optPenaltyGGMmixture.kCVauto: Automatic search for optimal penalty parameter (mixture of...
In porridge: Ridge-Type Penalized Estimation of a Potpourri of Models

optPenaltyGGMmixture.kCVauto

R Documentation

Automatic search for optimal penalty parameter (mixture of GGMs).

Description

Function that performs an automatic search for the optimal penalty parameter for the ridgeGGMmixture call by employing Brent's method to the calculation of a cross-validated (negative) log-likelihood score.

Usage

optPenaltyGGMmixture.kCVauto(Y, K, lambdaMin, lambdaMax, 
                lambdaInit=(lambdaMin+lambdaMax)/2, 
                fold=nrow(Y), target,               
                iWeights=matrix(sample(seq(0+1/nrow(Y), 
                                1-1/nrow(Y), by=1/(2*nrow(Y))), 
                                nrow(Y)*K, replace=TRUE), 
                                nrow=nrow(Y), ncol=K),
                nInit=100, minSuccDiff=10^(-10),
                minMixProp=0.01)

Arguments

`Y`	Data `matrix` with samples as rows and variates as columns.
`K`	A `numeric`, specifying the number of mixture components.
`lambdaMin`	A `numeric` giving the minimum value for the penalty parameter.
`lambdaMax`	A `numeric` giving the maximum value for the penalty parameter.
`lambdaInit`	A `numeric` giving the initial (starting) value for the penalty parameter.
`fold`	A `numeric` or `integer` specifying the number of folds to apply in the cross-validation.
`target`	A semi-positive definite target `matrix` towards which the estimate is shrunken.
`iWeights`	Sample-specific positive component weight `matrix`. Rows correspond to samples, while columns to components.
`nInit`	A `numeric` specifying the number of iterations.
`minSuccDiff`	A `numeric`: minimum successive difference (in terms of their penalized loglikelihood) between two succesive estimates to be achieved.
`minMixProp`	Smallest mixing probability tolerated.

Value

The function returns a positive numeric, the cross-validated optimal penalty parameter.

Note

The elements of iWeights may be larger than one as they are rescaled internally to sum to one.

Author(s)

W.N. van Wieringen, M. Aflakparast.

References

Aflakparast, M., de Gunst, M.C.M., van Wieringen, W.N. (2018), "Reconstruction of molecular network evolution from cross-sectional omics data", Biometrical Journal, 60(3), 547-563.

Examples

# define mixing proportions
pis <- c(0.2, 0.3, 0.4)

# set dimension and sample size
p <- 5
n <- 100

# define population covariance matrices
diags       <- list(rep(1,    p), 
                    rep(0.5,  p-1), 
                    rep(0.25, p-2), 
                    rep(0.1,  p-3))
Omega       <- as.matrix(Matrix::bandSparse(p, 
                                            k   =-c(0:3), 
                                            diag=c(diags), 
                                            symm=TRUE))
Sigma1      <- solve(Omega)
Omega       <- matrix(0.3, p, p)
diag(Omega) <- 1
Sigma2      <- solve(Omega)
Sigma3      <- cov(matrix(rnorm(p*n), ncol=p))

# mean vectors
mean1 <- rep(0,p)
mean2 <- rexp(p)
mean3 <- rnorm(p)

# draw data data from GGM mixture
Z <- sort(sample(c(1:3), n, prob=pis, replace=TRUE))
Y <- rbind(mvtnorm::rmvnorm(sum(Z==1), mean=mean1, sigma=Sigma1),
           mvtnorm::rmvnorm(sum(Z==2), mean=mean2, sigma=Sigma2),
           mvtnorm::rmvnorm(sum(Z==3), mean=mean3, sigma=Sigma3))

# find optimal penalty parameter
### optLambda <- optPenaltyGGMmixture.kCVauto(Y,  K=3,         
###                                          0.00001, 100,    
###                                          10, fold=5,      
###                                          target=0*Sigma1) 

# ridge penalized estimation of the GGM mixture
### ridgeGGMmixFit <- ridgeGGMmixture(Y, 3, optLambda, target=0*Sigma1)

porridge documentation built on Oct. 16, 2023, 1:06 a.m.