Struct NormalizingFlowVI 

pub struct NormalizingFlowVI {
    pub base: MeanFieldGaussian,
    pub flows: Vec<FlowLayer>,
    pub dim: usize,
}

Normalizing flow variational family (placeholder/scaffold)

This provides a framework for flow-based variational inference where the posterior is represented as a composition of invertible transformations applied to a base distribution (typically a standard Gaussian).

q(z) = q_0(f^{-1}(z)) * |det(df^{-1}/dz)|

Currently supports:

• Planar flows: f(z) = z + u * h(w^T z + b)
• Radial flows: f(z) = z + beta * h(alpha, r)(z - z0)

Fields

base: MeanFieldGaussian

Base distribution (mean-field Gaussian)

flows: Vec<FlowLayer>

Flow layers

dim: usize

Dimensionality

Implementations

impl NormalizingFlowVI

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

Create a new normalizing flow VI with a base mean-field Gaussian

pub fn add_planar_flow( &mut self, u: Array1<f64>, w: Array1<f64>, b: f64, ) -> Result<()>

Add a planar flow layer

pub fn add_radial_flow( &mut self, z0: Array1<f64>, log_alpha: f64, beta: f64, ) -> Result<()>

Add a radial flow layer

pub fn transform(&self, z0: &Array1<f64>) -> Result<(Array1<f64>, f64)>

Transform a sample through all flow layers, returning the transformed sample and the sum of log-abs-det-Jacobians

pub fn sample(&self, epsilon: &Array1<f64>) -> Result<(Array1<f64>, f64)>

Sample from the flow-transformed distribution

pub fn n_flow_params(&self) -> usize

Get the number of flow parameters

Trait Implementations

impl Clone for NormalizingFlowVI

fn clone(&self) -> NormalizingFlowVI

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for NormalizingFlowVI

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

Formats the value using the given formatter.