Module loss 

Loss functions for training machine learning models.

Usage

use aprender::loss::{mse_loss, mae_loss, huber_loss};
use aprender::primitives::Vector;

let y_true = Vector::from_slice(&[1.0, 2.0, 3.0]);
let y_pred = Vector::from_slice(&[1.1, 2.1, 2.9]);

let mse = mse_loss(&y_pred, &y_true);
let mae = mae_loss(&y_pred, &y_true);
let huber = huber_loss(&y_pred, &y_true, 1.0);

Structs

CTCLoss

Connectionist Temporal Classification (CTC) Loss.

DiceLoss

Dice loss struct wrapper.

FocalLoss

Focal loss function (struct wrapper).

HingeLoss

Hinge loss struct wrapper.

HuberLoss

Huber loss function (struct wrapper).

InfoNCELoss

InfoNCE / NT-Xent loss function (struct wrapper).

MAELoss

Mean Absolute Error loss function (struct wrapper).

MSELoss

Mean Squared Error loss function (struct wrapper).

TripletLoss

Triplet loss function (struct wrapper).

WassersteinLoss

Wasserstein Loss struct wrapper.

Traits

Loss

Trait for loss functions.

Functions

cross_entropy_loss

Cross-entropy loss with one-hot (soft) targets.

dice_loss

Dice loss for segmentation tasks.

focal_loss

Focal loss for class imbalance (spec: more-learning-specs.md §18).

gradient_penalty

Gradient penalty for WGAN-GP. Enforces Lipschitz constraint via gradient norm penalty.

hinge_loss

Hinge loss for SVM-style margin classification.

huber_loss

Huber loss (smooth approximation of MAE).

info_nce_loss

InfoNCE (Noise Contrastive Estimation) loss for contrastive learning.

kl_divergence

KL Divergence loss between two probability distributions.

mae_loss

Mean Absolute Error (MAE) loss.

mse_loss

Mean Squared Error (MSE) loss.

squared_hinge_loss

Squared hinge loss (smoother gradient).

triplet_loss

Triplet loss for metric learning.

wasserstein_discriminator_loss

Wasserstein loss for discriminator (critic). Maximizes distance between real and fake.

wasserstein_generator_loss

Wasserstein loss for generator. Minimizes negative fake score.

wasserstein_loss

Wasserstein (Earth Mover’s) Distance Loss.