Outline

Why

demo where lm fails
vs 'outliers'
vs transformation fail

Introduce with standard polynomial

where works
where overfits
where isn't applicable/fails

Introduce GAM

Interpretation

Extensions

smooths

Technicals

penalized regression
choice of smooth
choice/estimation of df

Ties to other

Random effects
GP
Boosting/RF
Spatial