Related Anomalies

Mona now automatically detects correlated metrics with similar anomalies for generated insights to help with faster investigations of root causes.

Hi everyone, we are happy to announce our long awaited feature - “Related Anomalies”.
Besides proactively alerting on anomalous behavior, Mona now automatically detects related anomalies in correlated metrics for each generated insight and makes investigating the root cause much easier!

For example, let’s say you are tracking the average probability score for a detected category in a text category classifier. Let's assume that this metric is correlated with the length of the text. The shorter the text, usually the lower the probability score. So if Mona found a segment in which this metric’s average decreases, then if the average text length on this same segment decreased as well, it might be the cause for the decline in probability score and Mona will suggest that.

The related anomalies are found inside each insight page:

While this feature is activated automatically, you can easily tweak its behavior with the following params:
“find_related_anomalies_for” - lets you define which metrics you wish to get related anomalies for. If left empty, Mona will search for related anomalies in all metrics.
“avoid_related_anomalies_for” - lets you define which metrics you want to avoid searching for related anomalies for.
“related_anomalies_min_correlation” - lets you set a threshold for the related anomalies by the correlation between both metrics.

Note: this feature is currently only available for AverageDrift verse types, but will gradually be available for all verse types.