-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathanalysis_correct.py
200 lines (138 loc) · 5.34 KB
/
analysis_correct.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
import pandas as pd
import numpy as np
from river import drift
from scipy import stats
def retrieve_data(dataset_name):
labels = np.load(f"datasets/labels_SEA_{dataset_name}.npy")
epsilon_data = pd.read_csv(f"results/SEA_{dataset_name}_correct.csv")
epsilons = epsilon_data["epsilon"].to_list()
print(len(epsilons[900:]))
return labels, epsilons[900:]
def update_det2(epsilon, idx, reference, detection, window_size, fp_count):
#first fill reference window
tp = None
if len(reference) != window_size:
reference.append(epsilon)
else:
if len(detection)==window_size:
res = stats.mannwhitneyu(reference, detection)
if res[1]< 0.005:
if idx>=5000:
tp = idx
return reference, detection, tp, fp_count
else:
fp_count+=1
detection.pop(0)
detection.append(epsilon)
else:
detection.append(epsilon)
return reference, detection, tp, fp_count
def detect_drift2(labels, epsilons, drift_index, window_size):
reference_true, reference_false = [], []
detection_true, detection_false = [], []
fp_count = 0
tp = None
class_label = None
for idx, (label, epsilon) in enumerate(zip(labels[1900:6000], epsilons)):
idx+=1900
#split based on class
if label==True:
reference_true, detection_true, tp, fp_count = update_det2(epsilon, idx, reference_true, detection_true, window_size, fp_count)
if tp != None:
class_label = True
else:
reference_false, detection_false, tp, fp_count = update_det2(epsilon, idx, reference_false, detection_false, window_size, fp_count)
if tp != None:
class_label = False
if tp != None:
print("drift detection index: ", tp)
print("FPS: ", fp_count)
print("Detected class: ", class_label)
break
print("No drift detected within 1000 instances")
print("FPS: ", fp_count)
def update_det(epsilon, idx, reference, detection, window_size, fp_count):
#first fill reference window
tp = None
if len(reference) != window_size:
reference.append(epsilon)
else:
if len(detection)==window_size:
res = stats.ks_2samp(reference, detection)
if res[1]< 0.005:
if idx>=5000:
tp = idx
return reference, detection, tp, fp_count
else:
fp_count+=1
detection.pop(0)
detection.append(epsilon)
else:
detection.append(epsilon)
return reference, detection, tp, fp_count
def detect_drift(labels, epsilons, drift_index, window_size):
reference_true, reference_false = [], []
detection_true, detection_false = [], []
fp_count = 0
tp = None
class_label = None
for idx, (label, epsilon) in enumerate(zip(labels[1900:6000], epsilons)):
idx+=1900
#split based on class
if label==True:
reference_true, detection_true, tp, fp_count = update_det(epsilon, idx, reference_true, detection_true, window_size, fp_count)
if tp != None:
class_label = True
else:
reference_false, detection_false, tp, fp_count = update_det(epsilon, idx, reference_false, detection_false, window_size, fp_count)
if tp != None:
class_label = False
if tp != None:
print("drift detection index: ", tp)
print("FPS: ", fp_count)
print("Detected class: ", class_label)
break
print("No drift detected within 1000 instances")
print("FPS: ", fp_count)
def detect_drift_3(labels, epsilons, drift_index, window_size):
fp_count = 0
tp = None
class_label = None
ddm_true = drift.binary.DDM()
ddm_false = drift.binary.DDM()
for idx, (label, epsilon) in enumerate(zip(labels[1900:6000], epsilons)):
idx+=1900
if label == True:
ddm_true.update(epsilon)
if ddm_true.drift_detected:
if idx >=drift_index:
tp = idx
class_label = True
print(tp, fp_count, class_label)
else:
fp_count+=1
else:
ddm_false.update(epsilon)
if ddm_false.drift_detected:
if idx>=drift_index:
tp = idx
class_label = False
print(tp, fp_count, class_label)
else:
fp_count+=1
print(tp, fp_count, class_label, "no drift")
def main():
#params
dataset_name = "2_3_5"
drift_index = 5000
window_size = 100
labels, epsilons = retrieve_data(dataset_name)
# detect_drift(labels, epsilons, drift_index, window_size)
# detect_drift2(labels, epsilons, drift_index, window_size)
# detect_drift_3(labels, epsilons, drift_index, window_size)
mtd = [40,53,109,101,81]
fac = [0,2,0,0,50]
print(np.mean(mtd), np.std(mtd))
print(np.mean(fac), np.std(fac))
if __name__ == '__main__':
main()