filter_glm: Adaptive knockoff filter With GLM (Generalized Linear Model)

In zhimeir/adaptiveKnockoffs: Implementing knockoffs with side information.

Description

filter_glm returns a set of rejections with FDR controlled at custom target

Usage

filter_glm(W, z, alpha = 0.1, offset = 1, reveal_prop = 0.5, mute = TRUE)

Arguments

W	vector of length p, denoting the imporatence statistics calculated by knockoff.filter.
z	p-by-r matrix of side information.
alpha	target FDR level (default is 0.1)
offset	either 0 or 1 (default: 1). The offset used to compute the rejection threshold on the statistics. For details, see knockoff.threshold.
reveal_prop	The proportion of hypotheses revealed at intialization (default is 0.5).
mute	whether \hat{fdp} of each iteration is printed (defalt is TRUE).

Value

A list of the following:

nrejs	The number of rejections for each specified target fdr (alpha) level
rejs	Rejsction set fot each specified target fdr (alpha) level
rej.path	The order of the hypotheses (used for diagnostics)
unrevealed.id	id of the hypotheses that are nor revealed in the end (used for diagnostics)
tau	Threshold of each target FDR level (used for diagnostics)
acc	The accuracy of classfication at each step (used for diagnostics)

See Also

Other filter: filter_EM(), filter_gam(), filter_glmnet(), filter_randomForest_getorder(), filter_randomForest()

Examples

#Generating data
p=100;n=100;k=40;
mu = rep(0,p); Sigma = diag(p)
X = matrix(rnorm(n*p),n)
nonzero = 1:k
beta = 5*(1:p%in%nonzero)*sign(rnorm(p))/ sqrt(n)
y = X%*%beta + rnorm(n,1)

#Generate knockoff copy
Xk = create.gaussian(X,mu,Sigma)

#Gnerate importance statistic using knockoff package
W = stat.glmnet_coefdiff(X,Xk,y)

#Using filer_gam to obtain the final rejeciton set
z = 1:p #Use the location of the hypotheses as the side information
result = filter_glm(W,z)

zhimeir/adaptiveKnockoffs documentation built on Oct. 6, 2021, 9:41 p.m.