filter_EM: Adaptive knockoff filter based on Bayeian two group model

In zhimeir/adaptiveKnockoffs: Implementing knockoffs with side information.

Description

filter_EM takes the inmportance statistic W as input. The filter determines the order of the statistic by the Bayesian posterior probability of being a null.

Usage

filter_EM(
  W,
  U,
  alpha = 0.1,
  offset = 1,
  mute = TRUE,
  df = 3,
  R = 1,
  s0 = 0.005,
  cutoff = NULL
)

Arguments

W	vector of length p, denoting the imporatence statistics calculated by knockoff.filter.
U	p-by-r matrix of side information.
alpha	target FDR level (default is 0.1).
mute	whether \hat{fdp} of each iteration is printed (defalt is TRUE).
df	Degree of freedom of the splines (default is 3).
R	Number of iterations in the EM algorithm
reveal_prop	The proportion of hypotheses revealed at intialization (default is 0.4).

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_gam(), filter_glmnet(), filter_glm(), 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)

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

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

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