Model triggers execute when the time-series data deviates from expected values.

1. Click MetricBase > Flow Designer > Trigger Definitions and click New.
2. Click Model Trigger - Detects and takes action when a metric value deviates from the trained model.
   The MetricBase Model Trigger form specifies the table and metric used to train the model.

   

3. On the form, fill in the fields.

   Field	Description
   Name	Name the model trigger.
   Model	Trained model this trigger uses.
   Metric	Automatically filled in based on the model.
   Table name	Automatically filled in based on the model.
   Active	Option to activate the trigger.
   Description	Explanation of the trigger.
   Spike Direction	Trigger execution criteria. Specify whether you want to trigger only on anomalies that occur above predicted values, only on anomalies that occur below predicted values, or both.
   Window	Amount of data to use when calculating the moving, weighted average; the longer the window, the more data. The recommendation is 30 times the sampling interval. For example, if the time-series data is sampled every minute, make this value 30 minutes. Values must be multiples of the sampling interval.
   Active	Turns on the trigger.

   Note: For the Window value, very short durations make the weighted average move with the data itself and fail to detect anomalies. Very long durations make the weighted average unresponsive to current trends in the data and may cause false triggers.
4. Click Save.
5. In the MetricBase Model Trigger Levels area, double-click each cell to add values that specify trigger parameters.

   Column	Description
   Level	Numbers that indicate increasing severity. For example, you might define level 1 to be 3 standard deviations from the mean. Level 2 might be 5 standard deviations. Each level should trigger a different Flow Designer flow. Level is often used in Conditional Scripts.
   Function	Always Greater than or is, meaning that in order to trigger, the time-series metric must be greater than or equal to the number of standard deviations from the mean specified in Number of Standard Deviations.
   Number of Standard Deviation	Float value specifying the number of standard deviations the time-series metric must be away from the mean to trigger an alert at this level.

6. (Optional) Add additional rows in the table to create multi-layered triggering behavior.
   Each successive row should indicate a more severe condition.
7. Click Update.
   The MetricBase Trigger Definitions page displays the list of defined triggers.
8. Test your model and trigger on real data before deploying to production.
9. (Optional) Add a triggering condition that determines whether or not a trigger executes a Flow Designer flow.
   For more information, see Create a triggering condition.