Skip to content

Live Monitoring for Anomalies

Live Monitoring (via UI)

Start Monitoring

Using the UI, from the All Signals view in the Insights tool, users can selectively start live anomaly monitoring using the latest available Insights model. Once monitoring begins, assessment outputs are written to the common model M[0]. Each monitored signal will have a corresponding output signal named <signal name>/anomaly score. Output will no longer be written to the specific model path <signal name>/M[#]/anomaly score.

Stop Monitoring

To update an out-of-date anomaly detection model, select the signal and choose the Stop Monitoring action. Once monitoring is stopped, no new assessment output will be generated, neither on the common model M[0] nor on the previously active model.

Resume Monitoring

If a new version of the anomaly detection model for a signal is available, it will not go live until the user manually restarts live monitoring using the following two-step process:

  1. Select the signal with the updated (and likely improved) anomaly model
  2. Choose the Start Monitoring action (after ensuring that monitoring of the previous model has been stopped)

Once restarted, the system will begin recording assessment output to the common model path M[0], accessible at <signal name>/anomaly score.

Making the Model Live (via API)

To apply the Insights anomaly detection continuously on incoming real-time data, user should make the correct model live.

Start Monitoring

  • Use the STARTLIVEMODEL API flow to deploy your Insights model in the live monitoring. This will enable it to continuously evaluate incoming data and generate anomaly scores in near real-time. The anomaly score (assessment output) will be recorded on the common model M[0] accessible at <signal name>/anomaly score.
  • Once live, the Insights Dashboard will begin updating to display the anomaly heatmaps for your streamed data, and you will see the live model count update in the Insights tab.

Stop Monitoring

  • Use the STOPLIVEMODEL API flow to undeploy your Insights model and stop the live monitoring. This will stop any further evaluation of the incoming data. The anomaly score (assessment output) will no longer be recorded on the common model M[0].

Resume Monitoring

Use the STARTLIVEMODEL API on the appropriate Model.