Monitoring

Understand the concepts

Let's quickly go over some key concepts. For further information, go to the Superwise glossary overview.

  • Project - A group of models that are of interest to a specific group of stakeholders in the company.
  • Models - The algorithms that are providing predictions and being monitored using Superwise.
  • Entities - All the data entities with which the model interacts, including features, predictions, labels, etc.
  • Metrics - All the calculations and statistics that are being applied on the models' entities.
  • Segments - Sub-populations within the data (based on feature filtering).
  • Policy - An automatic monitoring mechanism that continuously checks a specific metric to see that it's behaving as expected and creates an alert once an anomaly is detected.
  • Control limits - The boundaries within which the metric values are considered to be normal behavior.
  • Sensitivity - How sensitive the thresholds are when declaring an anomaly.
  • Normal period - How far back in time Superwise should observe the data to determine what normal data looks like. For example, when looking at vaccination rates, data earlier than January 2019 isn't relevant anymore due to COVID implications.
  • Incident length - How long an anomaly should last for an alert to be sent out.
  • Incident - A group of anomalies that are thought to be symptoms of the same problem.

Policies are defined at the project/model/segment level and can be created/edited/consumed from the Policies screen.

Incidents are consumed and managed from the Incidents screen.