bayesfn

Bayesian methods for parameter estimation of the folded normal distribution and two-sample mean testing, with maximum likelihood estimation (MLE) provided as a baseline for comparison.

Features

• Bayesian estimation: Estimate mu and sigma of the folded normal via Stan (fn_bayes_est()).
• Bayesian two-sample mean test: Compare two folded-normal means using posterior probabilities (fn_ttest_bayes()).
• MLE baseline: Fast maximum likelihood estimator for mu and sigma (fn_mle()).

Installation

This package depends on rstan and a working C++14 toolchain.

• From a local checkout:

# From the project root
install.packages("devtools") # if not installed
devtools::install_local(".")
# or during development
devtools::load_all(".")

System requirements (Windows)

• R and RTools with C++14 support
• rstan configured (see the RStan installation guide)

Quick start

1) Simulate folded-normal data

set.seed(1)
mu    <- 1.0
sigma <- 0.5
x <- abs(rnorm(200, mean = mu, sd = sigma))

2) MLE baseline

library(bayesfn)
mle_fit <- fn_mle(x)
mle_fit$mu     # MLE for mu
mle_fit$sigma  # MLE for sigma

3) Bayesian estimation

# By default uses a cached Stan model; adjust iteration settings as needed
bayes_fit <- fn_bayes_est(x, iter = 2000, warmup = 1000, chains = 4)

bayes_fit$mu            # posterior mean of mu
bayes_fit$sigma         # posterior mean of sigma
bayes_fit$Rhat_mu       # R-hat diagnostic for mu
bayes_fit$Rhat_sigma    # R-hat diagnostic for sigma
length(bayes_fit$mu_samples)     # number of posterior samples
length(bayes_fit$sigma_samples)  # number of posterior samples

4) Bayesian two-sample mean test

x1 <- abs(rnorm(150, mean = 0.8, sd = 0.4))
x2 <- abs(rnorm(150, mean = 1.1, sd = 0.5))

# Test H0: mu1 < mu2; returns posterior P(mu1 > mu2) and a decision
tt <- fn_ttest_bayes(x1, x2, confidence_level = 0.05, iter = 2000, warmup = 1000)

tt$p_value   # P(mu1 > mu2)
tt$decision  # 0 accept H0, 1 accept H1

Stan model compilation & caching

The package uses a cached compiled Stan model for speed.

• get_stan_model() automatically retrieves a cached model if available.
• If needed, compile and cache once:

library(bayesfn)
mdl <- compile_stan_model(stan_code)  # provided in the package

On first use, compilation may take some time depending on your toolchain.

License

GPL-3