slope::AUC Class Reference

Area Under the ROC Curve (AUC-ROC) scoring metric. More...

#include <score.h>

Inheritance diagram for slope::AUC:

Collaboration diagram for slope::AUC:

Public Member Functions
double	eval (const Eigen::MatrixXd &eta, const Eigen::MatrixXd &y, const std::unique_ptr< Loss > &loss) const override
double	eval (const Eigen::MatrixXd &eta, const Eigen::MatrixXd &y, const std::unique_ptr< Loss > &loss, const Eigen::VectorXd &observation_weights) const override
Public Member Functions inherited from slope::MaximizeScore
bool	isWorse (double a, double b) const override
double	initValue () const override
Public Member Functions inherited from slope::Score
std::function< bool(double, double)>	getComparator () const

Additional Inherited Members
Static Public Member Functions inherited from slope::Score
static std::unique_ptr< Score >	create (const std::string &metric)

Detailed Description

Area Under the ROC Curve (AUC-ROC) scoring metric.

Computes the area under the Receiver Operating Characteristic curve, which plots the true positive rate against false positive rate at various classification thresholds. Inherits from MaximizeScore since higher AUC values indicate better discriminative ability.

For binary classification:

• AUC of 1.0 represents perfect prediction
• AUC of 0.5 represents random prediction
• AUC of 0.0 represents perfectly inverse prediction

For multi-class problems:

• Computes the one-vs-rest AUC for each class
• Returns the average across all classes

Definition at line 353 of file score.h.

Member Function Documentation

eval() [1/2]

double slope::AUC::eval ( const Eigen::MatrixXd & eta, const Eigen::MatrixXd & y, const std::unique_ptr< Loss > & loss ) const

overridevirtual

Evaluates the AUC-ROC score for predictions.

Parameters

eta	Matrix of model predictions/probabilities
y	Matrix of true labels (one-hot encoded)
loss	Loss function used to transform predictions if needed

Returns

AUC-ROC score (between 0 and 1)

Implements slope::Score.

eval() [2/2]

double slope::AUC::eval ( const Eigen::MatrixXd & eta, const Eigen::MatrixXd & y, const std::unique_ptr< Loss > & loss, const Eigen::VectorXd & observation_weights ) const

overridevirtual

Evaluates the metric with observation weights. Empty weights select the unweighted metric. Weights must be finite and nonnegative, and their sum must be positive. Multiplying every weight by a positive constant leaves the score unchanged. Custom metrics can override this overload; the default implementation accepts uniform weights only.

Reimplemented from slope::Score.

---

The documentation for this class was generated from the following file:

• include/slope/score.h