def handleRecord(self, inputData):
"""Returns a tuple (anomalyScore).
The anomalyScore is the tail probability of the gaussian (normal) distribution
over a sliding window of inputData values. The tail probability is based on the
Q-function. The windowSize has been tuned to give best performance on NAB.
"""
anomalyScore = 0.0
inputValue = inputData["value"]
if len(self.windowData) > 0:
anomalyScore = 1 - anomaly_likelihood.normalProbability(
inputValue, {
"mean": self.mean,
"stdev": self.std
})
if len(self.windowData) < self.windowSize:
self.windowData.append(inputValue)
self._updateWindow()
else:
self.stepBuffer.append(inputValue)
if len(self.stepBuffer) == self.stepSize:
# slide window forward by stepSize
self.windowData = self.windowData[self.stepSize:]
self.windowData.extend(self.stepBuffer)
# reset stepBuffer
self.stepBuffer = []
self._updateWindow()
return (anomalyScore, )
def handleRecord(self, inputData):
"""Returns a tuple (anomalyScore).
The anomalyScore is the tail probability of the gaussian (normal) distribution
over a sliding window of inputData values. The tail probability is based on the
Q-function. The windowSize has been tuned to give best performance on NAB.
"""
anomalyScore = 0.0
inputValue = inputData["value"]
if len(self.windowData) > 0:
anomalyScore = 1 - anomaly_likelihood.normalProbability(
inputValue, {"mean": self.mean,"stdev": self.std})
if len(self.windowData) < self.windowSize:
self.windowData.append(inputValue)
self._updateWindow()
else:
self.stepBuffer.append(inputValue)
if len(self.stepBuffer) == self.stepSize:
# slide window forward by stepSize
self.windowData = self.windowData[self.stepSize:]
self.windowData.extend(self.stepBuffer)
# reset stepBuffer
self.stepBuffer = []
self._updateWindow()
return (anomalyScore, )
def testNormalProbability(self):
"""
Test that the normalProbability function returns correct normal values
"""
# Test a standard normal distribution
# Values taken from http://en.wikipedia.org/wiki/Standard_normal_table
p = {"name": "normal", "mean": 0.0, "variance": 1.0, "stdev": 1.0}
self.assertWithinEpsilon(an.normalProbability(0.0, p), 0.5)
self.assertWithinEpsilon(an.normalProbability(0.3, p), 0.3820885780)
self.assertWithinEpsilon(an.normalProbability(1.0, p), 0.1587)
self.assertWithinEpsilon(1.0 - an.normalProbability(1.0, p),
an.normalProbability(-1.0, p))
self.assertWithinEpsilon(an.normalProbability(-0.3, p),
1.0 - an.normalProbability(0.3, p))
# Non standard normal distribution
p = {"name": "normal", "mean": 1.0, "variance": 4.0, "stdev": 2.0}
self.assertWithinEpsilon(an.normalProbability(1.0, p), 0.5)
self.assertWithinEpsilon(an.normalProbability(2.0, p), 0.3085)
self.assertWithinEpsilon(an.normalProbability(3.0, p), 0.1587)
self.assertWithinEpsilon(an.normalProbability(3.0, p),
1.0 - an.normalProbability(-1.0, p))
self.assertWithinEpsilon(an.normalProbability(0.0, p),
1.0 - an.normalProbability(2.0, p))
# Non standard normal distribution
p = {
"name": "normal",
"mean": -2.0,
"variance": 0.5,
"stdev": math.sqrt(0.5)
}
self.assertWithinEpsilon(an.normalProbability(-2.0, p), 0.5)
self.assertWithinEpsilon(an.normalProbability(-1.5, p), 0.241963652)
self.assertWithinEpsilon(an.normalProbability(-2.5, p),
1.0 - an.normalProbability(-1.5, p))
def testNormalProbability(self):
"""
Test that the normalProbability function returns correct normal values
"""
# Test a standard normal distribution
# Values taken from http://en.wikipedia.org/wiki/Standard_normal_table
p = {"name": "normal", "mean": 0.0, "variance": 1.0, "stdev": 1.0}
self.assertWithinEpsilon(an.normalProbability(0.0, p), 0.5)
self.assertWithinEpsilon(an.normalProbability(0.3, p), 0.3820885780)
self.assertWithinEpsilon(an.normalProbability(1.0, p), 0.1587)
self.assertWithinEpsilon(1.0 - an.normalProbability(1.0, p),
an.normalProbability(-1.0, p))
self.assertWithinEpsilon(an.normalProbability(-0.3, p),
1.0 - an.normalProbability(0.3, p))
# Non standard normal distribution
p = {"name": "normal", "mean": 1.0, "variance": 4.0, "stdev": 2.0}
self.assertWithinEpsilon(an.normalProbability(1.0, p), 0.5)
self.assertWithinEpsilon(an.normalProbability(2.0, p), 0.3085)
self.assertWithinEpsilon(an.normalProbability(3.0, p), 0.1587)
self.assertWithinEpsilon(an.normalProbability(3.0, p),
1.0 - an.normalProbability(-1.0, p))
self.assertWithinEpsilon(an.normalProbability(0.0, p),
1.0 - an.normalProbability(2.0, p))
# Non standard normal distribution
p = {"name": "normal", "mean": -2.0, "variance": 0.5,
"stdev": math.sqrt(0.5)}
self.assertWithinEpsilon(an.normalProbability(-2.0, p), 0.5)
self.assertWithinEpsilon(an.normalProbability(-1.5, p), 0.241963652)
self.assertWithinEpsilon(an.normalProbability(-2.5, p),
1.0 - an.normalProbability(-1.5, p))