Struct PolyLoss 

pub struct PolyLoss {
    pub epsilon: f64,
    pub poly_coeff: f64,
}

Poly Loss - Polynomial Expansion of Cross-Entropy Loss.

Paper: “PolyLoss: A Polynomial Expansion Perspective of Classification Loss Functions” (Leng et al., 2022) https://arxiv.org/abs/2204.12511

PolyLoss adds polynomial terms to cross-entropy to provide better gradient flow for well-classified examples. It helps with:

• Label noise robustness
• Improved generalization
• Better handling of class imbalance

The loss is defined as: L_poly = CE + ε₁(1 - p_t) + ε₂(1 - p_t)² + … + εⱼ(1 - p_t)^j

where p_t is the predicted probability of the target class, and εⱼ are polynomial coefficients. In practice, Poly-1 (j=1) is most commonly used.

Fields

epsilon: f64

Epsilon for numerical stability

poly_coeff: f64

Polynomial coefficient (typically between 0.5 and 2.0)

Implementations

impl PolyLoss

pub fn new(poly_coeff: f64) -> Self

Create a new Poly Loss with custom coefficient.

Trait Implementations

impl Clone for PolyLoss

fn clone(&self) -> PolyLoss

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for PolyLoss

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

Formats the value using the given formatter.

impl Default for PolyLoss

fn default() -> Self

Returns the “default value” for a type.

impl Loss for PolyLoss

fn compute( &self, predictions: &ArrayView<'_, f64, Ix2>, targets: &ArrayView<'_, f64, Ix2>, ) -> TrainResult<f64>

Compute loss value.

fn gradient( &self, predictions: &ArrayView<'_, f64, Ix2>, targets: &ArrayView<'_, f64, Ix2>, ) -> TrainResult<Array<f64, Ix2>>

Compute loss gradient with respect to predictions.

fn name(&self) -> &str

Get the name of the loss function.