Loss functions

enum elsa::ml::LossReduction

Reduction types for loss functions.

Values:

enumerator Sum

reduce loss by summing up across batches

enumerator SumOverBatchSize

reduce loss by summing up over batches

template<typename data_t = real_t>
class elsa::ml::Loss

Base class for all loss functions

Author

David Tellenbach

Public Functions

Loss() = default

default constructor

~Loss() = default

virtual destructor

data_t getLoss(const DataContainer<data_t> &x, const DataContainer<data_t> &y) const

Return

the loss between predictions x and labels y

data_t operator()(const DataContainer<data_t> &x, const DataContainer<data_t> &y)

Return

the loss between predictions x and labels y

DataContainer<data_t> getLossGradient(const DataContainer<data_t> &x, const DataContainer<data_t> &y) const

Return

the loss-gradient between predictions x and labels y

std::string getName() const

Return

the name of this loss function

template<typename data_t = real_t>
class elsa::ml::BinaryCrossentropy : public elsa::ml::Loss<real_t>

Computes the cross-entropy loss between true labels and predicted labels.

Use this cross-entropy loss when there are only two label classes (assumed to be 0 and 1). For each example, there should be a single floating-point value per prediction.

Author

David Tellenbach

Public Functions

BinaryCrossentropy(LossReduction reduction = LossReduction::SumOverBatchSize)

Construct a BinaryCrossEntropy loss by optionally specifying the reduction type.

template<typename data_t = real_t>
class elsa::ml::CategoricalCrossentropy : public elsa::ml::Loss<real_t>

Computes the crossentropy loss between the labels and predictions.

Use this crossentropy loss function when there are two or more label classes. We expect labels to be provided in a one-hot representation. If you don’t want to use one-hot encoded labels, please use

SparseCategoricalCrossentropy loss. There should be # classes floating point values per feature.

Author

David Tellenbach

Public Functions

CategoricalCrossentropy(LossReduction reduction = LossReduction::SumOverBatchSize)

Construct a CategoricalCrossentropy loss by optionally specifying the reduction type.

template<typename data_t = real_t>
class elsa::ml::SparseCategoricalCrossentropy : public elsa::ml::Loss<real_t>

Computes the crossentropy loss between the labels and predictions.

Use this crossentropy loss function when there are two or more label classes. If you want to provide labels using one-hot representation, please use

CategoricalCrossentropy loss. There should be # classes floating point values per feature for x and a single floating point value per feature for y.

Author

David Tellenbach

Public Functions

SparseCategoricalCrossentropy(LossReduction reduction = LossReduction::SumOverBatchSize)

Construct a SparseCategoricalCrossentropy loss by optionally specifying the reduction type.

template<typename data_t = real_t>
class elsa::ml::MeanSquaredError : public elsa::ml::Loss<real_t>

Computes the mean squared error between labels y and predictions x:

\[ \text{loss} = (y - x)^2 \]

Author

David Tellenbach

Public Functions

MeanSquaredError(LossReduction reduction = LossReduction::SumOverBatchSize)

Construct a MeanSquaredError loss by optionally specifying the reduction type.