Variational inference for binomial logistic regression model

This is an experimental function to perform variational inference for binomial logistic regression models.

Usage

vb_logistic_n(
  X,
  y,
  n,
  mu0,
  Sigma0,
  mu_init,
  Sigma_init,
  tol = 1e-08,
  maxiter = 1000L,
  maxiter_jj = 25L,
  alg = "jj",
  verbose = FALSE
)

Arguments

X: The design matrix
y: The response vector
n: The trial vector
mu0: The prior mean for beta paramter
Sigma0: The prior variance for beta parameter
mu_init: Initial value for mu for optimisation.
Sigma_init: Initial value for Sigma for optimisation.
tol: The tolerance level to assess convergence
maxiter: The maximum number of iterations
maxiter_jj: The maximum number of Jaakkola-Jordan iterations to initialise estimation
alg: The algorithm used for final estimation of variational parameters. Must be one of jj, sj, or kmw.
verbose: Print trace of the lower bound to console. Default is FALSE.

Value

A list containing:

converged: Indicator for algorithm convergence.
jj_converged: Indicator for convergence of initial Jaakkola-Jordan iterations.
elbo: Vector of the ELBO sequence.
mu: The optimised value of mu.
Sigma: The optimised value of Sigma.

Details

By default, the algorithm always intialises with Jaakkola-Jordan updates until convergence or maxiter_jj.