remove sigmoid params array, make methods pure virtual in Calibration

src/shogun/evaluation/Calibration.h

@@ -70,44 +70,26 @@ namespace shogun
 		* @param targets The true labels corresponding to the predictions
 		* @return boolean indicating whether the calibration was succesful
 		**/
-		virtual bool fit_binary(CBinaryLabels* predictions, CBinaryLabels* targets)
-		{
-			SG_NOTIMPLEMENTED
-
-			return true;
-		}
+		virtual bool fit_binary(CBinaryLabels* predictions, CBinaryLabels* targets) = 0;
 
 		/** Calibrate binary predictions based on parameters learned by calling fit.
 		* @param predictions The predictions outputted by the machine
 		* @return Calibrated binary labels
 		**/
-		virtual CBinaryLabels* calibrate_binary(CBinaryLabels* predictions)
-		{
-			SG_NOTIMPLEMENTED
-			return NULL;
-		}
+		virtual CBinaryLabels* calibrate_binary(CBinaryLabels* predictions) = 0;
 
 		/** Fit calibration parameters for multiclass labels.
 		* @param predictions The predictions outputted by the machine
 		* @param targets The true labels corresponding to the predictions
 		* @return boolean indicating whether the calibration was succesful
 		**/
-		virtual bool fit_multiclass(CMulticlassLabels* predictions, CMulticlassLabels* targets)
-		{
-			SG_NOTIMPLEMENTED
-
-			return true;
-		}
+		virtual bool fit_multiclass(CMulticlassLabels* predictions, CMulticlassLabels* targets) = 0;
 
 		/** Calibrate multiclass predictions based on parameters learned by calling fit.
 		* @param predictions The predictions outputted by the machine
 		* @return Calibrated binary labels
 		**/
-		virtual CMulticlassLabels* calibrate_multiclass(CMulticlassLabels* predictions)
-		{
-			SG_NOTIMPLEMENTED
-			return NULL;
-		}
+		virtual CMulticlassLabels* calibrate_multiclass(CMulticlassLabels* predictions) = 0;
 	};
 }
 #endif

src/shogun/evaluation/SigmoidCalibration.cpp

@@ -42,7 +42,7 @@
 
 using namespace shogun;
 
-CSigmoidCalibration::CSigmoidCalibration()
+CSigmoidCalibration::CSigmoidCalibration() : CCalibration()
 {
 	init();
 }
@@ -53,12 +53,15 @@ CSigmoidCalibration::~CSigmoidCalibration()
 
 void CSigmoidCalibration::init()
 {
-	m_sigmoid_parameters.resize(1);
+	m_sigmoid_as.resize_vector(1);
+	m_sigmoid_bs.resize_vector(1);
 	m_maxiter = 100;
 	m_minstep = 1E-10;
 	m_sigma = 1E-12;
 	m_epsilon = 1E-5;
 
+	SG_ADD(&m_sigmoid_as, "m_sigmoid_as", "Vector of paramter A of sigmoid for each class.", MS_NOT_AVAILABLE);
+	SG_ADD(&m_sigmoid_bs, "m_sigmoid_bs", "Vector of paramter B of sigmoid for each class.", MS_NOT_AVAILABLE);
 	SG_ADD(&m_maxiter, "m_maxiter", "Maximum number of iteration for search.", MS_NOT_AVAILABLE);
 	SG_ADD(&m_minstep, "m_minstep", "Minimum step taken in line search.", MS_NOT_AVAILABLE);
 	SG_ADD(&m_sigma, "m_sigma", "Positive parameter to ensure positive semi-definite Hessian.", MS_NOT_AVAILABLE);
@@ -107,19 +110,24 @@ float64_t CSigmoidCalibration::get_epsilon()
 
 bool CSigmoidCalibration::fit_binary(CBinaryLabels* predictions, CBinaryLabels* targets)
 {
-	m_sigmoid_parameters.resize(1);
-
-	m_sigmoid_parameters[0] = CStatistics::fit_sigmoid(predictions->get_values(), targets->get_labels(), 
+	m_sigmoid_as.resize_vector(1);
+	m_sigmoid_bs.resize_vector(1);
+	auto sigmoid_params = CStatistics::fit_sigmoid(predictions->get_values(), targets->get_labels(), 
 		m_maxiter, m_minstep, m_sigma, m_epsilon);
 
+	m_sigmoid_as[0] = sigmoid_params.a;
+	m_sigmoid_bs[0] = sigmoid_params.b;
+
 	return true;
 }
 
