What are the advantages and disadvantages of a very flexible $($versus a less flexible$)$ approach for regression or classification? Under what circumstances might a more flexible approach be preferred to a less flexible approach? When might a less flexible approach be preferred?