Learn to evaluate machine-learning models rigorously and honestly with Python and scikit-learn. You will build trustworthy train/validation/test workflows, read confusion matrices, choose the right metric for each problem, and diagnose over- and underfitting so your reported numbers reflect real-world performance.
Watch the free preview
Why evaluation is hard — train, validation and test — free to watch, no account needed.
What you'll learn
- Evaluate models rigorously and honestly using disciplined validation workflows
- Split data into train, validation and test sets and run k-fold cross-validation without leakage
- Interpret confusion matrices and compute precision, recall, F1, ROC and AUC correctly
- Choose regression metrics and classification metrics that match the real problem
- Read learning curves to detect over- and underfitting and act on the diagnosis
Syllabus
Honest Evaluation Foundations
Why evaluation is hard — train, validation and testFree preview
Cross-validation for stable estimates
Classification Metrics That Tell the Truth
The confusion matrix
Precision, recall and F1
Thresholds, Ranking and Regression
ROC, AUC and the decision threshold
Regression metrics
Diagnosing and Choosing
Learning curves and over/underfitting
Choosing the right metric for the problem
Lab — 10 Exercises & Solutions
Exercises 1–5
Exercises 6–10