Model Serving & Monitoring on GCP—Machine Learning Platforms Engineering

This article discusses the two critical components of an ML model's lifecycle: model serve and monitor.

The article covers the flow of concepts from introduction, tooling, GCP service, and how to use it at scale in an ML system engineering article. It explains the differences between offline model serve/batch prediction and prediction service, synonymous terms often used interchangeably in the realm of ML. It also covers the concept of datum/feature drift, and methods to identify and deal with it.

The article also describes the concept of model drift, model biases, and datum schema changes. It also lists the tools and infrastructure needed to support an ML platform.