A directed acyclic graph (DAG) connects nodes with one-way edges and contains no cycles. In an ML pipeline, it orders tasks and runs each task only after its dependencies complete. section 5.1.2
A Kubeflow pipeline uses the DAG to:
- Represent components as executable tasks. section 5.1.2
- Enforce dependencies by passing one task's output into another task's input. section 5.1.2
Income classifier pipeline
Four tasks fetch data, retrieve features, run inference, and write results. The feature task consumes fetch_data_task.outputs["data_output"], and inference consumes the feature task's output.
