Variational and Monte Carlo Inference for Bayesian State-Space Models of Count Data

Bayesian inference for a general class of state-space models for count data, providing a unified interface to variational Bayes and Markov chain Monte Carlo. Models are assembled from modular components for the observation distribution, link function, gain function, lag distribution, system equation, and system error, and support Poisson and negative binomial observations together with seasonality terms. The model class and the hybrid variational approximation are described in Tang and Prado (2026) .

dgtf

Variational and MCMC inference for Bayesian state-space models of count data.

dgtf provides a unified R interface to variational Bayes and MCMC for a general class of dynamic generalized transfer function models for count data.

Installation

1. Clone the repository from a terminal.

git clone [email protected]:minimeini/dgtf.git

1. Change into the repository with cd dgtf.
2. Install the R package.
   • In terminal: R CMD INSTALL .
   • Or in R: devtools::install(".")

System requirements

• C++17 compiler
• LAPACK and BLAS, both provided by R

The current release builds without OpenMP, so the parallel sections of the sequential Monte Carlo filter are evaluated serially, and no OpenMP runtime or omp.h header is needed.

R package dependencies

• LinkingTo: Rcpp, RcppArmadillo, BH, RcppProgress, pg.
• Imports: Rcpp, ggplot2 (plus base stats, utils, graphics, and grDevices).

License

MIT (see LICENSE).