Losses
-
Loss *eddl::getLoss(string type)
Get Loss object by its name.
- Parameters
type – Loss name/type
- Returns
Selected Loss
Mean Squared Error
Aliases: mean_squared_error and mse.
Creates a criterion that measures the mean squared error (squared L2 norm) between each element in the input
Example:
myloss = getLoss("mean_squared_error");
Binary Cross-Entropy
Aliases: binary_cross_entropy, and bce.
Creates a criterion that measures the Binary Cross Entropy between the target and the output. Values are encoded as the probability of the positive class.
Example:
myloss = getLoss("binary_cross_entropy");
Categorical Cross-Entropy
Aliases: categorical_cross_entropy, cce, cross_entropy, and ce.
Creates a criterion that measures the Categorical Cross Entropy between the target and the output. Values are encoded as vector of probabilities that sum is equal to one.
Example:
myloss = getLoss("categorical_cross_entropy");
Softmax Cross-Entropy
Aliases: softmax_cross_entropy, soft_cross_entropy, and sce.
This is the optimized version of the CategoricalCrossEntropy when the last layer is a Softmax layer. It bypasses the
backward of the Softmax layer by applying the simplified derivative of f(x)=CE(Softmax(x)); df/dx= y_pred-y_target.
Example:
myloss = getLoss("softmax_cross_entropy");