03 — Machine learning classification basics

flowchart LR
    TRAIN["labeled beats\ntrain pile"] --> MODEL["model\nadjusts knobs"]
    MODEL --> TEST["hidden beats\ntest pile"]
    TEST --> SCORE["grade = generalization"]

flowchart TD
    BAD["random beat split"] --> LEAK["same patient both piles\nmodel memorizes patients"]
    GOOD["patient-wise split\nde Chazal DS1 train / DS2 test"] --> FAIR["model must learn disease"]

• Label: the right answer attached to a training example (N/V/S/F/Q).
• Train / test: teach pile vs hidden grading pile.
• Overfitting: perfect on seen beats, useless on new patients.
• Leakage: test info secretly present in training (same patient on both sides).
• Features: computed summaries exposing what matters (our scalogram, lesson 07).

Classification here: one beat window in, one of N/V/S/F/Q out. The only grade that counts is hidden patients — everything else is practice.

Next: 04-metrics.md