bnns 1.0.0 is now on CRAN

I am happy to share that bnns version 1.0.0 is now available on CRAN:

https://cran.r-project.org/web/packages/bnns/index.html

CRAN lists the package publication date as June 7, 2026.

You can install it directly from CRAN:

install.packages("bnns")

What is `bnns`?

bnns is an R package for building Bayesian neural networks with a familiar formula interface. Under the hood, it uses Stan, so the model output is not just a point prediction: it carries posterior uncertainty with it.

That makes it useful when we care about both prediction and uncertainty quantification, especially in settings where decisions should not be based on a single fitted value alone.

What the CRAN release includes

The 1.0.0 release supports:

Formula-based model specification for Bayesian neural networks.
User-chosen priors for weights and biases.
Regression, binary classification, and multi-class classification workflows.
Posterior predictive summaries that make uncertainty easier to inspect.
A workflow designed for applied problems in clinical trials, finance, and other decision-heavy domains.

A small example

The basic workflow is intentionally close to standard R modeling:

library(bnns)

fit <- bnns(
  y ~ .,
  data = training_data,
  L = 2,
  nodes = c(16, 8),
  act_fn = c(4, 1),
  iter = 2000,
  warmup = 1000,
  chains = 4
)

pred <- predict(fit, newdata = test_data)

From there, predictions can be summarized using posterior medians, credible intervals, or any other posterior quantity that is useful for the analysis.

Why this matters

Neural networks are often treated as black-box prediction engines. Bayesian neural networks give us a richer object: a model that can learn non-linear relationships while still expressing uncertainty about its predictions.

That is the main motivation behind bnns: to make Bayesian neural networks more accessible to R users who want flexible modeling without giving up probabilistic interpretation.