Trivial anomalies are NOT detected

[AmitaiPerlstein]

x = 1:5000 x[4900:4910] = 3000 AnomalyDetectionVec(x, period=1440, direction = 'both', e_value = T, plot = T)

I get the following disappointing result: $anoms data frame with 0 columns and 0 rows

[AmitaiPerlstein]

Even a single straight forward deviation from a completely flat signal creates unexpected made up anomalies:

y = rep(0, 5000) y[4900]=10 AnomalyDetectionVec(y, period=1440, direction = 'both', e_value = T, plot = T) $anoms index anoms expected_value 1 580 0 4 2 2020 0 4 3 3460 0 5 4 4900 10 5

[AmitaiPerlstein]

Most surprising: Even when the expected values (returned by the parameter e_value = T), $anoms returns "anomalies" whose expected values are _identical _to the originally supplied ones:

z = rep(0,3000) z[2900] = -5 AnomalyDetectionVec(z, period=1440, direction = 'both', e_value = T, plot = T) $anoms index anoms expected_value 1 1291 0 0 2 1292 0 0 3 1293 0 0 4 1294 0 0 ...

[JasonAizkalns]

@AmitaiPerlstein perhaps you can can contain each of the examples in one post and format the code as you do on your SO post? In addition, you may want to change your title to "Trivial anomalies are not being detected."

[samuel-liyi]

I think this is due to the test this alogrithm is using, it does not handle when the data has zero sigma: https://github.com/twitter/AnomalyDetection/blob/master/R/detect_anoms.R

# protect against constant time series
data_sigma <- func_sigma(data[[2L]])
if(data_sigma == 0) 
       break

maybe in this case ,there should be a more intuitive result instead of an empty data_frame