Vertex AI Feature Store

We’ve done a lot of feature engineering work in this chapter. Bear in mind that we performed data transformations and engineered new features because we had reason to believe that the raw data was insufficient to train a machine learning model to suit our business case. This means that the raw data our model will see in the real world would usually not contain the enhancements we performed on the data during training. After all of that work, we would generally want to save the updated features we’ve engineered so that our model can reference them when it needs to make predictions. Vertex AI Feature Store was created for this purpose. We briefly mentioned Vertex AI Feature Store in Chapter 3, and in this section, we will dive into more detail regarding what it is and how we can use it to store and serve features for both training and inference.

Introduction to Vertex AI Feature Store

Here’s the official definition from the Google Cloud...