Struct Huber 

pub struct Huber { /* private fields */ }

Huber’s loss with tuning k (default 1.345, ≈ 95% Gaussian efficiency). ρ(r) = r²/2 for |r| ≤ k, else k|r| − k²/2.

Implementations

impl Huber

pub fn new(k: f64) -> Result<Huber, RobustError>

Create a Huber loss with the given positive tuning constant.

Trait Implementations

impl Clone for Huber

fn clone(&self) -> Huber

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for Huber

impl Debug for Huber

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

Formats the value using the given formatter.

impl Default for Huber

fn default() -> Huber

k = 1.345.

impl RhoFunction for Huber

fn rho(&self, r: f64) -> f64

The loss ρ(r).

fn psi(&self, r: f64) -> f64

The score ψ(r) = ρ'(r): the (unstandardized) influence-function shape.

fn weight(&self, r: f64) -> f64

The IRLS weight w(r) = ψ(r)/r, with the removable singularity at r = 0 resolved by the limit ψ'(0).

fn psi_prime(&self, r: f64) -> f64

The derivative ψ'(r) = ρ''(r): the curvature of the loss and the shape of the E[ψ'] term in the asymptotic variance. (That E[ψ'] is computed via Stein’s identity E[X·ψ(X)], not by integrating this, so kinked scores stay exact; psi_prime is exposed for callers and completeness.)

fn tuning(&self) -> f64

The tuning constant controlling the efficiency/robustness trade-off (f64::NAN for losses without one, e.g. L1/least squares).

fn is_redescending(&self) -> bool

Whether ψ redescends to zero (⇒ ρ non-convex ⇒ needs a good start).

fn rho_sup(&self) -> Option<f64>

sup ρ = ρ(∞): Some(_) for bounded (redescending) losses, None for unbounded ones (Huber, least squares). Consumed by the S-scale.