Struct DirichletProcessMixtureModel 

pub struct DirichletProcessMixtureModel {
    pub alpha: f64,
    pub truncation: usize,
    pub max_iter: usize,
    pub tol: f64,
    pub activity_threshold: f64,
}

Dirichlet Process Mixture Model via truncated stick-breaking variational inference.

The model assumes a DP with concentration alpha, truncated at T components. Each component is a spherical Gaussian N(x | mu_k, sigma_k^2 I). Variational mean-field approximation is used, updating:

1. Component responsibilities (E-like step).
2. Posterior stick-breaking parameters (M-like step for the DP prior).
3. Posterior component parameters (M-like step for the Gaussian likelihood).

The number of effective components is inferred automatically from the data.

Fields

alpha: f64

DP concentration parameter (larger => more components).

truncation: usize

Truncation level.

max_iter: usize

Maximum variational EM iterations.

tol: f64

Convergence tolerance on the ELBO.

activity_threshold: f64

Minimum component weight threshold to declare a component active.

Implementations

impl DirichletProcessMixtureModel

pub fn new(alpha: f64, truncation: usize) -> Self

Create a new DPMM estimator.

pub fn fit(&self, data: ArrayView2<'_, f64>) -> Result<DpmmResult>

Fit the DPMM to data.