Feature Store MLOps Pipeline

[MacerPacer]

Hey,

i'm trying to understand the MLOps Pipeline with the CI/CD-Automation (Stage 2 Maturity Level) and struggle with the Feature Store as the component feeding the Automated Pipeline with data. What i found out in the internet was, that Feature Stores extract data from different sources, transform them and create training data which can be used to train the model (retraining with new data). But in the pipeline the steps like Data preperation and Data extraction come after the Feature Store.

Can somebody explain to me, whats the output of the Feature Store and how it is used to serve the data for the Automated Pipeline and the Prediction Service?

Thanks in advance

[Joevanie · Accepted solution]

The Feature Store is just a centralized repository of features. By that, its output is just a set of features typically used to train an ML model. Depending on your specific needs, you can serve the ingested data in the Feature Store to the model right away (in what is called feature serving) or export feature values and do further preparation of data.