MAELoss

Mean Absolute Error (L1) Loss.

MAE(pred, target) = |pred - target|

MAE is less sensitive to outliers compared to MSE because it doesn't square the error term. This makes it more robust for data with outliers.

Constructors

constructor()

open override fun <T : DType, V> forward(preds: Tensor<T, V>, targets: Tensor<out DType, *>, ctx: ExecutionContext, reduction: Reduction): Tensor<T, V>