Prefix OnlineSupervisedMStep with Fast

angeloskath · Sep 9, 2016 · b116ca1 · b116ca1
1 parent 034d42f
commit b116ca1
Show file tree

Hide file tree

Showing 7 changed files with 50 additions and 49 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -42,11 +42,11 @@ set(SOURCES
     src/ldaplusplus/em/AbstractEStep.cpp
     src/ldaplusplus/em/CorrespondenceSupervisedEStep.cpp
     src/ldaplusplus/em/CorrespondenceSupervisedMStep.cpp
+    src/ldaplusplus/em/FastOnlineSupervisedMStep.cpp
     src/ldaplusplus/em/FastSupervisedEStep.cpp
     src/ldaplusplus/em/FastSupervisedMStep.cpp
     src/ldaplusplus/em/MultinomialSupervisedEStep.cpp
     src/ldaplusplus/em/MultinomialSupervisedMStep.cpp
-    src/ldaplusplus/em/OnlineSupervisedMStep.cpp
     src/ldaplusplus/em/SemiSupervisedEStep.cpp
     src/ldaplusplus/em/SemiSupervisedMStep.cpp
     src/ldaplusplus/em/SupervisedEStep.cpp

diff --git a/include/ldaplusplus/LDABuilder.hpp b/include/ldaplusplus/LDABuilder.hpp
@@ -432,10 +432,10 @@ class LDABuilder : public LDABuilderInterface<Scalar>
         }
 
         /**
-         * Create an OnlineSupervisedMStep without specifying class weights.
+         * Create an FastOnlineSupervisedMStep without specifying class weights.
          *
          * You can also see a description of the parameters at
-         * OnlineSupervisedMStep::OnlineSupervisedMStep.
+         * FastOnlineSupervisedMStep::FastOnlineSupervisedMStep.
          *
          * @param num_classes            The number of classes
          * @param regularization_penalty The L2 penalty for the logistic
@@ -447,25 +447,25 @@ class LDABuilder : public LDABuilderInterface<Scalar>
          * @param eta_learning_rate      The learning rate for the SGD
          *                               update of \f$\eta\f$
          * @param beta_weight            The weight for the online update
-         *                                   of \f$\beta\f$
+         *                               of \f$\beta\f$
          */
-        std::shared_ptr<em::MStepInterface<Scalar> > get_supervised_online_m_step(
+        std::shared_ptr<em::MStepInterface<Scalar> > get_fast_supervised_online_m_step(
             size_t num_classes,
             Scalar regularization_penalty = 1e-2,
             size_t minibatch_size = 128,
             Scalar eta_momentum = 0.9,
             Scalar eta_learning_rate = 0.01,
             Scalar beta_weight = 0.9
         );
-        LDABuilder & set_supervised_online_m_step(
+        LDABuilder & set_fast_supervised_online_m_step(
             size_t num_classes,
             Scalar regularization_penalty = 1e-2,
             size_t minibatch_size = 128,
             Scalar eta_momentum = 0.9,
             Scalar eta_learning_rate = 0.01,
             Scalar beta_weight = 0.9
         ) {
-            return set_m(get_supervised_online_m_step(
+            return set_m(get_fast_supervised_online_m_step(
                 num_classes,
                 regularization_penalty,
                 minibatch_size,
@@ -476,10 +476,10 @@ class LDABuilder : public LDABuilderInterface<Scalar>
         }
 
         /**
-         * Create an OnlineSupervisedMStep.
+         * Create an FastOnlineSupervisedMStep.
          *
          * You can also see a description of the parameters at
-         * OnlineSupervisedMStep::OnlineSupervisedMStep.
+         * FastOnlineSupervisedMStep::FastOnlineSupervisedMStep.
          *
          * @param class_weights          Weights to account for class
          *                               imbalance
@@ -494,23 +494,23 @@ class LDABuilder : public LDABuilderInterface<Scalar>
          * @param beta_weight            The weight for the online update
          *                                   of \f$\beta\f$
          */
-        std::shared_ptr<em::MStepInterface<Scalar> > get_supervised_online_m_step(
+        std::shared_ptr<em::MStepInterface<Scalar> > get_fast_supervised_online_m_step(
             std::vector<Scalar> class_weights,
             Scalar regularization_penalty = 1e-2,
             size_t minibatch_size = 128,
             Scalar eta_momentum = 0.9,
             Scalar eta_learning_rate = 0.01,
             Scalar beta_weight = 0.9
         );
-        LDABuilder & set_supervised_online_m_step(
+        LDABuilder & set_fast_supervised_online_m_step(
             std::vector<Scalar> class_weights,
             Scalar regularization_penalty = 1e-2,
             size_t minibatch_size = 128,
             Scalar eta_momentum = 0.9,
             Scalar eta_learning_rate = 0.01,
             Scalar beta_weight = 0.9
         ) {
-            return set_m(get_supervised_online_m_step(
+            return set_m(get_fast_supervised_online_m_step(
                 class_weights,
                 regularization_penalty,
                 minibatch_size,
@@ -521,10 +521,10 @@ class LDABuilder : public LDABuilderInterface<Scalar>
         }
 
