Generalisation, Splits, and Leakage
Learners can identify invalid evaluation, repair leakage, and explain a clean train/validation/test protocol.
Executable courseRead the lesson pages, run the browser practice notebook, and keep the downloadable fallback pack if you want a local copy.
Module sequence
- Why a model must work beyond the rows it sawUnit 1 · 20-35 minutes
- Training, validation, and test setsUnit 2 · 20-35 minutes
- Overfitting, underfitting, and model complexityUnit 3 · 20-35 minutes
- Cross-validation and variation across foldsUnit 4 · 20-35 minutes
- Time-aware and group-aware splittingUnit 5 · 20-35 minutes
- Leakage from columns, preprocessing, duplicates, and future informationUnit 6 · 20-35 minutes
- Learning curves and simple complexity checksUnit 7 · 20-35 minutes
- Project step: repair an invalid experimentUnit 8 · 20-35 minutes
- Activity: Repair an Invalid ExperimentPractice activity · 45-75 minutes
- Module 05 Assessment: Generalisation and Leakage CheckModule check · 35-50 minutes