Struct DirichletRegression 

pub struct DirichletRegression {
    pub coefficients: Vec<f64>,
    pub precision: f64,
    pub n_parts: usize,
    pub n_covariates: usize,
    pub log_likelihood: f64,
}

A Dirichlet GLM fitted by Iteratively Reweighted Least Squares (IRLS).

The model is: ln(E[yⱼ]) = Xβⱼ + offset, with y ~ Dir(φ · μ) where φ (precision) is estimated jointly.

For simplicity this implementation uses a reduced model:

• Intercept-only mean model for each part
• Single precision parameter φ estimated from variance

For a full covariate model, use DirichletRegression which supports design matrices.

Fields

coefficients: Vec<f64>

Intercept coefficients for each part (on log-ratio / softmax scale).

precision: f64

Precision parameter φ = Σ αⱼ > 0.

n_parts: usize

Number of parts D.

n_covariates: usize

Number of covariates (including intercept).

log_likelihood: f64

Log-likelihood at fitted parameters.

Implementations

impl DirichletRegression

pub fn fit(responses: &[Vec<f64>], covariates: &[Vec<f64>]) -> StatsResult<Self>

Fit a Dirichlet regression model to compositional response data.

responses: N × D matrix (as Vec<Vec<f64>>) of compositional observations. covariates: N × P matrix of covariates (each row is one observation’s features). Pass an empty inner vec or &[] to fit an intercept-only model.

Uses IRLS to maximise the Dirichlet log-likelihood.

Errors

Returns an error if:

• responses is empty or has inconsistent dimensions.
• Any response row contains non-positive components.
• The algorithm encounters a numerical degeneracy.

Examples

use scirs2_stats::compositional::DirichletRegression;
let responses = vec![
    vec![0.3, 0.4, 0.3],
    vec![0.2, 0.5, 0.3],
    vec![0.4, 0.3, 0.3],
    vec![0.25, 0.35, 0.4],
];
let covariates: Vec<Vec<f64>> = vec![vec![]; 4]; // intercept-only
let model = DirichletRegression::fit(&responses, &covariates).unwrap();
assert_eq!(model.n_parts, 3);
assert!(model.precision > 0.0);

pub fn predict(&self, x: &[f64]) -> StatsResult<Vec<f64>>

Predict the expected composition for a new covariate vector x.

Returns a closed composition summing to 1.

Errors

Returns an error if x has the wrong length.

Examples

use scirs2_stats::compositional::DirichletRegression;
let responses = vec![
    vec![0.3, 0.4, 0.3],
    vec![0.2, 0.5, 0.3],
    vec![0.4, 0.3, 0.3],
    vec![0.25, 0.35, 0.4],
];
let covariates: Vec<Vec<f64>> = vec![vec![]; 4];
let model = DirichletRegression::fit(&responses, &covariates).unwrap();
let pred = model.predict(&[]).unwrap();
let sum: f64 = pred.iter().sum();
assert!((sum - 1.0).abs() < 1e-12);

Trait Implementations

impl Clone for DirichletRegression

fn clone(&self) -> DirichletRegression

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for DirichletRegression

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Display for DirichletRegression

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.