         /**
-         * Create an OnlineSupervisedMStep.
+         * Create an FastOnlineSupervisedMStep.
          *
          * You can also see a description of the parameters at
-         * OnlineSupervisedMStep::OnlineSupervisedMStep.
+         * FastOnlineSupervisedMStep::FastOnlineSupervisedMStep.
          *
          * @param class_weights          Weights to account for class
          *                               imbalance
@@ -539,23 +539,23 @@ class LDABuilder : public LDABuilderInterface<Scalar>
          * @param beta_weight            The weight for the online update
          *                                   of \f$\beta\f$
          */
-        std::shared_ptr<em::MStepInterface<Scalar> > get_supervised_online_m_step(
+        std::shared_ptr<em::MStepInterface<Scalar> > get_fast_supervised_online_m_step(
             Eigen::Matrix<Scalar, Eigen::Dynamic, 1> class_weights,
             Scalar regularization_penalty = 1e-2,
             size_t minibatch_size = 128,
             Scalar eta_momentum = 0.9,
             Scalar eta_learning_rate = 0.01,
             Scalar beta_weight = 0.9
         );
-        LDABuilder & set_supervised_online_m_step(
+        LDABuilder & set_fast_supervised_online_m_step(
             Eigen::Matrix<Scalar, Eigen::Dynamic, 1> class_weights,
             Scalar regularization_penalty = 1e-2,
             size_t minibatch_size = 128,
             Scalar eta_momentum = 0.9,
             Scalar eta_learning_rate = 0.01,
             Scalar beta_weight = 0.9
         ) {
-            return set_m(get_supervised_online_m_step(
+            return set_m(get_fast_supervised_online_m_step(
                 class_weights,
                 regularization_penalty,
                 minibatch_size,

diff --git a/.../ldaplusplus/em/OnlineSupervisedMStep.hpp → ...plusplus/em/FastOnlineSupervisedMStep.hpp b/.../ldaplusplus/em/OnlineSupervisedMStep.hpp → ...plusplus/em/FastOnlineSupervisedMStep.hpp
@@ -1,5 +1,5 @@
-#ifndef _ONLINE_SUPERVISED_M_STEP_HPP_
-#define _ONLINE_SUPERVISED_M_STEP_HPP_
+#ifndef _LDAPLUSPLUS_EM_FASTONLINESUPERVISEDMSTEP_HPP_
+#define _LDAPLUSPLUS_EM_FASTONLINESUPERVISEDMSTEP_HPP_
 
 #include "ldaplusplus/em/MStepInterface.hpp"
 
@@ -8,7 +8,7 @@ namespace em {
 
 
 /**
- * OnlineSupervisedMStep is an online implementation of the classical
+ * FastOnlineSupervisedMStep is an online implementation of the classical
  * categorical supervised LDA.
  *
  * m_step() is called by doc_m_step() according to the minibatch_size
@@ -24,14 +24,14 @@ namespace em {
  * SupervisedMStep.
  */
 template <typename Scalar>
-class OnlineSupervisedMStep : public MStepInterface<Scalar>
+class FastOnlineSupervisedMStep : public MStepInterface<Scalar>
 {
     typedef Eigen::Matrix<Scalar, Eigen::Dynamic, Eigen::Dynamic> MatrixX;
     typedef Eigen::Matrix<Scalar, Eigen::Dynamic, 1> VectorX;
 
     public:
         /**
-         * Create an OnlineSupervisedMStep that accounts for class imbalance by
+         * Create an FastOnlineSupervisedMStep that accounts for class imbalance by
          * weighting the classes.
          *
          * @param class_weights          Weights to account for class
@@ -47,7 +47,7 @@ class OnlineSupervisedMStep : public MStepInterface<Scalar>
          * @param beta_weight            The weight for the online update
          *                                   of \f$\beta\f$
          */
-        OnlineSupervisedMStep(
+        FastOnlineSupervisedMStep(
             VectorX class_weights,
             Scalar regularization_penalty = 1e-2,
             size_t minibatch_size = 128,
@@ -56,7 +56,7 @@ class OnlineSupervisedMStep : public MStepInterface<Scalar>
             Scalar beta_weight = 0.9
         );
         /**
-         * Create an OnlineSupervisedMStep that uses uniform weights for the
+         * Create an FastOnlineSupervisedMStep that uses uniform weights for the
          * classes.
          *
          * @param num_classes            The number of classes
@@ -71,7 +71,7 @@ class OnlineSupervisedMStep : public MStepInterface<Scalar>
          * @param beta_weight            The weight for the online update
          *                                   of \f$\beta\f$
          */
-        OnlineSupervisedMStep(
+        FastOnlineSupervisedMStep(
             size_t num_classes,
             Scalar regularization_penalty = 1e-2,
             size_t minibatch_size = 128,
@@ -134,4 +134,4 @@ class OnlineSupervisedMStep : public MStepInterface<Scalar>
 }  // namespace em
 }  // namespace ldaplusplus
 
