-
Notifications
You must be signed in to change notification settings - Fork 8
Error Handling
Quantile Method
Based on the parameter alpha
[default: 0.01], the anomaly flag gets set according
error < quantile(alpha/2) => flag = -1
error > quantile(1 - alpha/2) => flag = 1
where alpha/2
defines the upper and lower cutoff value (e.g. alpha=0.02
leads to anomalies if the error is above the 99% and below the 1% quantile). Smaller values of alpha
lead to a less sensitive alerting.
Tukeys Method
Similar to the Plugin itself, the 25th and 75th quantiles of the error distribution are used to calculate the Inter Quartile Range iqr = quantile_75 - quantile_25
and the flag gets set according
error < quantile_25 - scaling * iqr => flag = -1
error > quantile_75 + scaling * iqr => flag = 1
where scaling
[default: 1.5] is an optional parameter. Bigger values of scaling
lead to a less sensitive alerting.
Additional thresholds
For both methods minimum deviation thresholds can be set (see configuration example of the plugin). If those thresholds are active, the error flag is only set, if error not only fulfills the above requirements but also is above the minimum_upper_threshold
or below the minimum_lower_threshold
respectively.