dpd.sis: DPD-SIS and DPD-CSIS under a Generalized Linear Model (GLM)...

In abhianik/dpdSIS: Robust Variable Screening with DPD-SIS and DPD-CSIS

DPD-SIS and DPD-CSIS under a Generalized Linear Model (GLM) with no variance parameter

Description

Performs the Density Power Divergence based Robust Variable Screening (DPD-SIS) and associated Robust Conditional Variable Screening (DPD-CSIS) under a GLM with no variance parameter, using parallel computation.

Usage

dpd.sis(d, y, X,  alpha,  reg = c("lrm", "logistic", "poisson"),  XC = matrix(rep(1, nrow(X))), Initial = matrix(rep(1, ncol(X))), Method = "L-BFGS-B")

Arguments

d	Integer. A list of the data matrices. Number of features to be selected. (1<= d <=p)
y	Vector. The response vector [n X 1].
X	Matrix. Covariate Matrix [n X p]. It shoudl only include variables to be considered for screening. So, it should not contain the intercept or any conditioning variables.
alpha	Numeric. The DPD tuning parameter (0<= alpha <=1)
reg	A string indiacting the regression model. Possible options are "lrm" (default), "logistic" and "poisson", indicating the linear regression (with unit error variance), logistic regression and poisson regression, respectively.
Initial	Vector. Initial values of the marginal slope parameter for estimation process. Default: 1 for all parameters. Optional.
Method	String. Numerical optimization method to be used for computation of marginal slopes. Possible options are "L-BFGS-B","Nelder-Mead", "BFGS", "CG", which are the same as the input of 'optim' function in R. Optional. Default is "L-BFGS-B".
X_C	Matrix. Conditioning Covariate Matrix [n X q] in DPD-CSIS. Default: only the intercept variable (q=1). Optional for DPD-SIS.

Details

Reference: Ghosh A, Ponzi E, Sandanger T, Thoresen M. Robust Sure Independence Screening for Non-polynomial dimensional Generalized Linear Models. arXiv preprint 2021; arXiv:2005.12068v2.

Value

SIS Vector. A d X 2 vector where the first column contains the variable index (ranked in decreasing order of importance) and the second column consist of the corresponding MDPDE of the slope

Examples

n <- 50; p <- 500;
beta <- rep(0,p); beta[c(1:5)] <- c(1,1,1,1,1);
d <- floor(n/log(n));
Sigma <- diag(p-1);
X <- mvrnorm(n, mu=rep(0,p-1), Sigma=Sigma);
X0 <- cbind(1,X)
Y <- drop(X0 %*% beta + 2*rnorm(n))

alpha <- 0.3
SIS<-dpd.sis(d,Y,X,alpha, reg='lrm')

abhianik/dpdSIS documentation built on Sept. 5, 2022, 12:40 p.m.