Machine learning: patterns and evidence

Learn the shapes different models can represent, how data becomes a training signal, and how to tell whether learning transfers to new cases.

Fit a useful baseline

Start with linear models and trees on a small table. Ask what structure each model assumes before comparing scores.

Clusters and projections depend on distances and assumptions. Draw a counterexample where a method gives a misleading picture.

Split data the way new examples arrive. Trace leakage, select task-relevant metrics, and examine important subgroups.