-#endif  // _ONLINE_SUPERVISED_M_STEP_HPP_
+#endif  // _LDAPLUSPLUS_EM_FASTONLINESUPERVISEDMSTEP_HPP_
diff --git a/slda.cpp b/slda.cpp
@@ -346,7 +346,7 @@ LDA<double> create_lda_for_training(
             std::stof(args["--regularization_penalty"].asString())
         );
     } else if (args["--online_m_step"].asBool()) {
-        builder.set_supervised_online_m_step(
+        builder.set_fast_supervised_online_m_step(
             create_class_weights(y),
             std::stof(args["--regularization_penalty"].asString()),
             args["--batch_size"].asLong(),

diff --git a/src/ldaplusplus/LDABuilder.cpp b/src/ldaplusplus/LDABuilder.cpp
@@ -1,10 +1,10 @@
 #include "ldaplusplus/LDABuilder.hpp"
 #include "ldaplusplus/em/CorrespondenceSupervisedEStep.hpp"
 #include "ldaplusplus/em/CorrespondenceSupervisedMStep.hpp"
+#include "ldaplusplus/em/FastOnlineSupervisedMStep.hpp"
 #include "ldaplusplus/em/FastSupervisedMStep.hpp"
 #include "ldaplusplus/em/MultinomialSupervisedEStep.hpp"
 #include "ldaplusplus/em/MultinomialSupervisedMStep.hpp"
-#include "ldaplusplus/em/OnlineSupervisedMStep.hpp"
 #include "ldaplusplus/em/SemiSupervisedEStep.hpp"
 #include "ldaplusplus/em/SemiSupervisedMStep.hpp"
 #include "ldaplusplus/em/SupervisedEStep.hpp"
@@ -175,15 +175,15 @@ std::shared_ptr<em::MStepInterface<Scalar> > LDABuilder<Scalar>::get_supervised_
 }
 
 template <typename Scalar>
-std::shared_ptr<em::MStepInterface<Scalar> > LDABuilder<Scalar>::get_supervised_online_m_step(
+std::shared_ptr<em::MStepInterface<Scalar> > LDABuilder<Scalar>::get_fast_supervised_online_m_step(
     size_t num_classes,
     Scalar regularization_penalty,
     size_t minibatch_size,
     Scalar eta_momentum,
     Scalar eta_learning_rate,
     Scalar beta_weight
 ) {
-    return std::make_shared<em::OnlineSupervisedMStep<Scalar> >(
+    return std::make_shared<em::FastOnlineSupervisedMStep<Scalar> >(
         num_classes,
         regularization_penalty,
         minibatch_size,
@@ -194,7 +194,7 @@ std::shared_ptr<em::MStepInterface<Scalar> > LDABuilder<Scalar>::get_supervised_
 }
 
 template <typename Scalar>
-std::shared_ptr<em::MStepInterface<Scalar> > LDABuilder<Scalar>::get_supervised_online_m_step(
+std::shared_ptr<em::MStepInterface<Scalar> > LDABuilder<Scalar>::get_fast_supervised_online_m_step(
     std::vector<Scalar> class_weights,
     Scalar regularization_penalty,
     size_t minibatch_size,
@@ -208,7 +208,7 @@ std::shared_ptr<em::MStepInterface<Scalar> > LDABuilder<Scalar>::get_supervised_
         weights[i] = class_weights[i];
     }
 
-    return get_supervised_online_m_step(
+    return get_fast_supervised_online_m_step(
         weights,
         regularization_penalty,
         minibatch_size,
@@ -219,15 +219,15 @@ std::shared_ptr<em::MStepInterface<Scalar> > LDABuilder<Scalar>::get_supervised_
 }
 
 template <typename Scalar>
-std::shared_ptr<em::MStepInterface<Scalar> > LDABuilder<Scalar>::get_supervised_online_m_step(
+std::shared_ptr<em::MStepInterface<Scalar> > LDABuilder<Scalar>::get_fast_supervised_online_m_step(
     Eigen::Matrix<Scalar, Eigen::Dynamic, 1> class_weights,
     Scalar regularization_penalty,
     size_t minibatch_size,
     Scalar eta_momentum,
     Scalar eta_learning_rate,
     Scalar beta_weight
 ) {
-    return std::make_shared<em::OnlineSupervisedMStep<Scalar> >(
+    return std::make_shared<em::FastOnlineSupervisedMStep<Scalar> >(
         class_weights,
         regularization_penalty,
         minibatch_size,

