RidgeFusion (version 1.0-3)

RidgeFused: Ridged Fused Inverse Covariance Matrix Estimation

Description

Calculates the ridge fusion precision estimator for multiple classes

Usage

RidgeFused(S,lambda1,lambda2,nc,tol=10^(-7), maxiter=1e3,warm.start=NULL,scale=FALSE)

Arguments

S

A list of length J that contains the sample covariance estimators of each class

lambda1

Ridge tuning parameter, must be greater than or equal to 0

lambda2

Ridge Fusion tuning parameter, must be greater than or equal to 0

nc

A vector of length J that contains the sample size of each class

tol

Convergence tolerance for blockwise coordinate descent algorithm

maxiter

The number of iterations the algorithm will run if convergence tolerance is not met

warm.start

If NULL no warm start is used. If initialized with a list of positive definite inverse covariance matrix estimates of length J, will use them as initialization for the algorithm.

scale

If FALSE scale invariant method is used

Value

An object of class RidgeFusion, basically a list including elements

Omega

a list where each element is the inverse covariance matrix estimate for the corresponding element of S

Ridge

lambda1

FusedRidge

lambda2

iter

Number of iterations until convergence

Examples

## Creating a toy example with 5 variables
library(mvtnorm)
set.seed(526)
p=5
    Sig1=matrix(0,p,p)
	for(j in 1:p){
		for(i in j:p){          
            Sig1[j,i]=.7^abs(i-j)
            Sig1[i,j]=Sig1[j,i]
            
		}
	}
    Sig2=diag(c(rep(2,p-5),rep(1,5)),p,p)
X1=rmvnorm(100,rep(2*log(p)/p,p),Sig1)
Y=rmvnorm(100,,Sig2)
## Creating a list to use as S
S=list(0,0)	
S[[1]]=(99/100)*cov(X1)
S[[2]]=(99/100)*cov(Y)

## Creating the vector of sample sizes
nc2=c(100,100)

## Running RidgeFused scale invariant method for tuning parameters lambda1=1 ,lambda2=2
A=RidgeFused(S,1,2,nc2,scale=TRUE)
A
names(A)