Additive models and generalised additive models from MATH5806: one smooth per predictor, a basis and a penalty for each, penalised least squares or penalised IRLS to fit, and GCV to choose the smoothing.
Cross-Validation
Choosing among regression models and estimating the error of the one chosen: decision-theoretic error definitions, the optimism of the training error, cross-validation with the leave-one-out shortcut, bootstrap error estimates, information criteria and subset selection, worked on the Cheddar data.