Struct CategoricalLikelihood

Source

pub struct CategoricalLikelihood {
    pub number_of_categories: i64,
    pub function_name: String,
    pub description: String,
    pub formula: String,
}

Expand description

Likelihood for categorical outcomes.

Fields§

§number_of_categories: i64

Number of categories.

§function_name: String

Unique name identifying a likelihood function.

§description: String

Free-text description.

§formula: String

Mathematical formula.

Implementations§

Source §

impl CategoricalLikelihood

Source

pub fn log_likelihood( &self, n_categories: usize, params: &[f64], data: &[f64], ) -> f64

Evaluates the categorical log-likelihood Σⱼ countⱼ · ln pⱼ of data under the probability vector params.

The number of categories k is passed explicitly rather than read from the struct (see the module docs); the frozen LogLikelihood trait is instead implemented on CategoricalLikelihoodModel.

§Arguments

n_categories — the number of categories k.
params — the probability vector [p₀ … p_{k−1}]; length must equal n_categories and the entries must sum to 1 within SIMPLEX_TOLERANCE.
data — the observed category indices, each a non-negative integer below k stored as f64.

§Returns

The log-likelihood, or f64::NEG_INFINITY when params has the wrong length, is not normalized, assigns a non-positive probability to a non-empty category, or data contains an invalid index. A category with zero observed count contributes 0 (the 0 · ln 0 ≔ 0 convention).

§Examples

use stats_claw::likelihood::CategoricalLikelihood;

let model = CategoricalLikelihood::default();
// counts = [2, 1, 3] over k = 3; ℓ = 2·ln0.2 + 1·ln0.3 + 3·ln0.5.
let ll = model.log_likelihood(3, &[0.2, 0.3, 0.5], &[0.0, 0.0, 1.0, 2.0, 2.0, 2.0]);
assert!((ll - (-6.502_290_170_873_972)).abs() < 1e-10, "ll was {ll}");

Source

pub fn fit(&self, n_categories: usize, data: &[f64]) -> Result<MleFit>

Closed-form maximum-likelihood fit of the category probabilities, p̂ⱼ = countⱼ / n.

§Arguments

n_categories — the number of categories k.
data — the observed category indices, each a non-negative integer below k stored as f64.

§Returns

An MleFit whose parameters are the empirical frequencies (length n_categories, summing to 1) and whose log-likelihood is ℓ(p̂; data). Categories with zero observed count are allowed and receive p̂ⱼ = 0, contributing 0 to the log-likelihood.

§Errors

Error::InsufficientData if data is empty.
Error::InvalidInput if any datum is not a non-negative integer below n_categories.

§Examples

use stats_claw::likelihood::CategoricalLikelihood;

let model = CategoricalLikelihood::default();
let fit = model.fit(3, &[0.0, 0.0, 1.0, 2.0, 2.0, 2.0])?;
assert!((fit.params()[2] - 0.5).abs() < 1e-12, "p2 was {}", fit.params()[2]);