A team is getting ready to begin working on a ML project. They need to build a data preparation pipeline and someone on the team suggests they reuse the same pipeline they created for their last project. What's wrong with this suggestion?

Data pipelines are tightly linked to the specific data patterns to be detected and the exact machine-learning model requirements. Reusing a pipeline built for a different pattern or model risks improper feature engineering and data bias.

A – Operationalization specificity is secondary; pattern/model needs drive pipeline design. C – Even if the high-level pattern seems similar, differing model requirements alter transformations. D – Blind reuse ignores unique schema, quality and feature needs of each project.

1. Cognilytica. “CPMAI v7 Guidebook,” Sec 7.2 “Data Pipelines,” p. 105.

2. Stanford CS229 Lecture Notes, “Feature Processing & Pipelines,” 2020, pp. 3-4.