Variational Bayes for linear mixed model (random intercept and coefficient only).

Performs variational inference for random intercept and coefficient model. Currently assumes that all groups have same number of parameters. That is, that all Zlist elements are of equal dimension.

Usage

vb_lmm_randintslope(
  Xlist,
  Zlist,
  ylist,
  beta_mu0,
  beta_sigma0,
  nu_Omega0,
  lambda_Omega0,
  pr_Omega = 1L,
  sigma_a0 = 0.01,
  sigma_b0 = 0.01,
  pr_sigma = 1L,
  tol = 1e-08,
  maxiter = 100L,
  verbose = FALSE,
  trace = FALSE,
  streamlined = FALSE,
  use_elbo = TRUE
)

Arguments

Xlist

A list of subject specific design matrices

Zlist

A list of subject specific group matrices

ylist

A list of subject specific responses

beta_mu0

Prior mean for beta

beta_sigma0

Prior covariance for beta

nu_Omega0

Prior df for Omega

lambda_Omega0

Prior scale matrix for Omega

pr_Omega

Prior type for Omega - 1 is IW(nu, lambda), 2 is HW(nu, 2*nu*diag(1/lambda^2))

sigma_a0

The first hyper-parameter for prior on sigma

sigma_b0

The second hyper-parameter for prior on sigma

pr_sigma

The prior to use for sigma_epsilon - 1 is IG(a0,b0) and 2 is Half-t(a0, b0)

tol

Tolerance level for assessing convergence

maxiter

Maximum number of fixed-update iterations

verbose

Print trace of ELBO

trace

Return trace of parameters beta, gamma

streamlined

Use streamlined updates (more efficient if dim(Zlist) is large).

use_elbo

Should the ELBO be calculated and used for convergence checks?