def test_estimate_anomaly():
f1 = estimate_anomaly(dummy_features,
{'spike': dummy_params['spike']})
f2 = estimate_anomaly(pd.DataFrame(dummy_features),
{'spike': dummy_params['spike']})
assert ma.allclose(f1, f2)
assert ma.allclose(f1,
ma.masked_values([-999, 0.0, -5.797359001920061,
-57.564627324851145, -999, -9.626760611162082], -999))
def test_estimate_anomaly():
f1 = estimate_anomaly(dummy_features, {'spike': dummy_params['spike']})
f2 = estimate_anomaly(pd.DataFrame(dummy_features),
{'spike': dummy_params['spike']})
assert ma.allclose(f1, f2)
assert ma.allclose(
f1,
ma.masked_values([
-999, 0.0, -5.797359001920061, -57.564627324851145, -999,
-9.626760611162082
], -999))
def test_estimate_anomaly_dict():
features = {'spike': [0, 1, -.5], 'tukey53H_norm': [-1, 0, 2]}
output = estimate_anomaly(features, dummy_params)
assert len(output) == len(features[features.keys()[0]])
assert type(output) == ma.MaskedArray
assert ~ma.getmaskarray(output).any()
def test_estimate_anomaly_dict():
features = {'spike': [0, 1, -.5],
'tukey53H_norm': [-1, 0, 2]
}
output = estimate_anomaly(features, dummy_params)
assert len(output) == len(features[features.keys()[0]])
assert type(output) == ma.MaskedArray
assert ~ma.getmaskarray(output).any()
def test_estimate_anomaly_maskedarray():
features = {'spike': ma.MaskedArray([0, 1, -.5], mask=[False, True, False]),
'tukey53H_norm': ma.MaskedArray([-1, 0, 2], mask=[False, True, False]),
}
output = estimate_anomaly(features, dummy_params)
assert len(output) == len(features[features.keys()[0]])
assert type(output) == ma.MaskedArray
# If all features for one measurement are masked, the estimated probability
# will be masked, like in the second position of this example
assert (output.mask == [False, True, False]).all()
features['spike'].mask[1] = False
output = estimate_anomaly(features, dummy_params)
assert len(output) == len(features[features.keys()[0]])
assert type(output) == ma.MaskedArray
# But if is there at least one of features valid, it returns a valid value
# Compare this example with the previous where spike[1] is not masked
# anymore.
assert ~ma.getmaskarray(output).any()
def test_estimate_anomaly_maskedarray():
features = {
'spike': ma.MaskedArray([0, 1, -.5], mask=[False, True, False]),
'tukey53H_norm': ma.MaskedArray([-1, 0, 2], mask=[False, True, False]),
}
output = estimate_anomaly(features, dummy_params)
assert len(output) == len(features[features.keys()[0]])
assert type(output) == ma.MaskedArray
# If all features for one measurement are masked, the estimated probability
# will be masked, like in the second position of this example
assert (output.mask == [False, True, False]).all()
features['spike'].mask[1] = False
output = estimate_anomaly(features, dummy_params)
assert len(output) == len(features[features.keys()[0]])
assert type(output) == ma.MaskedArray
# But if is there at least one of features valid, it returns a valid value
# Compare this example with the previous where spike[1] is not masked
# anymore.
assert ~ma.getmaskarray(output).any()
def anomaly_detection(features, cfg):
"""
Must decide where to set the flags.
"""
prob = estimate_anomaly(features, params = cfg['features'])
#flag = np.zeros(self.input[v].shape, dtype='i1')
flag = np.zeros(features[features.keys()[0]].shape, dtype='i1')
flag[np.nonzero(prob >= cfg['threshold'])] = 1
flag[np.nonzero(prob < cfg['threshold'])] = 4
return flag
def anomaly_detection(features, cfg):
"""
Must decide where to set the flags.
"""
prob = estimate_anomaly(features, params=cfg['features'])
#flag = np.zeros(self.input[v].shape, dtype='i1')
flag = np.zeros(prob.shape, dtype='i1')
flag[np.nonzero(prob >= cfg['threshold'])] = 1
flag[np.nonzero(prob < cfg['threshold'])] = 4
return prob, flag