ARMADA.select: Covariates selection via 8 selection methods

In armada: A Statistical Methodology to Select Covariates in High-Dimensional Data under Dependence

Description

Covariates selection via 8 selection methods

Usage

ARMADA.select(X, X.decorrele, Y, test, type.cor.test = NULL,
  type.measure_glmnet = c("deviance", "class"),
  family_glmnet = c("gaussian", "binomial", "multinomial"),
  clusterType = c("PSOCK", "FORK"), parallel = c(FALSE, TRUE))

Arguments

X	the matrix (or data.frame) of covariates, dimension n*p (n is the sample size, p the number of covariates). X must have rownames and colnames.
X.decorrele	the matrix of decorrelated covariates, dimension n*p (n is the sample size, p the number of covariates). X.decorrele has been obtained by the function X_decor.
Y	the vector of the response, length n.
test	the type of test to apply ("wilox.test" or "t.test" if Y is a binary variable; "kruskal.test" or "anova" if Y is a factor with more than 2 levels; "cor.test" if Y is a continuous variable).
type.cor.test	if test="cor.test", precise the type of test (possible choices: "pearson","kendall", "spearman"). Default value is NULL, which corresponds to "pearson".
type.measure_glmnet	argument for the lasso regression. The lasso regression is done with the function cv.glmnet (package glmnet), and you can precise the type of data in cv.glmnet. Possible choices for type.measure_glmnet: "deviance" (for gaussian models, logistic, regression and Cox), "class" (for binomial or multinomial regression).
family_glmnet	argument for the lasso regression. The lasso regression is done with the function glmnet. Possible choices for family_glmnet: "gaussian" (if Y is quantitative), "binomial" (if Y is a factor with two levels), "multinomial" (if Y is a factor with more than two levels).
clusterType	to precise the type of cluster of the machine. Possible choices: "PSOCK" or "FORK" (for UNIX or MAC systems, but not for WINDOWS).
parallel	TRUE if the calculus are made in parallel.

Details

The function ARMADA.select applies 8 selection methods on the decorrelated covariates (named X.decorrele), given the variable of interest Y. It resturns a list of 8 vectors of the selected covariates, each vector correspond to one selection method. The methods are (in the order): Random forest (threshold step), Random forest (interpretation step), Lasso, multiple testing with Bonferroni, multiple testing with Benjamini-Hochberg, multiple testing with qvalues, multiple testing with localfdr, FAMT.

Value

a list with 8 vectors, called: genes_rf_thres, genes_rf_interp, genes_lasso, genes_bonferroni, genes_BH, genes_qvalues, genes_localfdr, genes_FAMT. The 8 vectors are the selected covariates by the corresponding selection methods.

armada documentation built on May 2, 2019, 6:37 a.m.