Struct HuberT 

pub struct HuberT<F>
where
    F: Float + 'static + Display,
{
    pub t: F,
    pub weight_func: fn(F, F) -> F,
}

Huber loss function and constants for robust regression

Fields

t: F

The parameter that controls the transition between squared and absolute error

weight_func: fn(F, F) -> F

The weight function used for iteratively reweighted least squares

Implementations

impl<F> HuberT<F>

where F: Float + 'static + Display,

pub fn new() -> Self

Create a new Huber T object with default parameters

The default tuning constant t=1.345 provides 95% efficiency compared to ordinary least squares when the errors are normally distributed.

pub fn with_t(t: F) -> Self

Create a new Huber T object with custom parameter

Arguments

• t - The tuning constant that controls the transition between L1 and L2 loss
  • Smaller values (e.g., 1.0) make the estimator more robust to outliers but less efficient
  • Larger values (e.g., 2.0) make the estimator more similar to ordinary least squares
  • The recommended range is 1.0 to 2.0

pub fn loss(&self, r: F) -> F

Calculate the Huber loss for a given residual

The Huber loss function is defined as:

• For |r| ≤ t: L(r) = 0.5 * r² (quadratic/L2 loss)
• For |r| > t: L(r) = t * |r| - 0.5 * t² (linear/L1 loss with an offset)

Arguments

• r - The residual value

Returns

The Huber loss value for the residual

Trait Implementations

impl<F> Default for HuberT<F>

where F: Float + 'static + Display,

fn default() -> Self

Returns the “default value” for a type.