 CBinaryLabels*
 CSigmoidCalibration::calibrate_binary(CBinaryLabels* predictions)
 {
-	auto params = m_sigmoid_parameters[0];
-
+	CStatistics::SigmoidParamters params; 
+	params.a = m_sigmoid_as[0];
+	params.b = m_sigmoid_bs[0];
+
 	/** Convert predictions to probabilties. */
 	auto values = calibrate_values(predictions->get_values(), params);
 
@@ -133,7 +141,8 @@ bool CSigmoidCalibration::fit_multiclass(CMulticlassLabels* predictions, CMultic
 	index_t num_classes =
 	    predictions->get_num_classes();
 
-	m_sigmoid_parameters.resize(num_classes);
+	m_sigmoid_bs.resize_vector(num_classes);
+	m_sigmoid_bs.resize_vector(num_classes);
 
 	 /** Fit and store parameters for for each class seperately. */
 
@@ -144,8 +153,10 @@ bool CSigmoidCalibration::fit_multiclass(CMulticlassLabels* predictions, CMultic
 		auto pred_values = class_predictions->get_values();
 		auto target_labels = class_targets->get_labels();
 
-		m_sigmoid_parameters[i] = CStatistics::fit_sigmoid(pred_values, target_labels, 
+		auto sigmoid_params = CStatistics::fit_sigmoid(pred_values, target_labels, 
 			m_maxiter, m_minstep, m_sigma, m_epsilon);
+		m_sigmoid_as[i] = sigmoid_params.a;
+		m_sigmoid_bs[i] = sigmoid_params.b;
 	}
 
 	return true;
@@ -165,8 +176,12 @@ CSigmoidCalibration::calibrate_multiclass(CMulticlassLabels* predictions)
 		auto binary_predictions = predictions->get_binary_for_class(i);
 		auto class_values = binary_predictions->get_values();
 
+		CStatistics::SigmoidParamters sigmoid_params;
+		sigmoid_params.a = m_sigmoid_as[i];
+		sigmoid_params.b = m_sigmoid_bs[i];
+
 		SGVector<float64_t> calibrated_values =
-		    calibrate_values(class_values, m_sigmoid_parameters[i]);
+		    calibrate_values(class_values, sigmoid_params);
 
 		confidence_values.set_column(i, calibrated_values);
 	}

src/shogun/evaluation/SigmoidCalibration.h

@@ -97,42 +97,42 @@ namespace shogun
 	 	/** Set maximum number of iterations
 	 	* @param maxiter maximum number of iterations
 	 	*/
-	 	virtual void set_maxiter(index_t maxiter);
+	 	void set_maxiter(index_t maxiter);
 
 	 	/** Get max iterations
 	 	* @return maximum number of iterations 
 	 	*/
-	 	virtual index_t get_maxiter();
+	 	index_t get_maxiter();
 
 	 	/** Set min step
 	 	* @param minstep min step taken in line search
 	 	*/
-	 	virtual void set_minstep(float64_t minstep);
+	 	void set_minstep(float64_t minstep);
 
 	 	/** Get min step 
 	 	* @return minimum steps taken in line search
 	 	*/
-	 	virtual float64_t get_minstep();
+	 	float64_t get_minstep();
 
 	 	/** Set sigma
 	 	* @param sigma Set to a value greater than 0 to ensure that the Hessian matrix is positive semi-definite
 	 	*/
-	 	virtual void set_sigma(float64_t sigma);
+	 	void set_sigma(float64_t sigma);
 
 	 	/** Get sigma
 	 	* @return sigma
 	 	*/
-	 	virtual float64_t get_sigma();
+	 	float64_t get_sigma();
 
 	 	/** Get epsilon
 	 	* @param epsilon stopping criteria
 	 	*/
-	 	virtual void set_epsilon(float64_t epsilon);
+	 	void set_epsilon(float64_t epsilon);
 
 	 	/** Get epsilon
 	 	* @return stopping critera
 	 	*/
-	 	virtual float64_t get_epsilon();
+	 	float64_t get_epsilon();
 
 	private:
 		/** Initialize parameters */
@@ -145,8 +145,10 @@ namespace shogun
 		SGVector<float64_t> calibrate_values(SGVector<float64_t> values, CStatistics::SigmoidParamters params);
 
 	private:
-		/** Array to store sigmoid parameters for each class. In case of binary labels, only one pair of parameters are stored. */
-		std::vector<CStatistics::SigmoidParamters> m_sigmoid_parameters;
+		/** Vector to store parameter A of sigmoid for each class. In case of binary labels, only one pair of parameters are stored. */
+		SGVector<float64_t> m_sigmoid_as;
+		/** Vector to store parameter B of sigmoid for each class. */
+		SGVector<float64_t> m_sigmoid_bs;
 		/** Maximum number of iterations. */
 		index_t m_maxiter;
 		/** Minimum step taken in line search. */