Designing solutions for near real-time ML predictions

Sometimes machine learning applications demand high-throughput updates to features and near real-time access to the updated features. Timely access to fast-changing features is critical for the accuracy of predictions made by these applications. As an example, consider a machine learning application in a call center that predicts how to route the incoming customer calls to available agents. This application needs to have knowledge of the customer's latest web session clicks to make accurate routing decisions. If you capture a customer's web-click behavior as features, the features need to be updated instantly and the application needs access to the updated features in near-real time. Similarly, for weather prediction problems, you may want to capture the weather measurement features frequently for accurate weather predictions and need the ability to look up features in real time.

Let's look at some best practices...