ML Foundations & Evaluation: Cheat Sheet

The whole topic compressed for the night before the exam.

ML basics

≡FactsDefinitions worth quoting

• Mitchell - learning = improve P at task T with experience E, written ⟨P, T, E⟩.
• Supervised - labels given. Regression is continuous, classification is discrete.
• Unsupervised - no labels. Find hidden structure, e.g. clustering.
• Reinforcement - rewards from actions. Output a policy, state → action.

Metrics

$$\text{Precision} = \frac{TP}{TP+FP} \quad \text{Recall (TPR)} = \frac{TP}{TP+FN} \quad \text{Specificity} = \frac{TN}{TN+FP}$$ $$\text{Accuracy} = \frac{TP+TN}{\text{all}} \qquad F_1 = \frac{2PR}{P+R}$$

◎Exam cueThe three one-liners

• FP = Type-1 error, FN = Type-2 error.
• ROC plots TPR (y) against FPR (x) across thresholds; top-left is best.
• AUC near 1 is great, near 0.5 is useless.