Struct StochasticVI 

pub struct StochasticVI {
    pub variational: MeanFieldGaussian,
    pub config: SviConfig,
    pub diagnostics: VariationalDiagnostics,
    /* private fields */
}

Stochastic Variational Inference (SVI)

Implements scalable variational inference using stochastic gradient ascent on the ELBO. Supports:

• Mini-batch ELBO estimation for large datasets
• Natural gradient updates for faster convergence
• Adam adaptive learning rates
• Gradient clipping for stability

The model assumes a mean-field Gaussian posterior approximation q(z) = prod_i N(z_i; mu_i, sigma_i^2)

The user provides a log joint density function log p(z, x_batch) that takes the latent variables z and a mini-batch of data.

Fields

variational: MeanFieldGaussian

Variational distribution (mean-field Gaussian)

config: SviConfig

Configuration

diagnostics: VariationalDiagnostics

Diagnostics

Implementations

impl StochasticVI

pub fn new(dim: usize, config: SviConfig) -> Result<Self>

Create a new SVI instance

pub fn fit<F>( &mut self, data: ArrayView2<'_, f64>, log_joint: F, n_total: usize, ) -> Result<SviResult>

where F: Fn(&Array1<f64>, ArrayView2<'_, f64>) -> Result<(f64, Array1<f64>)>,

Run SVI optimization with a log joint density function

Arguments

• data - Full dataset (rows are observations)
• log_joint - Function computing log p(z, x_batch) given latent variables and a batch of data. Returns (log_prob, gradient_wrt_z).
• n_total - Total number of data points (for scaling mini-batch ELBO)

Returns

• The optimized variational distribution and diagnostics

pub fn variational_distribution(&self) -> &MeanFieldGaussian

Get the current variational distribution

pub fn diagnostics(&self) -> &VariationalDiagnostics

Get diagnostics

pub fn reset_optimizer(&mut self)

Reset the optimizer state (useful for warm restarts)

Trait Implementations

impl Clone for StochasticVI

fn clone(&self) -> StochasticVI

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for StochasticVI

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

Formats the value using the given formatter.