Huber

• namespace: Rindow\NeuralNetworks\Losses
• classname: Huber

Huber loss function.

For each value x in x = trues - predicts, d = delta Specify the delta value and use the two expressions properly

• loss = 0.5 * x^2 if |x| <= d
• loss = 0.5 * d^2 + d * (|x| - d) if |x| > d

Methods

constructor

$builer->Huber(
    float $delta=1.0
)

You can create a BinaryCrossEntropy loss function instances with the Losses Builder.

Examples

$model->compile(
    loss:$nn->losses()->Huber(delta:0.8),
);