Loss Functions Parallel
Located in Gradien.NN.Losses. Loss functions measure how far the model's predictions are from the targets.
Regression
mse_backward
Mean Squared Error.
lua
(pred: Tensor, target: Tensor) -> (number, Tensor)l1_backward
Mean Absolute Error (L1).
lua
(pred: Tensor, target: Tensor) -> (number, Tensor)huber_backward
Less sensitive to outliers than MSE.
lua
(pred: Tensor, target: Tensor, delta: number?) -> (number, Tensor)Classification
cross_entropy_backward
Combines Softmax and Cross Entropy. Expects raw logits.
lua
(logits: Tensor, targetIndices: {number}, smoothing: number?) -> (number, Tensor)bceWithLogits_backward
Binary Cross Entropy with Sigmoid built-in.
lua
(logits: Tensor, targets: {number}) -> (number, Tensor)focal_backward
Focuses training on hard examples. Useful for class imbalance.
lua
(logits: Tensor, targetIdx: {number}, alpha: number?, gamma: number?) -> (number, Tensor)kl_div_backward
Kullback-Leibler divergence.
lua
(pred: Tensor, target: Tensor, fromLogits: boolean?) -> (number, Tensor)