Crate robust_rs_core 

Shared spine of the robust-rs project: the M-estimator loss trait, robust scale estimation and the influence-function / efficiency / breakdown-point theory. Dependency-light (no linear algebra; only libm, num-traits, thiserror), so it builds for wasm32 and can be depended on for just the losses and theory. The estimators live in the robust-rs crate.

Modules

• rho: the organizing RhoFunction trait (ρ, ψ, weight, ψ', tuning, is_redescending, rho_sup) and its implementations: LeastSquares, L1, Huber, TukeyBiweight, Cauchy, Welsch, Andrews, Hampel.
• scale: the ScaleEstimator trait and the robust scales Mad, HuberProposal2, SScale, Qn, Sn.
• theory: influence_function, asymptotic_variance, gaussian_efficiency, breakdown_point and the gauss_hermite quadrature underneath them.
• solver: the Control stopping rule shared by the iterative estimators.
• types: the Scale, TuningConstant, RawResidual / ScaledResidual newtypes.
• error: the crate-wide RobustError.

use robust_rs_core::rho::{Huber, RhoFunction};
use robust_rs_core::theory::gaussian_efficiency;

let huber = Huber::default();          // k = 1.345
assert_eq!(huber.psi(10.0), 1.345);    // clipped score: bounded influence
assert!((gaussian_efficiency(&huber, 128) - 0.95).abs() < 0.01);

Modules

error

Error type for the robust-rs-core crate.

rho

The RhoFunction trait: the organizing abstraction of the crate and its standard implementations.

scale

The ScaleEstimator trait and standard robust scale estimates.

solver

Iterative solver utilities shared across estimators.

theory

Sampling-theory computations: influence function, asymptotic variance, Gaussian efficiency and breakdown point.

types

Strongly-typed wrappers that make illegal states unrepresentable and prevent conflating a raw residual with a scaled one.