Monitoring your model
In this last section, you will see how the platform automatically starts capturing the typical performance metrics of your model. The platform also helps you visualize the performance of the inference. The platform uses Seldon to package the model, and Seldon exposes default metrics to be captured. Seldon also allows you to write custom metrics for specific models; however, it is out of the scope of this book.
Let's start by understanding how the metrics capture and visualization work.
Understanding monitoring components
The way metrics capture works is that your model is wrapped by Seldon. Seldon then exposes the metrics to a well-defined URL endpoint, which was detailed in Chapter 7, Model Deployment and Automation. Prometheus harvests this information and stores it in its database. The platform's Grafana connects to Prometheus and helps you visualize the recorded metrics.
Figure 10.47 summarizes the relationship between the model and monitoring...
