results = []
for sample_size in sample_sizes:
print('\nstarting dataset', dataset, 'with sample size', sample_size)
baseline_cm = []
random_cm = []
unisel_cm = []
al_cm = []
al_unisel_cm = []
rand_n_selected_outliers = []
rand_n_selected_normals = []
unisel_n_selected_outliers = []
unisel_n_selected_normals = []
al_n_selected_outliers = []
al_n_selected_normals = []
al_unisel_n_selected_outliers = []
al_unisel_n_selected_normals = []
y = data['y'].ravel()
# create results dictionary
results = {'name':os.path.splitext(dataset)[0],
'n_obs':len(y),
'n_outliers':np.sum(y==1),
'n_samples':sample_size,
'baseline_cm':baseline_cm,
'rand_cm':random_cm,
'unisel_cm':unisel_cm,
'al_cm':al_cm,
'al_unisel_cm':al_unisel_cm,
'rand_n_selected_outliers':rand_n_selected_outliers,
'rand_n_selected_normals':rand_n_selected_normals,
'unisel_n_selected_outliers':unisel_n_selected_outliers,
'unisel_n_selected_normals':unisel_n_selected_normals,
'al_n_selected_outliers':al_n_selected_outliers,
'al_n_selected_normals':al_n_selected_normals,
'al_unisel_n_selected_outliers':al_unisel_n_selected_outliers,
'al_unisel_n_selected_normals':al_unisel_n_selected_normals}
if sample_size > len(y):
overall_results.append(results)
continue
for seed in tqdm(seeds):
# shuffle data
np.random.seed(seed)
X = data['X']
y = data['y'].ravel()
X, X_test, y, y_test = train_test_split(X, y, test_size=0.10,
random_state=seed, shuffle=True, stratify=y)
# define models
rf = RandomForestClassifier(n_estimators=100, class_weight=None,
n_jobs=-1, random_state=seed)
####### Baseline #######
# only run baseline for first case of sample_size since it is
# independent of sample_size
if sample_size == sample_sizes[0]:
# train and predict
rf.fit(X, y)
y_pred = rf.predict(X_test)
score = confusion_matrix(y_test, y_pred).ravel().tolist()
baseline_cm.append(score)
####### Random Sampling #######
sample_idx = np.random.choice(X.shape[0], size=sample_size,
replace=False).tolist()
nonsample_idx = list(set(np.arange(X.shape[0])) -
set(sample_idx))
# record selected sample labels
rand_n_selected_normals.append(np.sum(y[sample_idx] == 0))
rand_n_selected_outliers.append(np.sum(y[sample_idx] == 1))
# train and predict
rf.fit(X[sample_idx], y[sample_idx])
y_pred = rf.predict(X_test)
score = confusion_matrix(y_test, y_pred).ravel().tolist()
random_cm.append(score)
####### Unisel Method #######
sample_idx, nonsample_idx = unisel(X, sample_size, seed)
# record selected sample labels
unisel_n_selected_normals.append(np.sum(y[sample_idx] == 0))
unisel_n_selected_outliers.append(np.sum(y[sample_idx] == 1))
# train and predict
rf.fit(X[sample_idx], y[sample_idx])
y_pred = rf.predict(X_test)
score = confusion_matrix(y_test, y_pred).ravel().tolist()
unisel_cm.append(score)
####### Active Learning Method (random initialization) #######
first_half_sample_size = int(np.floor(sample_size/2))
second_half_sample_size = int(np.ceil(sample_size/2))
sample_idx = np.random.choice(X.shape[0], size=first_half_sample_size,
replace=False).tolist()
sample_idx, nonsample_idx = active_learning(rf, X, y, sample_idx,
n_iter=second_half_sample_size)
# record selected sample labels
al_n_selected_normals.append(np.sum(y[sample_idx] == 0))
al_n_selected_outliers.append(np.sum(y[sample_idx] == 1))
# train and predict
rf.fit(X[sample_idx], y[sample_idx])
y_pred = rf.predict(X_test)
score = confusion_matrix(y_test, y_pred).ravel().tolist()
al_cm.append(score)
####### Active Learning Method (unisel initialization) #######
sample_idx, _ = unisel(X, first_half_sample_size, seed)
sample_idx, nonsample_idx = active_learning(rf, X, y, sample_idx,
n_iter=second_half_sample_size)
# record selected sample labels
al_unisel_n_selected_normals.append(np.sum(y[sample_idx] == 0))
al_unisel_n_selected_outliers.append(np.sum(y[sample_idx] == 1))
# train and predict
rf.fit(X[sample_idx], y[sample_idx])
y_pred = rf.predict(X_test)
score = confusion_matrix(y_test, y_pred).ravel().tolist()
al_unisel_cm.append(score)
results['baseline_cm'] = baseline_cm
results['rand_cm'] = random_cm
results['unisel_cm'] = unisel_cm
results['al_cm'] = al_cm
results['al_unisel_cm'] = al_unisel_cm
results['rand_n_selected_normals'] = rand_n_selected_normals
results['rand_n_selected_outliers'] = rand_n_selected_outliers
results['unisel_n_selected_normals'] = unisel_n_selected_normals
results['unisel_n_selected_outliers'] = unisel_n_selected_outliers
results['al_n_selected_normals'] = al_n_selected_normals
results['al_n_selected_outliers'] = al_n_selected_outliers
results['al_unisel_n_selected_normals'] = al_unisel_n_selected_normals
results['al_unisel_n_selected_outliers'] = al_unisel_n_selected_outliers
overall_results.append(results)
output_file = results['name'] + '_' + time.strftime("%Y%m%d-%H%M%S") + '.csv'
output_file = os.path.join('output', output_file)
pd.DataFrame(overall_results).to_csv(output_file, index=False)