Struct ExponentialFamily 

pub struct ExponentialFamily {
    pub sufficient_stats: Vec<fn(&[f64]) -> f64>,
    pub log_partition: fn(&[f64]) -> f64,
}

An exponential family of distributions.

p(x; θ) = exp(θ · T(x) − A(θ)) where T are sufficient statistics and A is the log-partition (cumulant generating) function.

Fields

sufficient_stats: Vec<fn(&[f64]) -> f64>

Sufficient statistic functions T_i(x).

log_partition: fn(&[f64]) -> f64

Log-partition function A(θ).

Implementations

impl ExponentialFamily

pub fn new( sufficient_stats: Vec<fn(&[f64]) -> f64>, log_partition: fn(&[f64]) -> f64, ) -> Self

Create a new ExponentialFamily.

pub fn natural_params(&self, theta: &[f64]) -> Vec<f64>

Evaluate the natural parameters at theta (identity for canonical form).

pub fn moment_params(&self, theta: &[f64]) -> Vec<f64>

Compute the moment parameters μ_i = ∂A/∂θ_i by finite difference.

pub fn kl_divergence(&self, theta1: &[f64], theta2: &[f64]) -> f64

KL divergence KL(p_{θ1} ‖ p_{θ2}) = A(θ2) − A(θ1) − (θ2−θ1)·μ1.

pub fn fisher_info(&self, theta: &[f64]) -> Vec<Vec<f64>>

Fisher information matrix I(θ) = ∂²A/∂θ_i ∂θ_j (Hessian of A).