BayesPprobit

Overview

BayesPprobit is an advanced R package that implements Bayesian p-generalized probit regression models, incorporating efficient data compression techniques through the use of coresets. This package is designed to handle large-scale binary classification problems by providing scalable and efficient Bayesian inference methods.

Key features of BayesPprobit include:

• Flexible Modeling: Supports the p-generalized probit model, allowing for robustness to outliers and heavy-tailed data by adjusting the shape parameter ( p ).
• Efficient MCMC Sampling: Implements Metropolis-Hastings within Gibbs sampling for posterior inference, optimized for performance.
• Coreset Integration: Incorporates coreset construction methods to reduce data size while preserving essential statistical properties, enabling scalable inference on massive datasets.
• Diagnostic Tools: Provides functions for convergence diagnostics, posterior summary statistics, and visualization of MCMC results.
• Easy-to-Use Interface: Designed with a user-friendly API that integrates seamlessly with existing R workflows.

Installation

You can install the stable version of BayesPprobit from CRAN:

install.packages("BayesPprobit")

Or install the development version from GitHub to access the latest features and updates:

# Install devtools if you haven't already
install.packages("devtools")

# Install BayesPprobit from GitHub
devtools::install_github("zeyudsai/BayesPprobit")

Usage

Below is a basic example demonstrating how to use BayesPprobit for Bayesian p-generalized probit regression:

library(BayesPprobit)

# Simulate some data
set.seed(123)
n <- 1000  # Number of observations
d <- 5     # Number of predictors
X <- matrix(rnorm(n * d), n, d)
beta_true <- c(1.5, -0.8, 0.6, -0.4, 0.2)
p_true <- 2
eta <- X %*% beta_true
alpha <- p_scale(p_true)
prob <- gnorm::pgnorm(eta, mu = 0, alpha = alpha, beta = p_true)
y <- rbinom(n, 1, prob)

# Fit the Bayesian p-generalized probit model
fit <- multi_chain(
  n_sim = 5000,
  burn_in = 1000,
  X = X,
  y = y,
  initial_theta = rep(0, d),
  initial_p = 2,
  mh_iter = 100,
  p_range = c(1, 3),
  step_size = 0.05,
  n_chains = 3
)

# Print a summary of the results
print(fit)

# Plot diagnostic plots
plot(fit)

For more detailed examples and advanced usage, please refer to the package vignette:

vignette("BayesPprobit")

Features

• Bayesian Inference for Probit Models: Implements a Bayesian approach to probit regression, allowing uncertainty quantification and incorporation of prior information.
• p-Generalized Distribution: Utilizes the p-generalized normal distribution to model the latent variables, offering flexibility in capturing data characteristics.
• Metropolis-Hastings within Gibbs Sampling: Employs efficient MCMC algorithms for sampling from complex posterior distributions.
• Coreset Methods: Supports various coreset construction techniques such as leverage score sampling, uniform sampling, and oneshot methods to handle large datasets.
• Convergence Diagnostics: Integrates with the coda package for convergence assessment using Gelman-Rubin diagnostics and other MCMC diagnostic tools.
• Visualization: Provides functions for trace plots, posterior density plots, and comparison of true vs. estimated parameters.

Documentation

Comprehensive documentation is available for all functions within the package. Access the documentation using the standard R help system:

# For a general overview
?BayesPprobit

# For specific functions
?multi_chain
?gibbs_sampler

Support and Contributions

If you encounter any issues, have questions, or would like to contribute to the development of BayesPprobit, please visit our GitHub repository:

GitHub - zeyudsai/BayesPprobit

Feel free to open issues or submit pull requests. Contributions are welcome!

Citation

If you use BayesPprobit in your research or publications, please cite it as follows:

citation("BayesPprobit")

You can also visit CITATION This will provide the appropriate citation information, including authors and version number.

License

BayesPprobit is licensed under the MIT License. See the LICENSE file for more details.

Acknowledgments

We would like to thank all contributors and users who have provided feedback and suggestions to improve the package.

---

This README was generated to provide a comprehensive overview of the BayesPprobit package, its features, and how to get started. For more detailed information, please refer to the package documentation and vignettes.

---