POCRE (version 0.6.0)

gps: Screen Variables for Generalized Linear Models via Generalized POCRE

Description

A pre-specified number (i.e., maxvar) of covariates will be selected for generalized linear models by constructing maxcmp components with generalized POCRE. Each component will be constructed by selecting maxvar/macmp covariates which are most relevant to the response variable(s). Similar to pocrescreen, gps selects covariates for their top relevance to the response variable(s) without penalization.

Usage

gps(y, x, family="binomial", bc.method="optimal", x.include=NULL,
    weights=NULL, maxcmp=10, maxvar=NULL, tol = 1e-6, maxit = 100)

Arguments

y

n*q matrix, values of q response variables (allow for multiple response variables).

x

n*p matrix, values of p predicting variables (excluding the intercept).

family

Family objects as family. Currently only support "gaussian", "binomial" (by default), and "poisson".

bc.method

Bias correction method.

x.include

a vector of indices indicating covariates which should always be included in the model (so not counted into selected maxvar covariates).

weights

A vector, including a prespecified weight for each observation (set as 1/n by default).

maxcmp

maximum number of components to be constructed.

maxvar

maximum number of selected variables.

tol

tolerance of precision in iterations.

maxit

maximum number of iterations to be allowed.

Value

a vector of indices of selected covariates (excluding those in x.include).

References

Zhang D, Lin Y, and Zhang M (2009). Penalized orthogonal-components regression for large p small n data. Electronic Journal of Statistics, 3: 781-796.

See Also

pocrescreen.

Examples

# NOT RUN {
  # Binomial Data
  data(simbin)
  gps(simbin[,1], simbin[,-1], maxcmp=3, maxvar=9)
  gps(simbin[,1], simbin[,-1], x.include=103:104, maxcmp=3, maxvar=9)
 
 # Count Data
  data(simpoi)
  gps(simpoi[,1], simpoi[,-1], family='poisson',maxcmp=5,maxvar=10)
# }