A feature bundle consolidates related features from the same data sources when they share a key, lineage, and transformations. It replaces separate column-level pipelines with one computational unit. section 11.4.2
Related price features
Price before discount, price after discount, and price after applying a promotion can form one bundle because they share a key, lineage, and similar transforms or filters. section 11.4.2
A practical feature-bundle design should:
- Merge the computational graphs of features derived from the same sources. section 11.4.2
- Expose similar features as a set through an API or UI so they can be added to a dataset simultaneously. section 11.4.2
- Track lineage to order calculations, propagate feature-version updates, and investigate surprising behavior. section 11.4.2
Figure 11.15 A graph of feature dependencies
Use the dependency graph to inspect how bundled features relate to parent computations and update order. section 11.4.2
