Cost function interface. More...
#include <shark/ObjectiveFunctions/AbstractCost.h>
Inheritance diagram for shark::AbstractCost< LabelT, OutputT >:
Collaboration diagram for shark::AbstractCost< LabelT, OutputT >:
Public Types | |
| enum | Feature { HAS_FIRST_DERIVATIVE = 1, IS_LOSS_FUNCTION = 4 } |
| list of features a cost function can have More... | |
| typedef OutputT | OutputType |
| typedef LabelT | LabelType |
| typedef Batch< OutputType >::type | BatchOutputType |
| typedef Batch< LabelType >::type | BatchLabelType |
| typedef TypedFlags< Feature > | Features |
| typedef TypedFeatureNotAvailableException < Feature > | FeatureNotAvailableException |
Public Member Functions | |
| virtual | ~AbstractCost () |
| const Features & | features () const |
| virtual void | updateFeatures () |
| bool | hasFirstDerivative () const |
| returns true when the first parameter derivative is implemented More... | |
| bool | isLossFunction () const |
| returns true when the cost function is in fact a loss function More... | |
| void | configure (PropertyTree const &node) |
| configures the loss using informations in the property tree More... | |
| virtual double | eval (Data< LabelType > const &targets, Data< OutputType > const &predictions) const =0 |
| virtual double | evalDerivative (Data< LabelType > const &targets, Data< OutputType > const &predictions, Data< OutputType > &gradient) const |
| evaluates the cost and the derivative w.r.t. the prediction More... | |
| double | operator() (Data< LabelType > const &targets, Data< OutputType > const &predictions) const |
| Public Member Functions inherited from shark::INameable | |
| virtual | ~INameable () |
| virtual std::string | name () const |
| returns the name of the object More... | |
| Public Member Functions inherited from shark::IConfigurable | |
| virtual | ~IConfigurable () |
| Virtual d'tor. More... | |
Protected Attributes | |
| Features | m_features |
Detailed Description
template<class LabelT, class OutputT = LabelT>
class shark::AbstractCost< LabelT, OutputT >
Cost function interface.
- In Shark a cost function encodes the severity of a deviation of predictions from targets. This concept is more general than that or a loss function, because it does not necessarily amount to (uniformly) averaging a loss function over samples. In general, the loss depends on the true (training) label and the prediction in a not necessarily symmetric way. Also, in the most general case predictions can be in a different format than labels. E.g., the model prediction could be a probability distribution, while the label is a single value.
- The concept of an AbstractCost function is different from that encoded by the ErrorFunction class. A cost function compares model predictions to labels. It does not know about the model making the predictions, and thus it can not handle LabeledData directly. However, it is one of the components necessary to process LabeledData in an ErrorFunction.
Definition at line 68 of file AbstractCost.h.
Member Typedef Documentation
| typedef Batch<LabelType>::type shark::AbstractCost< LabelT, OutputT >::BatchLabelType |
Definition at line 74 of file AbstractCost.h.
| typedef Batch<OutputType>::type shark::AbstractCost< LabelT, OutputT >::BatchOutputType |
Definition at line 73 of file AbstractCost.h.
| typedef TypedFeatureNotAvailableException<Feature> shark::AbstractCost< LabelT, OutputT >::FeatureNotAvailableException |
Definition at line 86 of file AbstractCost.h.
| typedef TypedFlags<Feature> shark::AbstractCost< LabelT, OutputT >::Features |
Definition at line 86 of file AbstractCost.h.
| typedef LabelT shark::AbstractCost< LabelT, OutputT >::LabelType |
Definition at line 72 of file AbstractCost.h.
| typedef OutputT shark::AbstractCost< LabelT, OutputT >::OutputType |
Definition at line 71 of file AbstractCost.h.
Member Enumeration Documentation
| enum shark::AbstractCost::Feature |
list of features a cost function can have
| Enumerator | |
|---|---|
| HAS_FIRST_DERIVATIVE | |
| IS_LOSS_FUNCTION | |
Definition at line 80 of file AbstractCost.h.
