Prepare penalty arguments

Usage

penalty_args(args)

Arguments

args: A list of arguments. See See *Details*.

Value

A list of arguments

Details

The following arguments are supported:

name: The penalty function. Currently, only "adaptive_sparse_group_lasso" is supported.
alpha: The mixing parameter. Must be between 0 and 1. 1 is pure Lasso (default); 0 is pure group Lasso.
a: The SCAD parameter, requires >=2. Default: 3.7.
lambda: The penalty parameter. If NULL (default), a logarithmic grid search is used.
n_lambda: The number of penalty parameters to use if lambda is NULL (default.)
lambda_factor: The factor by which the penalty parameter is reduced overall (default: 1e-4.)
adaptive: The adaptive parameter. Must be non-negative (default: 0..)
penalize_intercept: Whether to penalize the intercept. Must be TRUE (default) or FALSE.
...: Additional arguments. Currently ignored.

For the grid search, we first obtain the minimum \(\lambda_\max\) for which the solution is zero. Then, a logarithmically-space sequence of \(\lambda\) values is generated starting from \(\lambda_\max\) and ending at \(\lambda_\max \times \texttt{lambda_factor}\).