logitPQbound

The repository provides the R functions implementing the novel piecewise quadratic bound for logistic log-likelihoods from the paper

• Anceschi, N., Castiglione, C., Rigon, T., Zanella, G., and Durante, D. (2025), Optimal and computationally tractable lower bounds for logistic log-likelihoods, arXiv:2410.10309.

The logitPQbound package can be installed by running the following R commands

# If the devtools R package is not already installed
# install.packages("devtools")

devtools::install_github("niccoloanceschi/logitPQbound")

Alternatively, if devtools::install_github() produces warnings or fails, the pak package can be used:

# If the pak R package is not already installed
# install.packages("pak")

pak::pak("niccoloanceschi/logitPQbound")

Structure

• src/ contains the Rcpp code scripts
• R/ contains the R source scripts
• man/ contains the documentation files for the exported R functions (generated via roxygen2)
• data/ contains example datasets included in the package (in .RData extension)
• tutorial/ contains scripts to reproduce the numerical results and figures from the main paper
• DESCRIPTION is the package metadata file specifying authors, maintainers, version, and dependencies
• NAMESPACE defines both the imported and exported functions
• LICENSE contains the licensing terms of the package (MIT license)
• .Rbuildignore specifies files and directories that should be excluded when building the package
• .gitignore specifies files and directories ignored by Git version control
• logitPQbound.Rproj is the RStudio project file for development and reproducibility

Requirements

Install packages:

install.packages(c("devtools", "pak", "RcppArmadillo", "Rmpfr", "Matrix", "sparseinv", "BayesLogit", "ggplot2", "dplyr"))