skainet-lang-core/sk.ainet.java/Losses

Losses

Java-friendly factory for loss functions.

Example usage from Java:

Loss loss = Losses.crossEntropy();
Loss mse = Losses.mse();
Loss bce = Losses.binaryCrossEntropy(1e-7f);

Functions

bceWithLogits

@JvmStatic

fun bceWithLogits(): Loss

Binary cross-entropy with logits (numerically stable).

binaryCrossEntropy

@JvmStatic

@JvmOverloads

fun binaryCrossEntropy(epsilon: Float = 1.0E-7f): Loss

Binary cross-entropy loss (predictions should be probabilities in 0, 1).

categoricalCrossEntropy

@JvmStatic

@JvmOverloads

fun categoricalCrossEntropy(dim: Int = -1): Loss

Categorical cross-entropy (alias for crossEntropy).

crossEntropy

@JvmStatic

@JvmOverloads

fun crossEntropy(dim: Int = -1): Loss

Cross-entropy loss (combines log-softmax with NLL).

hinge

@JvmStatic

@JvmOverloads

fun hinge(margin: Float = 1.0f): Loss

Hinge loss for SVM-style classification.

huber

@JvmStatic

@JvmOverloads

fun huber(delta: Float = 1.0f): Loss

Huber (Smooth L1) loss — quadratic for small errors, linear for large.

mae

@JvmStatic

fun mae(): Loss

Mean Absolute Error loss.

mse

@JvmStatic

fun mse(): Loss

Mean Squared Error loss.

poisson

@JvmStatic

@JvmOverloads

fun poisson(logInput: Boolean = true, epsilon: Float = 1.0E-8f): Loss

Poisson negative log-likelihood loss for count data.

sparseCategoricalCrossEntropy

@JvmStatic

@JvmOverloads

fun sparseCategoricalCrossEntropy(dim: Int = -1): Loss

Sparse categorical cross-entropy (integer target indices).

squaredHinge

@JvmStatic

@JvmOverloads

fun squaredHinge(margin: Float = 1.0f): Loss

Squared hinge loss.