def testCalcSweepScoreWindowScoreInteraction(self):
"""Scores inside a window should be positive; all others should be negative."""
numRows = 100
fakeAnomalyScores = [1 for _ in range(numRows)]
fakeTimestamps = [
i for i in range(numRows)
] # We'll use numbers, even though real data uses dates
fakeName = "TestDataSet"
windowA = (30, 39)
windowB = (75, 95)
windowLimits = [windowA, windowB]
expectedInWindowCount = (windowA[1] - windowA[0] +
1) + (windowB[1] - windowB[0] + 1)
# Standard profile
costMatrix = {
"tpWeight": 1.0,
"fnWeight": 1.0,
"fpWeight": 0.11,
}
probationPercent = 0.1
o = Sweeper(probationPercent=probationPercent, costMatrix=costMatrix)
scoredAnomalies = o.calcSweepScore(fakeTimestamps, fakeAnomalyScores,
windowLimits, fakeName)
# Check that correct number of AnomalyPoints returned
assert len(scoredAnomalies) == numRows
assert all(isinstance(x, AnomalyPoint) for x in scoredAnomalies)
# Expected number of points marked 'probationary'
probationary = [
x for x in scoredAnomalies if x.windowName == "probationary"
]
assert len(probationary) == o._getProbationaryLength(numRows)
# Expected number of points marked 'in window'
inWindow = [
x for x in scoredAnomalies
if x.windowName not in ("probationary", None)
]
assert len(inWindow) == expectedInWindowCount
# Points in window have positive score; others have negative score
for point in scoredAnomalies:
if point.windowName not in ("probationary", None):
assert point.sweepScore > 0
else:
assert point.sweepScore < 0
def testCalcSweepScoreWindowScoreInteraction(self):
"""Scores inside a window should be positive; all others should be negative."""
numRows = 100
fakeAnomalyScores = [1 for _ in range(numRows)]
fakeTimestamps = [i for i in range(numRows)] # We'll use numbers, even though real data uses dates
fakeName = "TestDataSet"
windowA = (30, 39)
windowB = (75, 95)
windowLimits = [windowA, windowB]
expectedInWindowCount = (windowA[1] - windowA[0] + 1) + (windowB[1] - windowB[0] + 1)
# Standard profile
costMatrix = {
"tpWeight": 1.0,
"fnWeight": 1.0,
"fpWeight": 0.11,
}
probationPercent = 0.1
o = Sweeper(probationPercent=probationPercent, costMatrix=costMatrix)
scoredAnomalies = o.calcSweepScore(fakeTimestamps, fakeAnomalyScores, windowLimits, fakeName)
# Check that correct number of AnomalyPoints returned
assert len(scoredAnomalies) == numRows
assert all(isinstance(x, AnomalyPoint) for x in scoredAnomalies)
# Expected number of points marked 'probationary'
probationary = [x for x in scoredAnomalies if x.windowName == "probationary"]
assert len(probationary) == o._getProbationaryLength(numRows)
# Expected number of points marked 'in window'
inWindow = [x for x in scoredAnomalies if x.windowName not in ("probationary", None)]
assert len(inWindow) == expectedInWindowCount
# Points in window have positive score; others have negative score
for point in scoredAnomalies:
if point.windowName not in ("probationary", None):
assert point.sweepScore > 0
else:
assert point.sweepScore < 0
def optimizeThreshold(args):
"""Optimize the threshold for a given combination of detector and profile.
@param args (tuple) Contains:
detectorName (string) Name of detector.
costMatrix (dict) Cost matrix to weight the
true positives, false negatives,
and false positives during
scoring.
resultsCorpus (nab.Corpus) Corpus object that holds the per
record anomaly scores for a
given detector.
corpusLabel (nab.CorpusLabel) Ground truth anomaly labels for
the NAB corpus.
probationaryPercent (float) Percent of each data file not
to be considered during scoring.
@return (dict) Contains:
"threshold" (float) Threshold that returns the largest score from the
Objective function.
"score" (float) The score from the objective function given the
threshold.
"""
(detectorName, costMatrix, resultsCorpus, corpusLabel,
probationaryPercent) = args
sweeper = Sweeper(probationPercent=probationaryPercent,
costMatrix=costMatrix)
# First, get the sweep-scores for each row in each data set
allAnomalyRows = []
for relativePath, dataSet in resultsCorpus.dataFiles.iteritems():
if "_scores.csv" in relativePath:
continue
# relativePath: raw dataset file,
# e.g. 'artificialNoAnomaly/art_noisy.csv'
relativePath = convertResultsPathToDataPath(
os.path.join(detectorName, relativePath))
windows = corpusLabel.windows[relativePath]
labels = corpusLabel.labels[relativePath]
timestamps = labels['timestamp']
anomalyScores = dataSet.data["anomaly_score"]
curAnomalyRows = sweeper.calcSweepScore(timestamps, anomalyScores,
windows, relativePath)
allAnomalyRows.extend(curAnomalyRows)
# Get scores by threshold for the entire corpus
scoresByThreshold = sweeper.calcScoreByThreshold(allAnomalyRows)
scoresByThreshold = sorted(scoresByThreshold,
key=lambda x: x.score,
reverse=True)
bestParams = scoresByThreshold[0]
print(
"Optimizer found a max score of {} with anomaly threshold {}.".format(
bestParams.score, bestParams.threshold))
return {"threshold": bestParams.threshold, "score": bestParams.score}
def optimizeThreshold(args):
"""Optimize the threshold for a given combination of detector and profile.
@param args (tuple) Contains:
detectorName (string) Name of detector.
costMatrix (dict) Cost matrix to weight the
true positives, false negatives,
and false positives during
scoring.
resultsCorpus (nab.Corpus) Corpus object that holds the per
record anomaly scores for a
given detector.
corpusLabel (nab.CorpusLabel) Ground truth anomaly labels for
the NAB corpus.
probationaryPercent (float) Percent of each data file not
to be considered during scoring.
@return (dict) Contains:
"threshold" (float) Threshold that returns the largest score from the
Objective function.
"score" (float) The score from the objective function given the
threshold.
"""
(detectorName,
costMatrix,
resultsCorpus,
corpusLabel,
probationaryPercent) = args
sweeper = Sweeper(
probationPercent=probationaryPercent,
costMatrix=costMatrix
)
# First, get the sweep-scores for each row in each data set
allAnomalyRows = []
for relativePath, dataSet in resultsCorpus.dataFiles.iteritems():
if "_scores.csv" in relativePath:
continue
# relativePath: raw dataset file,
# e.g. 'artificialNoAnomaly/art_noisy.csv'
relativePath = convertResultsPathToDataPath(
os.path.join(detectorName, relativePath))
windows = corpusLabel.windows[relativePath]
labels = corpusLabel.labels[relativePath]
timestamps = labels['timestamp']
anomalyScores = dataSet.data["anomaly_score"]
curAnomalyRows = sweeper.calcSweepScore(
timestamps,
anomalyScores,
windows,
relativePath
)
allAnomalyRows.extend(curAnomalyRows)
# Get scores by threshold for the entire corpus
scoresByThreshold = sweeper.calcScoreByThreshold(allAnomalyRows)
scoresByThreshold = sorted(
scoresByThreshold,key=lambda x: x.score, reverse=True)
bestParams = scoresByThreshold[0]
print("Optimizer found a max score of {} with anomaly threshold {}.".format(
bestParams.score, bestParams.threshold
))
return {
"threshold": bestParams.threshold,
"score": bestParams.score
}
