def mergeAnomaly(self, anoms, mergeLen=10):
accum_num = 0
find = 0
ret = []
l = len(anoms)
if l == 0:
return []
timediff = pd.Timedelta('15s') * 30
base = 10**9
for i, x in enumerate(anoms.index):
if x - anoms.index[i - 1] > timediff and i > 0:
ret.append(
Anomaly(anoms.index[find].value / base,
anoms.index[i - 1].value / base, 0, 0))
find = i
print find
ret.append(
Anomaly(anoms.index[find].value / base,
anoms.index[l - 1].value / base, 0, 0))
return ret
def mergeAnomaly(self, mergeLen=3):
accum_num = 0
find = -1
ret = []
l = len(self.anoms)
for i, x in enumerate(anoms):
if x and find == -1:
find = i
if not x or i == l - 1:
if i == l - 1:
i += 1
if i - find >= mergeLen and find > -1:
print i, find
ret.append(Anomaly(find, i, 0, 0))
find = -1
return ret
def _detect_anomalies(self):
"""
Detect anomalies using a threshold on anomaly scores.
"""
anom_scores = self.anom_scores
max_anom_score = anom_scores.max()
anomalies = []
if max_anom_score:
threshold = self.threshold or max_anom_score * self.score_percent_threshold
# Find all the anomaly intervals.
intervals = []
start, end = None, None
for timestamp, value in anom_scores.iteritems():
if value > threshold:
end = timestamp
if not start:
start = timestamp
elif start and end is not None:
intervals.append([start, end])
start = None
end = None
if start is not None:
intervals.append([start, end])
# Locate the exact anomaly point within each anomaly interval.
for interval_start, interval_end in intervals:
interval_series = anom_scores.crop(interval_start,
interval_end)
self.refine_algorithm_params['time_series'] = interval_series
refine_algorithm = self.refine_algorithm(
**self.refine_algorithm_params)
scores = refine_algorithm.run()
max_refine_score = scores.max()
# Get the timestamp of the maximal score.
max_refine_timestamp = scores.timestamps[scores.values.index(
max_refine_score)]
anomaly = Anomaly(interval_start, interval_end,
interval_series.max(), max_refine_timestamp)
anomalies.append(anomaly)
self.anomalies = anomalies