Constructor & Destructor Documentation
|
inlinevirtual |
Definition at line 76 of file AbstractCost.h.
Member Function Documentation
|
inlinevirtual |
configures the loss using informations in the property tree
- The default implementation is empty and actually most cost functions do not need to be configured. See the specific documentation of the corresponding sub-classes for more details.
- Parameters
-
node the configuration node
Reimplemented from shark::IConfigurable.
Definition at line 109 of file AbstractCost.h.
Referenced by shark::detail::ErrorFunctionWrapper< InputType, LabelType, OutputType >::configure(), shark::DenoisingAutoencoderError< InputType, RngType >::configure(), and shark::detail::NoisyErrorFunctionWrapper< InputType, LabelType, OutputType, RngType >::configure().
|
pure virtual |
Evaluates the cost of predictions, given targets.
- Parameters
-
targets target values predictions predictions, typically made by a model
Implemented in shark::NegativeWilcoxonMannWhitneyStatistic< LabelType, OutputType >, shark::AbstractLoss< LabelT, OutputT >, shark::AbstractLoss< LabelType >, shark::AbstractLoss< RealVector, RealVector >, shark::AbstractLoss< unsigned int, unsigned int >, shark::AbstractLoss< LabelType, OutputType >, shark::AbstractLoss< VectorType, VectorType >, shark::AbstractLoss< LabelType, LabelType >, shark::AbstractLoss< unsigned int, RealVector >, shark::AbstractLoss< InputType, InputType >, and shark::NegativeAUC< LabelType, OutputType >.
Referenced by shark::CrossValidationError< ModelTypeT, LabelTypeT >::eval(), shark::detail::CostBasedErrorFunctionImpl< InputType, LabelType, OutputType >::eval(), and shark::AbstractCost< InputType, InputType >::operator()().
|
inlinevirtual |
evaluates the cost and the derivative w.r.t. the prediction
- The default implementations throws an exception. If you overwrite this method, don't forget to set the flag HAS_FIRST_DERIVATIVE.
- Parameters
-
targets target values predictions predictions, typically made by a model gradient the gradient of the cost function with respect to the predictions
Reimplemented in shark::AbstractLoss< LabelT, OutputT >, shark::AbstractLoss< LabelType >, shark::AbstractLoss< RealVector, RealVector >, shark::AbstractLoss< unsigned int, unsigned int >, shark::AbstractLoss< LabelType, OutputType >, shark::AbstractLoss< VectorType, VectorType >, shark::AbstractLoss< LabelType, LabelType >, shark::AbstractLoss< unsigned int, RealVector >, and shark::AbstractLoss< InputType, InputType >.
Definition at line 126 of file AbstractCost.h.
|
inline |
Definition at line 86 of file AbstractCost.h.
|
inline |
returns true when the first parameter derivative is implemented
Definition at line 89 of file AbstractCost.h.
Referenced by shark::DenoisingAutoencoderError< InputType, RngType >::DenoisingAutoencoderError(), shark::detail::ErrorFunctionWrapper< InputType, LabelType, OutputType >::updateFeatures(), and shark::detail::NoisyErrorFunctionWrapper< InputType, LabelType, OutputType, RngType >::updateFeatures().
|
inline |
returns true when the cost function is in fact a loss function
Definition at line 98 of file AbstractCost.h.
Referenced by shark::ErrorFunction< InputType, LabelType >::ErrorFunction().
|
inline |
Definition at line 135 of file AbstractCost.h.
|
inlinevirtual |
Definition at line 86 of file AbstractCost.h.
Referenced by shark::detail::ErrorFunctionWrapper< InputType, LabelType, OutputType >::updateFeatures().
Member Data Documentation
|
protected |
Definition at line 86 of file AbstractCost.h.
Referenced by shark::AbstractLoss< InputType, InputType >::AbstractLoss(), shark::AbstractCost< InputType, InputType >::hasFirstDerivative(), and shark::AbstractCost< InputType, InputType >::isLossFunction().
The documentation for this class was generated from the following file:
- include/shark/ObjectiveFunctions/AbstractCost.h