Struct FlowVi 

pub struct FlowVi {
    pub config: FlowViConfig,
}

Flow-enhanced Variational Inference

Uses a normalizing flow on top of a mean-field Gaussian base distribution to produce a more flexible posterior approximation.

The ELBO becomes:

ELBO = E_{z_0 ~ q_0} [log p(x, z_K) - log q_0(z_0) + sum_k log|det J_k|]

where z_K = f_K . … . f_1(z_0) and q_0 = N(mu, diag(sigma^2)).

Fields

config: FlowViConfig

Configuration

Implementations

impl FlowVi

pub fn new(config: FlowViConfig) -> Self

Create a new flow-enhanced VI instance

Trait Implementations

impl Clone for FlowVi

fn clone(&self) -> FlowVi

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for FlowVi

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

Formats the value using the given formatter.

impl VariationalInference for FlowVi

fn fit<F>(&mut self, log_joint: F, dim: usize) -> StatsResult<PosteriorResult>

where F: Fn(&Array1<f64>) -> StatsResult<(f64, Array1<f64>)>,

Fit the variational approximation to a target log-joint distribution.