The R package gamsel implements an algorithm for generalized additive model selection developed by and described in Chouldechova & Hastie. The algorithm delivers a path of selected models that is parameterized by a positive scalar parameter, denoted by \(\lambda\). Higher values of \(\lambda\) correspond to sparser models.

In this vignette we work through some illustrative examples involving simulated and actual data. We explain how to deal with issues arising in actual data such as candidate predictors that are categorical, heavily skewed or ostensibly are continuous but have a low number of unique observed values.