gibbs_sampler: Gibbs two-step sampling procedure for parameter vector beta...

In kasparmartens/PolyaGamma: Bayesian Logistic Regression with Polya-Gamma latent variables

Description

Gibbs two-step sampling procedure for parameter vector beta and the latent Polya-Gamma variables

Usage

gibbs_sampler(y, X, lambda = 1e-04, b = rep(0, ncol(X)), B = lambda *
  diag(ncol(X)), n_iter_total = 200, burn_in = 100, naive = FALSE,
  naive_n_terms = 100, t = 0.64)

Arguments

y	binary vector of observations
X	design matrix of covariates
lambda	The diagonal element of the precision matrix of the prior distribution (see also parameter B)
b	The prior mean of the prior distribution. Defaults to a vector of zeros.
B	The prior precision of the prior distribution. Defaults to lambda * identity.
naive	Should the naive approximation be used to generate the Polya-Gamma distribution
naive_n_terms	If the naive approximation is used, then this specifies number of terms in the finite sum.
t	The parameter in the accept-reject algorithm for sampling from Polya-Gamma distribution (see paper for details).
n_iter	The total number of iterations in the MCMC chain

Value

list containing the MCMC samples from the posterior distribution of beta

Examples

data = generate_from_simple_logistic_model(n=100)
obj = gibbs_sampler(data$y, data$X, lambda=0.001, n_iter_total=100, burn_in=50)
plot(obj)

kasparmartens/PolyaGamma documentation built on May 20, 2019, 7:23 a.m.