Struct VariationalGaussianMixture 

pub struct VariationalGaussianMixture { /* private fields */ }

Variational Bayes Gaussian Mixture Model fitter (VBEM / VBGMM).

The algorithm performs coordinate-ascent variational inference on the mean-field factorisation

  q(pi, mu, z) = q(pi) * prod_k q(mu_k) * prod_n q(z_n)

updating the Dirichlet posterior q(pi), Gaussian posteriors q(mu_k), and categorical posteriors q(z_n) until the ELBO converges.

Example

use tensorlogic_quantrs_hooks::vmp::{VariationalGaussianMixture, VgmmConfig};

let config = VgmmConfig::new(2)
    .with_prior(0.0, 1e-3, 1.0)
    .with_observation_precision(1.0)
    .with_limits(100, 1e-6)
    .with_seed(42);

let vgmm = VariationalGaussianMixture::new(config).unwrap();
let data = vec![0.0, 0.1, 0.2, 10.0, 10.1, 10.2];
let result = vgmm.fit(&data).unwrap();
assert!(result.converged);

Implementations

impl VariationalGaussianMixture

pub fn new(config: VgmmConfig) -> Result<Self>

Construct a new fitter, validating the config immediately.

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

Run VBEM on data and return the variational posterior.

Errors

• PgmError::InvalidDistribution if data is empty or contains non-finite values.
• PgmError::ConvergenceFailure if the ELBO decreases by more than config.divergence_tolerance in a single iteration (numerical breakdown).

Trait Implementations

impl Clone for VariationalGaussianMixture

fn clone(&self) -> VariationalGaussianMixture

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for VariationalGaussianMixture

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

Formats the value using the given formatter.