bmbeR: Bayesian Model Building and Evaluation Repository

This repository contains a set of R scripts designed to build, evaluate, visualize, and perform sensitivity analysis on Bayesian models. These scripts use a mixture of rstan, rstanarm, and other Bayesian analysis libraries to facilitate the modeling process.

Scripts and Their Functions:

libraries_and_setup.R

• Purpose: Loads the necessary libraries and global settings for the entire modeling workflow.
• Libraries Used:
  • rstanarm
  • bayesplot
  • ggplot2
  • cowplot
  • purrr
  • rstan
  • loo

distributions.R

• Purpose: Defines a comprehensive set of prior distribution generators.
• Functions:
  • student_t_prior
  • normal_prior
  • cauchy_prior
  • uniform_prior
  • beta_prior
  • gamma_prior
  • binomial_prior
  • poisson_prior
  • lognormal_prior
  • bernoulli_prior
• Global Objects and Utilities:
  • original_distributions, distributions
  • add_distribution, reset_distributions, get_prior_distribution

utilities.R

• Purpose: Provides utility functions used across the repository.
• Functions:
  • validate_positive_integer
  • validate_numeric
  • %||% (null-coalescing operator)

empirical_bayes.R

• Purpose: Computes empirical Bayesian hyperparameter specifications from data.
• Functions:
  • empirical_bayes_priors: Fits a linear model and maps coefficient estimates and standard errors to hyperparameters for various prior distributions.

model_convergence.R

• Purpose: Checks convergence diagnostics of a fitted Bayesian model.
• Functions:
  • check_convergence: Assesses if a model has converged based on Rhat and effective sample size (ESS) metrics, and optionally generates trace plots.

model_fitting.R

• Purpose: Fits Bayesian models using specified prior configurations and data.
• Functions:
  • build_stanarm_priors: Converts hyperparameter lists into rstanarm prior objects.
  • fit_model_with_prior: Fits a Bayesian model using rstanarm::stan_glm with given priors, and checks for convergence.

model_visualization.R

• Purpose: Generates diagnostic and posterior visualization plots for fitted models.
• Functions:
  • generate_plot: Generates diagnostic plots (trace, histogram, density, and autocorrelation) for a model.
  • plot_posterior_distributions: Visualizes posterior distributions with 95% credible intervals using bayesplot.

model_sensitivity.R

• Purpose: Performs sensitivity analysis across different prior configurations.
• Functions:
  • sensitivity_analysis: Iterates through a list of prior configurations, fits models, and computes performance metrics (using LOO) to assess how changes in priors affect model outcomes.

model_evaluation.R

• Purpose: Evaluates the predictive performance of fitted models.
• Functions:
  • evaluate_model_performance: Computes performance metrics such as RMSE and MAE for regression tasks, or accuracy, precision, recall, and F1 Score for classification tasks.