rqPen (version 3.2.2)

QICD.nonpen: Implements QICD algorithm with some variables not being penalized

Description

Implements QICD algorithm with some variables not being penalized

Usage

QICD.nonpen(
  y,
  x,
  z,
  tau = 0.5,
  lambda,
  intercept = TRUE,
  penalty = "SCAD",
  initial_beta = NULL,
  maxin = 100,
  maxout = 20,
  eps = 1e-05,
  coef.cutoff = 1e-08,
  a = 3.7,
  method = "br",
  scalex = TRUE,
  ...
)

Arguments

y

response variable, length n vector

x

input nxp matrix, of dimension nobs x nvars; each row is an observation vector.

z

nxq matrix of bases; the coefficients for these columns will be unpenalized

tau

the quantile value

lambda

the tuning parameter (numeric value > 0)

intercept

a logical value,should intercept be fitted (default=TRUE) (intercept should be included when using splines)

penalty

The name of the penalty function ("SCAD", "MCP", "LASSO")

initial_beta

Vector containing initial values for intercept (if included) and x coefficients. Should be in the form (intercept, coefficients) intercept should be left out if intercept=FALSE. The intercept should be included to be consistent with other methods, but intercept and z coefficients will be initialized to by a rq() fit of residuals from initial beta against the unpenalized predictors, z.

maxin

maximum number of iterations for inside coordinate descent,default value is 100

maxout

maximum number of iterations for outside MM step, default value is 20

eps

The convergence threshold for coordinate descent and majorization minimization step

coef.cutoff

Threshold for determining nonzero coefficients

a

Scale parameter, the default value is 3.7 (>2 for SCAD, >1 for MCP, not used in LASSO)

method

quantile regression initialization method, can be "br" or "fn".

scalex

Whether predictors are centered and scaled

...

additional parameters