rbvs (version 1.0.2)

mcplus.coef: Measure an impact of the covariates on the response using MC+. This function evaluates the MC+ coefficients regressing y onto the design matrix x over subsamples in subsamples.

Description

Measure an impact of the covariates on the response using MC+. This function evaluates the MC+ coefficients regressing y onto the design matrix x over subsamples in subsamples.

Usage

mcplus.coef(x, y, subsamples, nonzero = NULL, family = c("gaussian", "binomial"), alpha = 1, gamma = 3, maxit = 500, tol = 0.01, lambda.ratio = 1e-06, nlam = 25, ...)

Arguments

x

Matrix with n observations of p covariates in each row.

y

Response vector with n observations.

subsamples

Matrix with m indices of N subsamples in each column.

nonzero

Number of non-zero coefficients estimated for each subsample.

family

Determines the likelihood optimised in the estimation procedure.

alpha

Scalar between 0 and 1 determining l2 penalty (see details).

gamma

Scalar greater than 1. The concacivity parameter (see details).

maxit

Maximum number of itarations when computing the MC+ coefficients.

tol

Scalar determining convergence of the MC+ algorithm used.

lambda.ratio

Scalar being a fraction of 1. Used in the MC+ algorithm

nlam

Number of penalty parameters used in the MC+ algorithm.

...

Not in use.

Details

To solve the MC+ problem, we implement the coordinate descent algorithm as in Breheny Jian (2011).

References

Zhang, Cun-Hui. "Nearly unbiased variable selection under minimax concave penalty." The Annals of Statistics (2010): 894-942.

Breheny, Patrick, and Jian Huang. "Coordinate descent algorithms for nonconvex penalized regression, with applications to biological feature selection." The Annals of Applied Statistics 5.1 (2011): 232.