def testFlatMetricScores(self):
"""
This calls estimateAnomalyLikelihoods with flat metric values. In this case
we should use the null distribution, which gets reasonably high likelihood
for everything.
"""
# Generate samples with very flat metric values
data1 = _generateSampleData(metricMean=42.0, metricVariance=1e-10)[0:1000]
likelihoods, _, estimatorParams = an.estimateAnomalyLikelihoods(data1)
# Check that we do indeed get reasonable likelihood values
self.assertEqual(len(likelihoods), len(data1))
self.assertTrue(likelihoods.sum() >= 0.4 * len(likelihoods))
# Check that we do indeed get null distribution
self.assertDictEqual(estimatorParams["distribution"], an.nullDistribution())
def testFlatMetricScores(self):
"""
This calls estimateAnomalyLikelihoods with flat metric values. In this case
we should use the null distribution, which gets reasonably high likelihood
for everything.
"""
# Generate samples with very flat metric values
data1 = _generateSampleData(metricMean=42.0,
metricVariance=1e-10)[0:1000]
likelihoods, _, estimatorParams = (
an.estimateAnomalyLikelihoods(data1))
# Check that we do indeed get reasonable likelihood values
self.assertEqual(len(likelihoods), len(data1))
self.assertTrue(likelihoods.sum() >= 0.4 * len(likelihoods))
# Check that we do indeed get null distribution
self.assertDictEqual(estimatorParams["distribution"],
an.nullDistribution())