The scikit-learn workflow

Why it matters

Every lesson in this phase used a piece of scikit-learn. Real projects need all the pieces at once, in an order that doesn't leak and in a form you can hand to someone else. That form is a pipeline: raw data in, predictions out, with every preparation step learned from training data only. It's also the shape of most production ML systems, whatever library they use.

What you'll be able to do

• Use scikit-learn's shared interface: fit, predict, transform, score
• Build one pipeline that handles missing values, categories and scaling
• Tune and compare different models in a single cross-validated search
• Save a trained pipeline, load it, and score new customers safely

Before this lesson

• Cross-validation
• Preparing data

Next in this phase: Capstone: churn prediction