Section 1

Supervised Learning

Section 1 builds supervised learning on 442 diabetes patients and 569 breast-cancer tumours, from the evaluation protocol the whole volume uses through least squares, logistic regression and trees to gradient boosting and the metrics, intervals and calibration that say what a classifier's numbers are worth. Each chapter ends on a choice a team has to make with numbers, from how to spend labelled data on holdout, cross-validation or a backtest in Chapter 1 to how many positive and negative cases a screening launch must label in Chapter 6.