Merge pull request #1366 from karlnapf/develop

Some minor changes towards x-validation for GPC

src/NEWS

@@ -8,6 +8,7 @@
 		- Added a new class for classical probability distribution that can be
 		  sampled and whose log-pdf can be evaluated. Added the multivariate
 		  Gaussian with various numerical flavours.
+		- Cross-validation framework works now with Gaussian Processes
 		- Added nu-SVR for LibSVR class
 		- Modelselection is now supported for parameters of sub-kernels of
 		  combined kernels in the MKL context. Thanks to Evangelos Anagnostopoulos

src/shogun/base/SGObject.cpp

@@ -1176,6 +1176,7 @@ void CSGObject::get_parameter_incremental_hash(Parameter* param,
 	for (index_t i=0; i<param->get_num_parameters(); i++)
 	{
 		TParameter* p = param->get_parameter(i);
+		SG_DEBUG("Updating hash for parameter \"%s\"\n", p->m_name ? p->m_name : "(nil)");
 
 		if (!p || !p->is_valid())
 			continue;

src/shogun/distributions/classical/GaussianDistribution.cpp

@@ -107,7 +107,7 @@ SGMatrix<float64_t> CGaussianDistribution::sample(int32_t num_samples,
 	return samples;
 }
 
-SGVector<float64_t> CGaussianDistribution::log_pdf(SGMatrix<float64_t> samples) const
+SGVector<float64_t> CGaussianDistribution::log_pdf_multiple(SGMatrix<float64_t> samples) const
 {
 	REQUIRE(samples.num_cols>0, "Number of samples must be positive, but is %d\n",
 			samples.num_cols);

src/shogun/distributions/classical/GaussianDistribution.h

@@ -79,7 +79,7 @@ class CGaussianDistribution: public CProbabilityDistribution
 	 * @param samples samples to compute log-pdf of (column vectors)
 	 * @return vector with log-pdfs of given samples
 	 */
-	virtual SGVector<float64_t> log_pdf(SGMatrix<float64_t> samples) const;
+	virtual SGVector<float64_t> log_pdf_multiple(SGMatrix<float64_t> samples) const;
 
 	/** @return name of the SGSerializable */
 	virtual const char* get_name() const

src/shogun/distributions/classical/ProbabilityDistribution.cpp

@@ -51,7 +51,7 @@ SGVector<float64_t> CProbabilityDistribution::sample() const
 	return result;
 }
 
-SGVector<float64_t> CProbabilityDistribution::log_pdf(
+SGVector<float64_t> CProbabilityDistribution::log_pdf_multiple(
 		SGMatrix<float64_t> samples) const
 {
 	SG_ERROR("Not implemented in sub-class\n");
@@ -66,7 +66,7 @@ float64_t CProbabilityDistribution::log_pdf(SGVector<float64_t> single_sample) c
 
 	SGMatrix<float64_t> s(m_dimension, 1);
 	memcpy(s.matrix, single_sample.vector, m_dimension*sizeof(float64_t));
-	return log_pdf(s)[0];
+	return log_pdf_multiple(s)[0];
 }
 
 void CProbabilityDistribution::init()

src/shogun/distributions/classical/ProbabilityDistribution.h

@@ -57,9 +57,10 @@ class CProbabilityDistribution: public CSGObject
 	 * @param samples samples to compute log-pdf of (column vectors)
 	 * @return vector with log-pdfs of given samples
 	 */
-	virtual SGVector<float64_t> log_pdf(SGMatrix<float64_t> samples) const;
+	virtual SGVector<float64_t> log_pdf_multiple(SGMatrix<float64_t> samples) const;
 
-	/** Computes the log-pdf for a single provided sample. Wrapper method.
+	/** Computes the log-pdf for a single provided sample. Wrapper method which
+	 * calls log_pdf_multiple
 	 *
 	 * @param sample sample to compute log-pdf for
 	 * @return log-pdf of the given sample

src/shogun/labels/DenseLabels.cpp

@@ -82,7 +82,7 @@ void CDenseLabels::set_labels(SGVector<float64_t> v)
 SGVector<float64_t> CDenseLabels::get_labels()
 {
 	if (m_subset_stack->has_subsets())
-		SG_ERROR("get_labels() is not possible on subset")
+		return get_labels_copy();
 
 	return m_labels;
 }

src/shogun/labels/DenseLabels.h

@@ -113,11 +113,9 @@ class CDenseLabels : public CLabels
 		 */
 		int32_t get_int_label(int32_t idx);
 
-		/** get labels
+		/** Getter for labels
 		 *
-		 * not possible with subset
-		 *
-		 * @return labels
+		 * @return labels, a copy if a subset is set
 		 */
 		SGVector<float64_t> get_labels();
 

src/shogun/machine/GaussianProcessMachine.h

@@ -101,6 +101,13 @@ class CGaussianProcessMachine : public CMachine
 		m_method->set_labels(lab);
 	}
 
+	/** Stores feature data of underlying model.
+	 * After this method has been called, it is possible to change
+	 * the machine's feature data and call apply(), which is then performed
+	 * on the training feature data that is part of the machine's model.
+	 */
+	virtual void store_model_features() { }
+
 private:
 	void init();
 

src/shogun/machine/gp/InferenceMethod.cpp

@@ -114,7 +114,7 @@ float64_t CInferenceMethod::get_log_ml_estimate(
 	 * log pdf of approximation, prior and likelihood */
 
 	/* log pdf q(f^i|y) */
-	SGVector<float64_t> log_pdf_post_approx=post_approx->log_pdf(samples);
+	SGVector<float64_t> log_pdf_post_approx=post_approx->log_pdf_multiple(samples);
 
 	/* dont need gaussian anymore, free memory */
 	SG_UNREF(post_approx);
@@ -133,7 +133,7 @@ float64_t CInferenceMethod::get_log_ml_estimate(
 
 	CGaussianDistribution* prior=new CGaussianDistribution(
 			m_mean->get_mean_vector(m_feature_matrix), scaled_kernel);
-	SGVector<float64_t> log_pdf_prior=prior->log_pdf(samples);
+	SGVector<float64_t> log_pdf_prior=prior->log_pdf_multiple(samples);
 	SG_UNREF(prior);
 	prior=NULL;
 

tests/unit/distribution/classical/GaussianDistribution_unittest.cc

@@ -45,7 +45,7 @@ TEST(GaussianDistribution,log_pdf_multiple_1d)
 	SGMatrix<float64_t> x(1,2);
 	x(0,0)=0;
 	x(0,1)=1;
-	SGVector<float64_t> result=(gauss)->log_pdf(x);
+	SGVector<float64_t> result=(gauss)->log_pdf_multiple(x);
 
 	EXPECT_NEAR(result[0], -1.5155121234846454, 1e-15);
 	EXPECT_NEAR(result[1], -1.2655121234846454, 1e-15);
@@ -92,7 +92,7 @@ TEST(GaussianDistribution,log_pdf_multiple_2d)
 	x(1,0)=2;
 	x(0,1)=3;
 	x(1,1)=4;
-	SGVector<float64_t> result=(gauss)->log_pdf(x);
+	SGVector<float64_t> result=(gauss)->log_pdf_multiple(x);
 
 	EXPECT_NEAR(result[0], -2.539698566136597, 1e-15);
 	EXPECT_NEAR(result[1], -3.620779647217678, 1e-15);