Vertex AI Prediction and Model Monitoring Guide
Training a model is only one part of machine learning. After a model is trained and validated, it must serve predictions in a real workflow. Then the team must monitor whether the model continues to behave well after users, systems, and data start changing. This guide explains batch prediction, online prediction, serving containers, and model monitoring in simple production-minded terms. Quick Answer Use batch prediction when you need many predictions at once and do not need an instant response. Use online prediction when an application needs a fast response from a deployed endpoint. Use model monitoring to detect training-serving skew, feature drift, data quality changes, unusual prediction patterns, and behavior that may reduce model trust. ...