cv.sparse.mediation.sgrplasso: Conduct K-fold cross validation for sparse mediation with...

In seonjoo/smm: Sparse Mediation For High-Dimensional Mediators

Description

Conduct K-fold cross validation for sparse mediation with group lasso with multiple tuning parameters

Usage

cv.sparse.mediation.sgrplasso(
  X,
  M,
  Y,
  tol = 10^(-5),
  K = 5,
  max.iter = 100,
  lambda1 = exp(-5:0),
  lambda2 = exp(seq(0, 0.5 * log(ncol(M)), length = 3)),
  alpha = c(0.5, 0.95),
  group.penalty.factor = c(1, rep(1, ncol(M))),
  penalty.factor = c(1, rep(1, ncol(M) * 2)),
  verbose = FALSE,
  multicore = 1,
  seednum = 1e+05,
  non.zeros.stop = ncol(M)
)

Arguments

X	One-dimensional predictor
M	Multivariate mediator
Y	Outcome
tol	(default -10^(-10)) convergence criterion
K	(default=5) number of cross-validation folds
max.iter	(default=100) maximum iteration
lambda1	(default=exp(-5:0)) tuning parameter for L1 penalization
lambda2	(default=exp(-5:0)) tuning parameter for L1 penalization
group.penalty.factor	(default=c(0,rep(1,V))) give different weight of penalization for the V+1 mediation paths.
penalty.factor	(default=c(0,rep(1,2*V))) give different weight of penalization for the 2V mediation paths.
multicore	(default=1) number of multicore
seednum	(default=10000) seed number for cross validation

Value

cv.lambda: optimal lambda

cv.tau: optimal tau

cv.alpha: optimal tau

cv.mse: minimum MSE value

mse: Array of MSE, length(alpha) x length(lambda) x length (tau)

lambda: vector of lambda

tau: vector of tau used

alpha: vector of alpha used

z: cross-valication results

Author(s)

Seonjoo Lee, [e-mail]

References

TBA

Examples

N=200
V=50
set.seed(1234)
a = c(rep(1,3),rep(0,V-3))*5;b<-a
X = rnorm(N)
M =  X %*% t(a)+ matrix(rnorm(N*V),N,V)
Y =  as.vector(X + M %*% b + rnorm(N))
system.time(cvfit<-cv.sparse.mediation.sgrplasso(X, M, Y, K = 4,multicore = 4, seednum = 20200317))
cvfit$cv.lambda
fit<-sparse.mediation.grplasso(X,M,Y,lambda1 = cvfit$cv.lambda1,lambda2 = cvfit$cv.lambda2)
nonzerogroups = 1-as.numeric((fit$hata!=0)+(fit$hatb!=0) ==0)
refit<-sparse.mediation.grplasso(X,M[,nonzerogroups==1],Y,lambda = 0,lambda2 = cvfit$cv.lambda2)

seonjoo/smm documentation built on Feb. 11, 2021, 5:54 a.m.