Closed
Description
In order to give users more insight into anomaly detection, we could be creating an annotation every time there is a significant change in the model calculation. Some ideas on what to report are:
- step change detected
- periodicity detected
- trend detected
How this could work technically:
- CPP emits log messages with special marker string denoting log message related to a significant model change
- Java adds a hook to
CppLogMessageHandler
class so that every time such message is discovered, an annotation is created - Annotation is automatically visible in the UI, just like user-created annotations or other system-created annotations (e.g.: model snapshot annotation)
- When job is deleted, we don't need to delete the annotations as they provide history that the user may want to retain