stouputils.data_science.models.keras_utils.losses.next_generation_loss module#

class NextGenerationLoss(alpha: float = 2.4092, name: str = 'ngl_loss')[source]#

Bases: Loss

Next Generation Loss with alpha = 2.4092.

Sources: - Code: ZKI-PH-ImageAnalysis/Next-Generation-Loss - Next Generation Loss Function for Image Classification: https://arxiv.org/pdf/2404.12948

name: str#

The name of the loss function.

alpha: float#

The alpha parameter.

call(y_true: Tensor, y_pred: Tensor) → Tensor[source]#

Compute the NGL loss.

Parameters:

  • y_true (tf.Tensor) – The true labels.

  • y_pred (tf.Tensor) – The predicted labels.

Returns:

The computed NGL loss.

Return type:

tf.Tensor