12.5 Categorical Distribution
12.5.1 Probability Mass Functions
If \(N \in \mathbb{N}\), \(N > 0\), and if \(\theta \in \mathbb{R}^N\) forms an \(N\)-simplex (i.e., has nonnegative entries summing to one), then for \(y \in \{1,\ldots,N\}\), \[ \text{Categorical}(y~|~\theta) = \theta_y. \] In addition, Stan provides a log-odds scaled categorical distribution, \[ \text{CategoricalLogit}(y~|~\beta) = \text{Categorical}(y~|~\text{softmax}(\beta)). \] See section softmax for the definition of the softmax function.
12.5.2 Sampling Statement
y ~ categorical(theta)
Increment target log probability density with categorical_lpmf(y | theta)
dropping constant additive terms.
12.5.3 Sampling Statement
y ~ categorical_logit(beta)
Increment target log probability density with categorical_logit_lpmf(y | beta)
dropping constant additive terms.
12.5.4 Stan Functions
All of the categorical distributions are vectorized so that the
outcome y can be a single integer (type int) or an array of integers
(type int[]).
real categorical_lpmf(ints y | vector theta)
The log categorical probability mass function with outcome(s) y in
\(1:N\) given \(N\)-vector of outcome probabilities theta. The parameter
theta must have non-negative entries that sum to one, but it need not
be a variable declared as a simplex.
real categorical_logit_lpmf(ints y | vector beta)
The log categorical probability mass function with outcome(s) y in
\(1:N\) given log-odds of outcomes beta.
int categorical_rng(vector theta)
Generate a categorical variate with \(N\)-simplex distribution parameter
theta; may only be used in transformed data and generated quantities blocks
int categorical_logit_rng(vector beta)
Generate a categorical variate with outcome in range \(1:N\) from
log-odds vector beta; may only be used in transformed data and generated
quantities blocks