Table of Contents

Retraining the internal model used by RAD

In some cases, it can be useful to retrain the internal model used by relational anomaly detection (RAD). This allows you to indicate the periods during which a parameter group was behaving as expected, so that RAD can better identify when the parameters deviate from that expected behavior in the future.

  1. Connect to the DMA hosting the grouped parameters using the SLNetClientTest tool.

  2. Go to the Build Message tab of the main window of the SLNetCLientTest tool.

  3. In the Message Type drop-down list, select the message Skyline.DataMiner.Analytics.MAD.RetrainMADModelMessage.

    In this message, "MAD" stands for "multivariate anomaly detection", which is another name for RAD.

  4. Configure the following fields:

    • GroupName: The name of the parameter group as configured in RelationalAnomalyDetection.xml.
    • StartTime: The start time of the period during which the parameter group was behaving as expected.
    • EndTime: The end time of the period during which the parameter group was behaving as expected.
  5. Click Send Message.

Note
  • From DataMiner 10.5.4/10.6.0 onwards, other messages are also available that can be used to add a parameter group, retrieve a parameter group, or retrieve all configuration information for a particular group (AddMADParameterGroupMessage, RemoveMADParameterGroupMessage, and GetMADParameterGroupInfoMessage, respectively). While you can configure and view the same things directly in RelationalAnomalyDetection.xml, with these messages you can for example configure a low-code app to visualize and manage the parameter groups.
  • Keep in mind that the group names need to be unique. Prior to DataMiner 10.5.4/10.6.0, casing is taken into account for this, but this no longer matters in later DataMiner versions.