phub: Modified EM algorithm for hub set selection in hub models...
In zhibinghe/Phub: modified EM algorithm for Penalized Hub Model

phub	R Documentation

Modified EM algorithm for hub set selection in hub models with the null component

Description

Modified EM algorithm for hub set selection in hub models with the null component

Usage

phub(
  G,
  A,
  rho,
  lam,
  pen.type = c("plog", "plasso", "log"),
  iter.max = 1000,
  tol = 1e-04
)

Arguments

`G`	observed group data
`A`	initial estimate of adjacent matrix, the first row contains the probabilities in non-hub group
`rho`	a vector of hub weight
`lam`	tuning parameter for component selection, degenrate to a standard EM without penalty if lam=0
`pen.type`	type of penalty, including 'log': penalization for logarithm of all components; 'plog': penalization for logarithm of partial components except the null component; 'plasso': penalization for lasso form of partial components
`iter.max`	maximum iteration steps
`tol`	threshold for shrinking rho to 0

Value

a list of components

`A`	a matrix containg estimated correlation among nodes
`rho`	a vector containing estimated component weight
`l`	log-likelihood
`iteration`	number of iterations used to converge

Examples

set.seed(2020)
n0 = 5; n=100; T=1000
A0 = GenA(n,n0,0.4,0.1,rep(0.05,n))
G0 = GenG(A0,T,c(0.2,rep(0.8/n0,n0)))
M = 10 
A = matrix(runif((M+1)*n),nrow=(M+1)); diag(A[-1,]) = 1
rho = runif(M+1); rho = rho/sum(rho)
phub(G0,A,rho,0.030,pen.type="log")$rho
phub(G0,A,rho,0.030,pen.type="plog")$rho
phub(G0,A,rho,0.8,pen.type="plasso")$rho

zhibinghe/Phub documentation built on Feb. 21, 2025, 11:52 a.m.