Improve GMM by registering m_components within the new parameter

framework. * Update also the associated observe() directives; * Improve GMM observable meta example;
shogun-toolbox · May 30, 2019 · 13287b8 · 13287b8
1 parent 4bbf17d
commit 13287b8
Show file tree

Hide file tree

Showing 2 changed files with 14 additions and 10 deletions.
diff --git a/examples/meta/src/observers/gaussian_mixture_models.sg b/examples/meta/src/observers/gaussian_mixture_models.sg
@@ -19,17 +19,22 @@ gmm.subscribe(obs)
 #![train_sample]
 gmm.put("features", features_train)
 gmm.train_em()
+RealVector output = gmm.sample()
 #![train_sample]
 
 #![training_smem]
 obs.clear()
 gmm.train_smem()
 #![training_smem]
 
-#![extract_params]
+#![extract_coeff]
 int num_observations = obs.get_int("num_observations")
-ObservedValue obs_val = obs.get_observation(num_observations-1)
-RealVector coeff =  obs_val.get_real_vector("coefficients")
-real log_likelihood = obs_val.get_real("log_likelihood")
-RealVector coef = gmm.get_real_vector("coefficients")
-#![extract_params]
+ObservedValue obs_val = obs.get_observation(num_observations-2)
+RealVector observed_coeff =  obs_val.get_real_vector("coefficients")
+RealVector coeff = gmm.get_real_vector("coefficients")
+#![extract_coeff]
+
+#![extract_likelihood]
+ObservedValue obs_val_2 = obs.get_observation(num_observations-1)
+real log_likelihood = obs_val_2.get_real("log_likelihood")
+#![extract_likelihood]
diff --git a/src/shogun/clustering/GMM.cpp b/src/shogun/clustering/GMM.cpp
@@ -187,7 +187,7 @@ float64_t CGMM::train_em(float64_t min_cov, int32_t max_iter, float64_t min_chan
 
 		this->observe<float64_t>(iter, "log_likelihood", "Log Likelihood", log_likelihood_cur);
 		this->observe<SGVector<float64_t>>(iter, "coefficients", "Mixture Coefficients", alpha);
-		this->observe<std::vector<CGaussian*>>(iter, "gaussians", "Gaussians Components", m_components);
+		this->observe<std::vector<CGaussian*>>(iter, "components");
 
 		iter++;
 	}
@@ -342,7 +342,7 @@ float64_t CGMM::train_smem(int32_t max_iter, int32_t max_cand, float64_t min_cov
 
 		this->observe<float64_t>(iter, "log_likelihood", "Log Likelihood", cur_likelihood);
 		this->observe<SGVector<float64_t>>(iter, "coefficients");
-		this->observe<std::vector<CGaussian*>>(iter, "gaussians", "Gaussians Components",m_components);
+		this->observe<std::vector<CGaussian*>>(iter, "components");
 
 		iter++;
 		pb.print_progress();
@@ -832,8 +832,7 @@ SGVector<float64_t> CGMM::cluster(SGVector<float64_t> point)
 
 void CGMM::register_params()
 {
-	//TODO serialization broken
-	//m_parameters->add((SGVector<CSGObject*>*) &m_components, "m_components", "Mixture components");
+	this->watch_param("components", &m_components, AnyParameterProperties("Mixture components"));
 	SG_ADD(
 	    &m_coefficients, "coefficients", "Mixture coefficients.");
 }