-
Notifications
You must be signed in to change notification settings - Fork 0
/
failure_detection.py
165 lines (115 loc) · 6.89 KB
/
failure_detection.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
import numpy as np
import pickle as pkl
from operator import itemgetter
from itertools import groupby
import pandas as pd
from scipy.stats import zscore
def extreme_anomaly(dist):
q25, q75 = np.quantile(dist, [0.25, 0.75])
return q75 + 3*(q75-q25)
def simple_lowpass_filter(arr, alpha):
y = arr[0]
filtered_arr = [y]
for elem in arr[1:]:
y = y + alpha * (elem - y)
filtered_arr.append(y)
return filtered_arr
def detect_failures(anom_indices):
failure_list = []
failure = set()
for i in range(len(anom_indices) - 1):
if anom_indices[i] == 1 and anom_indices[i + 1] == 1:
failure.add(i)
failure.add(i + 1)
elif len(failure) > 0:
failure_list.append(failure)
failure = set()
if len(failure) > 0:
failure_list.append(failure)
return failure_list
def failure_list_to_interval(cycle_dates, failures):
failure_intervals = []
for failure in failures:
failure = sorted(failure)
failure_intervals.append(pd.Interval(cycle_dates[failure[0]][0], cycle_dates[failure[-1]][1], closed="both"))
return failure_intervals
def collate_intervals(interval_list):
diff_consecutive_intervals = [(interval_list[i+1].left - interval_list[i].right).days for i in range(len(interval_list)-1)]
lt_1day = np.where(np.array(diff_consecutive_intervals) <= 1)[0]
collated_intervals = []
for k, g in groupby(enumerate(lt_1day), lambda ix: ix[0]-ix[1]):
collated = list(map(itemgetter(1), g))
collated_intervals.append(pd.Interval(interval_list[collated[0]].left, interval_list[collated[-1]+1].right, closed="both"))
collated_intervals.extend([interval_list[i] for i in range(len(interval_list)) if i not in lt_1day and i-1 not in lt_1day])
return sorted(collated_intervals)
def print_failures(cycle_dates, output):
failures = detect_failures(output)
failure_intervals = failure_list_to_interval(cycle_dates, failures)
collated_intervals = collate_intervals(failure_intervals)
for interval in collated_intervals:
print(interval)
##### Results from the main paper #####
def generate_intervals(granularity, start_timestamp, end_timestamp):
current_timestamp = start_timestamp
interval_length = pd.offsets.DateOffset(**granularity)
interval_list = []
while current_timestamp < end_timestamp:
interval_list.append(pd.Interval(current_timestamp, current_timestamp + interval_length, closed="left"))
current_timestamp = current_timestamp + interval_length
return interval_list
with open("data/training_chunk_dates.pkl", "rb") as chunk_dates_file:
training_chunk_dates = pkl.load(chunk_dates_file)
with open("data/test_chunk_dates.pkl", "rb") as chunk_dates_file:
test_chunk_dates = pkl.load(chunk_dates_file)
train_intervals = generate_intervals({"minutes": 5}, pd.Timestamp(training_chunk_dates[0][0]), pd.Timestamp(training_chunk_dates[-1][0]))
test_intervals = generate_intervals({"minutes": 5}, pd.Timestamp(test_chunk_dates[0][0]), pd.Timestamp(test_chunk_dates[-1][0]))
def map_cycles_to_intervals(interval_list, chunk_dates):
cycles_dates = list(map(lambda x: pd.Interval(pd.Timestamp(x[0]), pd.Timestamp(x[1]), closed="both"), chunk_dates))
return list(map(lambda x: np.where([x.overlaps(i) for i in cycles_dates])[0], interval_list))
train_chunks_to_intervals = map_cycles_to_intervals(train_intervals, training_chunk_dates)
test_chunks_to_intervals = map_cycles_to_intervals(test_intervals, test_chunk_dates)
alpha = 0.05
with open("results/final_chunks_complete_losses_WAE_LSTMDiscriminator_TCN_analog_feats_4_10_30_3_10.0_3_32_150_0.001_0.001_64.pkl", "rb") as loss_file:
tl = pkl.load(loss_file)
test_losses = tl["test"]
train_losses = tl["train"]
median_train_losses = np.array([np.median(np.array(train_losses["reconstruction"])[tc]) for tc in train_chunks_to_intervals if len(tc) > 0])
median_test_losses = np.array([np.median(np.array(test_losses["reconstruction"])[tc]) for tc in test_chunks_to_intervals if len(tc) > 0])
median_train_critic = np.array([np.median(np.array(train_losses["critic"])[tc]) for tc in train_chunks_to_intervals if len(tc) > 0])
median_test_critic = np.array([np.median(np.array(test_losses["critic"])[tc]) for tc in test_chunks_to_intervals if len(tc) > 0])
combine_critic_reconstruction = np.abs(list(map(lambda x: np.nan_to_num(zscore(median_test_critic[:x], ddof=1)[-1]),
range(1,len(median_test_critic)+1)))) * median_test_losses
combine_critic_reconstruction_train = np.abs(list(map(lambda x: np.nan_to_num(zscore(median_train_critic[:x], ddof=1)[-1]),
range(1,len(median_train_critic)+1)))) * median_train_losses
anom = extreme_anomaly(combine_critic_reconstruction_train)
binary_output = np.array(combine_critic_reconstruction > anom, dtype=int)
wae_gan_output = np.array(simple_lowpass_filter(binary_output,alpha))
print_failures(test_intervals, wae_gan_output)
with open("results/final_chunks_complete_losses_AE_tcn_ae_analog_feats_4_8_6_7_100_0.001_64.pkl", "rb") as loss_file:
tl = pkl.load(loss_file)
test_losses = tl["test"]
train_losses = tl["train"]
median_train_losses = np.array(
[np.median(np.array(train_losses)[tc]) for tc in train_chunks_to_intervals if len(tc) > 0])
median_test_losses = np.array([np.median(np.array(test_losses)[tc]) for tc in test_chunks_to_intervals if len(tc) > 0])
date_output_test = [interval.left for i, interval in enumerate(test_intervals) if len(test_chunks_to_intervals[i]) > 0]
date_output_train = [interval.left for i, interval in enumerate(train_intervals) if
len(train_chunks_to_intervals[i]) > 0]
anomaly_threshold = extreme_anomaly(median_train_losses)
binary_output = np.array(np.array(median_test_losses) > anomaly_threshold, dtype=int)
tcn_output = np.array(simple_lowpass_filter(binary_output, 0.03))
print_failures(test_intervals, tcn_output)
with open("results/final_chunks_complete_losses_AE_lstm_ae_analog_feats_4_5_150_0.001_64.pkl", "rb") as loss_file:
tl = pkl.load(loss_file)
test_losses = tl["test"]
train_losses = tl["train"]
median_train_losses = np.array(
[np.median(np.array(train_losses)[tc]) for tc in train_chunks_to_intervals if len(tc) > 0])
median_test_losses = np.array([np.median(np.array(test_losses)[tc]) for tc in test_chunks_to_intervals if len(tc) > 0])
date_output_test = [interval.left for i, interval in enumerate(test_intervals) if len(test_chunks_to_intervals[i]) > 0]
date_output_train = [interval.left for i, interval in enumerate(train_intervals) if
len(train_chunks_to_intervals[i]) > 0]
anomaly_threshold = extreme_anomaly(median_train_losses)
binary_output = np.array(np.array(median_test_losses) > anomaly_threshold, dtype=int)
lstm_output = np.array(simple_lowpass_filter(binary_output, 0.05))
print_failures(test_intervals, lstm_output)