bf.dist.multinomial_logits: Multinomial logit

In BayesForge: Bayesian Inference using 'numpyro' and 'XLA'

Multinomial logit

Description

A *multinomial logits* distribution refers to a categorical (or more generally multinomial) distribution over K classes whose probabilities are given via the softmax of a vector of logits. That is, given a vector of real-valued logits \ell = (\ell_1, \dots, \ell_K), the class probabilities are:

p_k = \frac{\exp(\ell_k)}{\sum_{j=1}^K \exp(\ell_j)}.

Then a single draw from the distribution yields one of the K classes (or for a multinomial count version, counts over the classes) with those probabilities.#' @export

Usage

bf.dist.multinomial_logits(
  logits,
  total_count = 1,
  total_count_max = py_none(),
  validate_args = py_none(),
  name = "x",
  obs = py_none(),
  mask = py_none(),
  sample = FALSE,
  seed = py_none(),
  shape = c(),
  event = 0,
  create_obj = FALSE,
  to_jax = TRUE
)

Arguments

logits	A numeric vector, matrix, or array representing the logits for each outcome.
total_count	A numeric vector, matrix, or array representing the total number of trials.
total_count_max	(int, optional): An optional integer providing an upper bound on 'total_count'. This is used for performance optimization with 'lax.scan' when 'total_count' is a dynamic JAX tracer, helping to avoid recompilation.
validate_args	Logical: Whether to validate parameter values. Defaults to 'reticulate::py_none()'.
name	A character string representing the name of the random variable within a model. This is used to uniquely identify the variable. Defaults to 'x'.
obs	A numeric vector or array of observed values. If provided, the random variable is conditioned on these values. If 'NULL', the variable is treated as a latent (unobserved) variable. Defaults to 'NULL'.
mask	A logical vector, matrix, or array to mask observations.
sample	A logical value that controls the function's behavior. If 'TRUE', the function will directly draw samples from the distribution. If 'FALSE', it will create a random variable within a model. Defaults to 'FALSE'.
seed	An integer used to set the random seed for reproducibility when 'sample = TRUE'. This argument has no effect when 'sample = FALSE', as randomness is handled by the model's inference engine. Defaults to 0.
shape	A numeric vector specifying the shape of the distribution. Use a vector (e.g., 'c(10)') to define the shape.
event	Integer specifying the number of batch dimensions to reinterpret as event dimensions (used in model building).
create_obj	Logical; If TRUE, returns the raw BI distribution object instead of creating a sample site.
to_jax	Boolean. Indicates whether to return a JAX array or not.

Value

- When sample=FALSE, a BI MultinomialLogits distribution object (for model building).

- When sample=TRUE, a JAX array of samples drawn from the MultinomialLogits distribution (for direct sampling).

- When create_obj=TRUE, the raw BI distribution object (for advanced use cases).

See Also

This is a wrapper of https://num.pyro.ai/en/stable/distributions.html#multinomiallogits

Examples

library(BayesForge)
m=importBF(platform='cpu')
bf.dist.multinomial_logits(logits =  c(0.2, 0.3, 0.5), total_count = 10, sample = TRUE)

BayesForge documentation built on June 9, 2026, 1:09 a.m.