A spectrum of models

Fully global models – e. g. Polynomial • May be slow to fit – Each parameter depends on all the data

Very local models – e.g. Nearest neighbors • Very fast to fit – Just store training cases • Local smoothing obviously improves things

y

y

x

x