filter_randomForest: Adaptive Knockoff Filter With Random Forest

In zhimeir/adaptiveKnockoffs: Implementing knockoffs with side information.

Description

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

Usage

filter_randomForest(
  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_glm(), filter_randomForest_getorder()

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.