diff --git a/src/ldaplusplus/em/OnlineSupervisedMStep.cpp → ...plusplus/em/FastOnlineSupervisedMStep.cpp b/src/ldaplusplus/em/OnlineSupervisedMStep.cpp → ...plusplus/em/FastOnlineSupervisedMStep.cpp
@@ -1,16 +1,15 @@
 #include <utility>
 
-#include "ldaplusplus/em/OnlineSupervisedMStep.hpp"
+#include "ldaplusplus/em/FastOnlineSupervisedMStep.hpp"
 #include "ldaplusplus/optimization/MultinomialLogisticRegression.hpp"
 #include "ldaplusplus/events/ProgressEvents.hpp"
 
 namespace ldaplusplus {
-
-using em::OnlineSupervisedMStep;
+namespace em {
 
 
 template <typename Scalar>
-OnlineSupervisedMStep<Scalar>::OnlineSupervisedMStep(
+FastOnlineSupervisedMStep<Scalar>::FastOnlineSupervisedMStep(
     VectorX class_weights,
     Scalar regularization_penalty,
     size_t minibatch_size,
@@ -28,14 +27,14 @@ OnlineSupervisedMStep<Scalar>::OnlineSupervisedMStep(
 {}
 
 template <typename Scalar>
-OnlineSupervisedMStep<Scalar>::OnlineSupervisedMStep(
+FastOnlineSupervisedMStep<Scalar>::FastOnlineSupervisedMStep(
     size_t num_classes,
     Scalar regularization_penalty,
     size_t minibatch_size,
     Scalar eta_momentum,
     Scalar eta_learning_rate,
     Scalar beta_weight
-) : OnlineSupervisedMStep(
+) : FastOnlineSupervisedMStep(
         VectorX::Constant(num_classes, 1),
         regularization_penalty,
         minibatch_size,
@@ -46,7 +45,7 @@ OnlineSupervisedMStep<Scalar>::OnlineSupervisedMStep(
 {}
 
 template <typename Scalar>
-void OnlineSupervisedMStep<Scalar>::doc_m_step(
+void FastOnlineSupervisedMStep<Scalar>::doc_m_step(
     const std::shared_ptr<corpus::Document> doc,
     const std::shared_ptr<parameters::Parameters> v_parameters,
     std::shared_ptr<parameters::Parameters> m_parameters
@@ -87,7 +86,7 @@ void OnlineSupervisedMStep<Scalar>::doc_m_step(
 }
 
 template <typename Scalar>
-void OnlineSupervisedMStep<Scalar>::m_step(
+void FastOnlineSupervisedMStep<Scalar>::m_step(
     std::shared_ptr<parameters::Parameters> parameters
 ) {
     // Check whether we should actually perform the m_step
@@ -126,7 +125,9 @@ void OnlineSupervisedMStep<Scalar>::m_step(
 
 
 // Instantiations
-template class OnlineSupervisedMStep<float>;
-template class OnlineSupervisedMStep<double>;
+template class FastOnlineSupervisedMStep<float>;
+template class FastOnlineSupervisedMStep<double>;
 
-}
+
+}  // namespace em
+}  // namespace ldaplusplus
diff --git a/test/test_online_maximization_step.cpp b/test/test_online_maximization_step.cpp
@@ -10,8 +10,8 @@
 
 #include "ldaplusplus/Parameters.hpp"
 #include "ldaplusplus/events/ProgressEvents.hpp"
-#include "ldaplusplus/em/SupervisedEStep.hpp"
-#include "ldaplusplus/em/OnlineSupervisedMStep.hpp"
+#include "ldaplusplus/em/FastOnlineSupervisedMStep.hpp"
+#include "ldaplusplus/em/FastSupervisedEStep.hpp"
 
 using namespace Eigen;
 using namespace ldaplusplus;
@@ -50,8 +50,8 @@ TYPED_TEST(TestOnlineMaximizationStep, Maximization) {
         MatrixX<TypeParam>::Zero(10, 6)
     );
 
-    em::SupervisedEStep<TypeParam> e_step(10, 1e-2, 10);
-    em::OnlineSupervisedMStep<TypeParam> m_step(
+    em::FastSupervisedEStep<TypeParam> e_step(10, 1e-2, 10);
+    em::FastOnlineSupervisedMStep<TypeParam> m_step(
         6,
         1e-2,
         25