Pass the exemplar into the NB restore constructor rather than persisting and restoring

Author: tveasey

include/maths/CNaiveBayes.h

@@ -149,7 +149,8 @@ class MATHS_EXPORT CNaiveBayes {
     explicit CNaiveBayes(const CNaiveBayesFeatureDensity& exemplar,
                          double decayRate = 0.0,
                          TOptionalDouble minMaxLogLikelihoodToUseFeature = TOptionalDouble());
-    CNaiveBayes(const SDistributionRestoreParams& params,
+    CNaiveBayes(const CNaiveBayesFeatureDensity& exemplar,
+                const SDistributionRestoreParams& params,
                 core::CStateRestoreTraverser& traverser);
     CNaiveBayes(const CNaiveBayes& other);
 

lib/maths/CNaiveBayes.cc

@@ -35,7 +35,6 @@ const std::string CLASS_MODEL_TAG{"c"};
 const std::string MIN_MAX_LOG_LIKELIHOOD_TO_USE_FEATURE_TAG{"d"};
 const std::string COUNT_TAG{"e"};
 const std::string CONDITIONAL_DENSITY_FROM_PRIOR_TAG{"f"};
-const std::string EXEMPLAR_FROM_PRIOR_TAG{"g"};
 }
 
 CNaiveBayesFeatureDensityFromPrior::CNaiveBayesFeatureDensityFromPrior(const CPrior& prior)
@@ -125,9 +124,10 @@ CNaiveBayes::CNaiveBayes(const CNaiveBayesFeatureDensity& exemplar,
       m_DecayRate{decayRate}, m_Exemplar{exemplar.clone()}, m_ClassConditionalDensities{2} {
 }
 
-CNaiveBayes::CNaiveBayes(const SDistributionRestoreParams& params,
+CNaiveBayes::CNaiveBayes(const CNaiveBayesFeatureDensity& exemplar,
+                         const SDistributionRestoreParams& params,
                          core::CStateRestoreTraverser& traverser)
-    : m_DecayRate{params.s_DecayRate}, m_ClassConditionalDensities{2} {
+    : m_DecayRate{params.s_DecayRate}, m_Exemplar{exemplar.clone()}, m_ClassConditionalDensities{2} {
     traverser.traverseSubLevel(boost::bind(&CNaiveBayes::acceptRestoreTraverser,
                                            this, boost::cref(params), _1));
 }
@@ -146,13 +146,6 @@ bool CNaiveBayes::acceptRestoreTraverser(const SDistributionRestoreParams& param
     do {
         const std::string& name{traverser.name()};
         RESTORE_BUILT_IN(CLASS_LABEL_TAG, label)
-        RESTORE_SETUP_TEARDOWN(
-            EXEMPLAR_FROM_PRIOR_TAG, CNaiveBayesFeatureDensityFromPrior density,
-            traverser.traverseSubLevel(
-                boost::bind(&CNaiveBayesFeatureDensityFromPrior::acceptRestoreTraverser,
-                            boost::ref(density), boost::cref(params), _1)),
-            m_Exemplar.reset(density.clone()))
-        // Add other implementations' restore code here.
         RESTORE_SETUP_TEARDOWN(
             CLASS_MODEL_TAG, CClass class_,
             traverser.traverseSubLevel(boost::bind(&CClass::acceptRestoreTraverser,
@@ -170,13 +163,6 @@ void CNaiveBayes::acceptPersistInserter(core::CStatePersistInserter& inserter) c
     using TSizeClassUMapCItr = TSizeClassUMap::const_iterator;
     using TSizeClassUMapCItrVec = std::vector<TSizeClassUMapCItr>;
 
-    if (dynamic_cast<const CNaiveBayesFeatureDensityFromPrior*>(m_Exemplar.get())) {
-        inserter.insertLevel(EXEMPLAR_FROM_PRIOR_TAG,
-                             boost::bind(&CNaiveBayesFeatureDensity::acceptPersistInserter,
-                                         m_Exemplar.get(), _1));
-    }
-    // Add other implementations' persist code here.
-
     TSizeClassUMapCItrVec classes;
     classes.reserve(m_ClassConditionalDensities.size());
     for (auto i = m_ClassConditionalDensities.begin();

lib/maths/CTrendComponent.cc

@@ -69,11 +69,14 @@ TOptionalDoubleDoublePr confidenceInterval(double prediction, double variance, d
     return TOptionalDoubleDoublePr{};
 }
 
+CNaiveBayesFeatureDensityFromPrior naiveBayesExemplar(double decayRate) {
+    return CNaiveBayesFeatureDensityFromPrior{CNormalMeanPrecConjugate::nonInformativePrior(
+        maths_t::E_ContinuousData, TIME_SCALES[NUMBER_MODELS - 1] * decayRate)};
+}
+
 CNaiveBayes initialProbabilityOfChangeModel(double decayRate) {
-    decayRate *= TIME_SCALES[NUMBER_MODELS - 1];
-    return CNaiveBayes{CNaiveBayesFeatureDensityFromPrior{CNormalMeanPrecConjugate::nonInformativePrior(
-                           maths_t::E_ContinuousData, decayRate)},
-                       decayRate, -20.0};
+    return CNaiveBayes{naiveBayesExemplar(decayRate),
+                       TIME_SCALES[NUMBER_MODELS - 1] * decayRate, -20.0};
 }
 
 CNormalMeanPrecConjugate initialMagnitudeOfChangeModel(double decayRate) {
@@ -157,7 +160,8 @@ bool CTrendComponent::acceptRestoreTraverser(const SDistributionRestoreParams& p
         RESTORE(VALUE_MOMENTS_TAG, m_ValueMoments.fromDelimited(traverser.value()))
         RESTORE_BUILT_IN(TIME_OF_LAST_LEVEL_CHANGE_TAG, m_TimeOfLastLevelChange)
         RESTORE_NO_ERROR(PROBABILITY_OF_LEVEL_CHANGE_MODEL_TAG,
-                         m_ProbabilityOfLevelChangeModel = CNaiveBayes(params, traverser))
+                         m_ProbabilityOfLevelChangeModel = std::move(CNaiveBayes(
+                             naiveBayesExemplar(m_DefaultDecayRate), params, traverser)))
         RESTORE_NO_ERROR(MAGNITUDE_OF_LEVEL_CHANGE_MODEL_TAG,
                          m_MagnitudeOfLevelChangeModel =
                              CNormalMeanPrecConjugate(params, traverser))

lib/maths/unittest/CNaiveBayesTest.cc

@@ -306,7 +306,8 @@ void CNaiveBayesTest::testPersist() {
     core::CRapidXmlStateRestoreTraverser traverser(parser);
 
     maths::SDistributionRestoreParams params{maths_t::E_ContinuousData, 0.1, 0.0, 0.0, 0.0};
-    maths::CNaiveBayes restoredNb{params, traverser};
+    maths::CNaiveBayes restoredNb{maths::CNaiveBayesFeatureDensityFromPrior(normal),
+                                  params, traverser};
 
     CPPUNIT_ASSERT_EQUAL(origNb.checksum(), restoredNb.checksum());