Which concept refers to tracking input features used by a model?

A feature store is a centralized repository that allows data scientists to share and reuse features. A key function of a feature store is to manage the entire lifecycle of a feature, which includes tracking its definition, lineage, and transformations from the raw input data. This capability ensures consistency, discoverability, and governance of features used across different models, directly addressing the need to track the input features used by a model.

B. Data labeling: This is the process of adding tags or annotations to raw data for supervised learning, not tracking features.

C. Pipeline trigger: This is a mechanism for automating the execution of a machine learning pipeline, not for feature management.

D. Batch scoring: This is a method for generating predictions on a set of observations at once, not for tracking feature inputs.

- Microsoft. (2023). What is a feature store in Azure Machine Learning? Azure Machine Learning Documentation. Retrieved from https://learn.microsoft.com/en-us/azure/machine-learning/concept-feature-store. The document states, "A feature store is a centralized repository of features that enables data scientists to find and reuse features." It also covers feature discovery and management.

- Nargundkar, S., & Malhorta, R. (2022). Engineering MLOps. O'Reilly Media. Chapter 4, "Feature Stores," discusses how feature stores are used to "track feature lineage" and "manage feature definitions."