Supervised Learning Mental Models

Supervised learning estimates a function (f(x) \to y) from labeled examples.

Core vocabulary

• Features (x): measurable inputs
• Label (y): the target you can supervise
• Hypothesis class: the family of functions your model can represent
• Loss: how wrong a prediction is

The workflow that actually matters

1. Define the decision the model will support
2. Collect representative labels
3. Split data to estimate generalization (not just fit)
4. Establish a baseline before complex models
5. Evaluate with metrics tied to business cost

Common failure modes

• Leakage from future or label-derived features
• Optimizing accuracy when recall/precision matter
• Ignoring distribution shift between train and production