Contract

• Input: dataset description, problem type, performance target.
• Output: pipeline architecture + training report + deployment checklist.
• Side effects: may create artifacts (models, logs) when executed.
• Dependencies: external ML framework (scikit-learn, PyTorch, TensorFlow, XGBoost) and data source.
• Stop condition: pipeline documented; model validated; checklist filled.
• Risk: medium — model decisions affect users; requires validation.
• Boundary: designs pipeline; does not train production models unless explicitly executed.

ML Pipeline Design

Build a machine-learning pipeline from data to deployed model with reproducibility and fairness checks.

Process

1. Frame the problem

State: supervised / unsupervised / reinforcement; classification / regression / clustering; time-series / tabular / image / text / tabular-time-series.