Chapter 7 - Moving beyond linearity

Question 1

If we fit curves where we use arg min RSS + penalty where the penalty is the area under the curve weighted with a parameter lambda, how is the curve affected?

Answer 1

• Larger lambda parameter gives more empathies on the penalty.
• Look for the deegre of the drivative in the penalty term - That is the degree we penalize, we measure the area under that curve.
• As lambda approaches inf, all empathies is on the penalty and the resulting model curve will have one degree less than the penalty term.

Question 2

If we fit curves where we use arg min RSS + penalty where the penalty is the area under the curve weighted with a parameter lambda, how is the flexibility affected?

Answer 2

Look at the penalty term - The term with a higher degree of derivative in the penalty term is more flexible.