mvdn: MVDN: Matrix-variate Differential Network Analysis.

In jijiadong/MVDN: Matrix-variate Differential Network Analysis

Description

Matrix-Variate Differential Network (MVDN) model which is particularly useful in modelling connectivity alteration for matrix-variate data such as in brain connectivity analysis.

Usage

mvdn(data1,data2,lambda=NULL,nlambda=20,method=c("none","aic","bic","nbic"),...)

Arguments

data1	a three-dimensional array (p*q*n1) containing n1 samples.
data2	a three-dimensional array (p*q*n2) containing n2 samples.
lambda	the tuning parameter of lasso penalty. user-supplied lambda sequence; default is NULL.
nlambda	the number of lambda values, default is 20.
method	the method used in the lambda selection.
...	Other arguments that can be passed to mvdn.cov

Value

mvdn	differential network.
lambda	the actual sequence of lambda values used.
nlambda	the number of lambda values used.
opt	indicating which one is the optimized lambda based on different norms ("max": max norm, "1": element-wise, "L1": matrix L1 max norm, "Spectral", "Frobenius", "Nuclear").

See Also

mvdn.cov

Examples

##-------------------------------------------------------------------------------
##generate data
p <- 6; q <- 20; n <- 15
set.seed(1212)
Theta <- graph.generate(p, graph="hub",m=2)
omega1 <- Theta * sample(c(-1, 1), p * p, replace = TRUE) * runif(p * p, 0.1, 0.5)
omega1[lower.tri(omega1, diag = FALSE)] <- 0
omega1 <- omega1 + t(omega1)
diag(omega1) <- abs(min(eigen(omega1)$values)) + 1
sigma1 <- cov2cor(solve(omega1))
omega1 <- solve(sigma1)
omega1[abs(omega1)<10^-4] <- 0
omega2 <- omega1
omega2[1:(p/2),1:(p/2)] <- -1*omega2[1:(p/2),1:(p/2)]
diag(omega2) <- diag(omega1)
sigma2 <- solve(omega2)
delta <- omega2-omega1
sigmaT1 <- 0.4^abs(outer(1:q,1:q,"-"))
sigmaT2 <- 0.5^abs(outer(1:q,1:q,"-"))

X1 <- rmatnorm(n,mean=matrix(0,p,q),sigmaS=sigma1,sigmaT=sigmaT1,method="chol")
X2 <- rmatnorm(n,mean=matrix(0,p,q),sigmaS=sigma2,sigmaT=sigmaT2,method="chol")
##generate data end
##-------------------------------------------------------------------------------

## ------ not run ------
obj <- mvdn(data1=X1,data2=X2,nlambda=10,method="bic")

obj$opt

# use the Frobenius norm
deltahat <- obj$mvdn[[obj$opt["Frobenius"]]]
deltahat

jijiadong/MVDN documentation built on Sept. 6, 2020, 7:15 p.m.