GCMlasso: Gaussian graphical model for ordered variables

In jialiwang1211/GCMlasso: Bayesian Gaussian graphical model with clustering structure for ordered variables.

Description

Perform Bayesian Gaussian graphical model for ordered variables with clustering structure. The conditional independence between variables are measured on the latent scale via the extended rank likelihood method. Shrinkage effects are applied on the precision matrix to handle multicollinearity. Clustering effects are modelled through the random effects. Missing data are allowed.

Usage

GCMlasso(data, var_ord, var_group, nsamp = 1000, odens = 1,
  nwarm = 500, seed = 1, s = 0.01, t = 0.01, verb = TRUE)

Arguments

data	an N by p data frame,
var_ord	indices of ordinal variables.
var_group	index of the variable that defines clusters.
nsamp	number of MCMC iterations.
odens	number of iterations between saved samples.
nwarm	number of MCMC iterations as burn-in.
seed	a random integer.
s	hyperparameter of lambda, degree of freedom in adaptive graphical lasso prior.
t	hyperparameter of lambda, shrinkage in adaptive graphical lasso prior.
verb	print progress of MCMC, logical TRUE or FALSE.

Details

GCMlasso function fits the Bayesian Gaussian copula model with graphical lasso prior for variables with ordering (continuous, ordinal and binary) in multilevel data sets. Adaptive graphical lasso prior is put on the precion matrix of the latent variables conditional on the random effects, where the latent variables are implied by the extended rank likelihood method.

var_group is the index of the varible that defines the clusters, and should be placed at the last column in the data. The coding for the clustering variable is from 1 to the total number of clusters. The binary variables in var_ord should be coded as 0 (control) or 1(case) in the case-control studies.

Missing data are allowed for ordered variables and should be denoted as NA.

Value

An object with S3 class "GCMlasso" is returned.

data_ordered	the same as the input data but ordered by the clustering variable.
Gamma.st	saved variance covariance matrices for latent variables.
Omega.st	saved precision matrices for latent variables.
psi.st	saved variance covariance matrices for random effects.
z.st	saved latent variables.
b.st	saved random effects.

Author(s)

Jiali Wang ([e-mail])

References

\insertRef

hoff2007extendingGCMlasso

\insertRef

wang2012bayesianGCMlasso

\insertRef

wang2017copulaGCMlasso

Examples

GCMlasso_obj<-GCMlasso(data=Framingham,var_ord=1:15,var_group=16,
  nsamp=1000,odens=1,nwarm=500,seed=1,s=1e-2,t=1e-2,verb=TRUE)

jialiwang1211/GCMlasso documentation built on May 14, 2019, 12:55 a.m.