Quote:
Originally Posted by rainbow
Hi,
In section 3.4.1, on page 104, it says "For instance, linear regression is often coupled with a feature transform to perform nonlinear regression".
As I understand it, the feature transform are on inputs, but the the model is still linear in the parameters w. Don't we then still have a linear model?

Correct. The term "nonlinear regression" is sometimes used when the dependence on the input variables is nonlinear, but indeed what matters as far as the algorithm is concerned is linearity in the parameters.