Description
Models to Systems Practical is a hands-on capstone for learners who want to move beyond model training and understand what makes a machine-learning workflow dependable in operation.
You will build a complete decision-support demonstration for prioritising fictional service records for human follow-up. The practical connects the model to the controls, evidence, monitoring, and operating decisions required around it.
What you will learn
- Define the decision purpose, users, boundaries, and prohibited uses.
- Create and validate a versioned input data contract.
- Build a reproducible preprocessing-and-classification pipeline.
- Evaluate discrimination, calibration, threshold trade-offs, and subgroup performance.
- Package the fitted pipeline with a manifest, software versions, data hashes, and checksums.
- Generate traceable batch predictions and an audit log.
- Monitor schema quality, missingness, feature distributions, and prediction behaviour.
- Define alerts, escalation, human review, rollback, retraining, and retirement decisions.
What is included
- A browser-based online practical with step-by-step guidance.
- Python scripts for synthetic data generation, training, model packaging, batch prediction, and monitoring.
- Bash scripts for the full workflow, shared-template copying, practical splitting, and customer ZIP creation.
- A machine-readable data contract.
- Generated evaluation, threshold, subgroup, audit, monitoring, and documentation outputs.
- A learner README and reproducible environment requirements.
Who this practical is for
Suitable for students, researchers, analysts, professionals, mentors, and learners who understand introductory machine learning and want practical experience connecting models to reliable systems.
Tools used
Python, pandas, NumPy, scikit-learn, joblib, Bash, and Quarto.
Important: The supplied dataset is synthetic and intended only for teaching and workflow testing. This practical is not a validated production system and must not be used to make real decisions about people.
This is a digital product. No physical item will be shipped.
