Create a MetricBase model trigger
Model triggers execute when the time-series data deviates from expected values.
Before you begin
Role required: admin
About this task
There are four model types supported: autoregressive integrated moving average (ARIMA), probabilistic exponentially weighted moving average (PEWMA), Holt-Winters, and seasonal trend decomposition using Loess (STL). MetricBase uses the model to create a definition of expected, normal values.
Procedure
Navigate to All > MetricBase > MetricBase Triggers > Trigger Definitions.
Select New.
On the MetricBase Trigger Creation form, select the Model Trigger option.
On the form, fill in the fields.
The trigger levels are not initially visible on this form. You configure the levels later.
| Field | Description |
|---|---|
| Name | Name the model trigger. |
| Model | Trained model this trigger uses. See Create and train a predictive model. |
| Metric | Table metric that you want to monitor. This value is set automatically based on the model, which means you cannot set it yourself. |
| Table name | Table in the MetricBase database that contains the metric that you want to monitor. This value is set automatically based on the model, which means you cannot set it yourself. |
| Active | Option to activate the trigger. |
| Description | Description of the trigger. |
| Spike Direction | Specify whether you want to trigger only on anomalies that occur above predicted values \(upward\), below predicted values \(downward\), or both \(either\). |
Select Save.
A new record is created.
In the MetricBase Model Trigger Levels area on the record you just created, 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 2 standard deviations from the mean. Level 2 might be 4 standard deviations. Each level should trigger a different Workflow Studio flow. Level is often used in Condition Scripts. See Execute triggers conditionally. 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. Add rows to the table to create multi-layered triggering behavior.
Typically, each additional row (level) indicates a more severe condition. The Workflow Studio flow associated with the level should warn with increasing severity.
Select Update.
Add a triggering condition that determines whether a trigger executes a Workflow Studio flow.
Test your model and trigger on real data before deploying to production.
Model Trigger form
Model Trigger form
What to do next
Associate this trigger with a Workflow Studio flow. For more information, see Assign a trigger to a flow.