Skip to contents

Derivative of the non-convex regularization penalty

Source: R/deriv.R
deriv.Rd

Compute the derivative of the specified non-convex regularization penalty.

Usage

deriv(penalty, Omega, lambda, gamma)

Arguments

penalty

A character string specifying the non-convex penalty to use. Available options include:

  1. "adapt": adaptive lasso zou2006adaptive,fan2009networkfcstat.

  2. "atan": arctangent type penalty wang2016variablefcstat.

  3. "exp": exponential type penalty wang2018variablefcstat.

  4. "mcp": minimax concave penalty zou2006adaptivefcstat.

  5. "scad": smoothly clipped absolute deviation fan2001variable,fan2009networkfcstat.

Omega

The precision matrix.

lambda

A scalar specifying the regularization parameter.

gamma

A scalar specifying the hyperparameter for the penalty function. The defaults are:

  1. "adapt": 0.5

  2. "atan": 0.005

  3. "exp": 0.01

  4. "mcp": 3

  5. "scad": 3.7

Value

A numeric matrix.

References