filter_gam: Adaptive knockoff filter based on GAM (Generalized Additive...

In zhimeir/adaptiveKnockoffs: Implementing knockoffs with side information.

Description

filter_gam takes the inmportance statistic W as input. The filter determines the order of the statistic by the probability of being a null. The probability is given by GAM.

Usage

filter_gam(
  W,
  U,
  alpha = 0.1,
  offset = 1,
  df_list = 6:10,
  reveal_prop = 0.1,
  mute = TRUE
)

Arguments

W	vector of length p, denoting the imporatence statistics calculated by knockoff.filter.
U	p-by-r matrix of side information.
alpha	(a vector of) target FDR levels (default is 0.1).
df_list	A list of candidates for the degree of freedom of the splines (default is 6:10).
reveal_prop	The proportion of hypotheses revealed according to |W| (default is 0.1).
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)

See Also

Other filter: filter_EM(), 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_gam to obtain the final rejeciton set
z = 1:p #Use the location of the hypotheses as the side information
result = filter_gam(W,z)

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