Make HybridFactorGraph just a FactorGraph<Factor> with extra methods

gtsam/hybrid/GaussianMixture.cpp

@@ -59,6 +59,8 @@ GaussianMixture::GaussianMixture(
                       Conditionals(discreteParents, conditionals)) {}
 
 /* *******************************************************************************/
+// TODO(dellaert): This is copy/paste: GaussianMixture should be derived from
+// GaussianMixtureFactor, no?
 GaussianFactorGraphTree GaussianMixture::add(
     const GaussianFactorGraphTree &sum) const {
   using Y = GraphAndConstant;

gtsam/hybrid/HybridFactorGraph.h

@@ -11,8 +11,9 @@
 
 /**
  * @file   HybridFactorGraph.h
- * @brief  Hybrid factor graph base class that uses type erasure
+ * @brief  Factor graph with utilities for hybrid factors.
  * @author Varun Agrawal
+ * @author Frank Dellaert
  * @date   May 28, 2022
  */
 
@@ -31,13 +32,11 @@ using SharedFactor = boost::shared_ptr<Factor>;
 
 /**
  * Hybrid Factor Graph
- * -----------------------
+ * Factor graph with utilities for hybrid factors.
- * This is the base hybrid factor graph.
- * Everything inside needs to be hybrid factor or hybrid conditional.
  */
-class HybridFactorGraph : public FactorGraph<HybridFactor> {
+class HybridFactorGraph : public FactorGraph<Factor> {
  public:
-  using Base = FactorGraph<HybridFactor>;
+  using Base = FactorGraph<Factor>;
   using This = HybridFactorGraph;              ///< this class
   using shared_ptr = boost::shared_ptr<This>;  ///< shared_ptr to This
 
@@ -140,8 +139,10 @@ class HybridFactorGraph : public FactorGraph<HybridFactor> {
   const KeySet discreteKeys() const {
     KeySet discrete_keys;
     for (auto& factor : factors_) {
-      for (const DiscreteKey& k : factor->discreteKeys()) {
+      if (auto p = boost::dynamic_pointer_cast<HybridFactor>(factor)) {
-        discrete_keys.insert(k.first);
+        for (const DiscreteKey& k : p->discreteKeys()) {
+          discrete_keys.insert(k.first);
+        }
       }
     }
     return discrete_keys;
@@ -151,8 +152,10 @@ class HybridFactorGraph : public FactorGraph<HybridFactor> {
   const KeySet continuousKeys() const {
     KeySet keys;
     for (auto& factor : factors_) {
-      for (const Key& key : factor->continuousKeys()) {
+      if (auto p = boost::dynamic_pointer_cast<HybridFactor>(factor)) {
-        keys.insert(key);
+        for (const Key& key : p->continuousKeys()) {
+          keys.insert(key);
+        }
       }
     }
     return keys;

gtsam/hybrid/HybridGaussianFactorGraph.cpp

@@ -79,48 +79,47 @@ static GaussianFactorGraphTree addGaussian(
 }
 
 /* ************************************************************************ */
-// TODO(dellaert): Implementation-wise, it's probably more efficient to first
+// TODO(dellaert): it's probably more efficient to first collect the discrete
-// collect the discrete keys, and then loop over all assignments to populate a
+// keys, and then loop over all assignments to populate a vector.
-// vector.
 GaussianFactorGraphTree HybridGaussianFactorGraph::assembleGraphTree() const {
+  using boost::dynamic_pointer_cast;
+
   gttic(assembleGraphTree);
 
   GaussianFactorGraphTree result;
 
   for (auto &f : factors_) {
     // TODO(dellaert): just use a virtual method defined in HybridFactor.
-    if (f->isHybrid()) {
+    if (auto gm = dynamic_pointer_cast<GaussianMixtureFactor>(f)) {
-      if (auto gm = boost::dynamic_pointer_cast<GaussianMixtureFactor>(f)) {
+      result = gm->add(result);
+    } else if (auto hc = dynamic_pointer_cast<HybridConditional>(f)) {
+      if (auto gm = hc->asMixture()) {
         result = gm->add(result);
+      } else if (auto g = hc->asGaussian()) {
+        result = addGaussian(result, g);
+      } else {
+        // Has to be discrete.
+        continue;
       }
-      if (auto gm = boost::dynamic_pointer_cast<HybridConditional>(f)) {
+    } else if (auto gf = dynamic_pointer_cast<HybridGaussianFactor>(f)) {
-        result = gm->asMixture()->add(result);
+      result = addGaussian(result, gf->inner());
-      }
+    } else if (dynamic_pointer_cast<DiscreteFactor>(f) ||
-
+               dynamic_pointer_cast<HybridDiscreteFactor>(f)) {
-    } else if (f->isContinuous()) {
-      if (auto gf = boost::dynamic_pointer_cast<HybridGaussianFactor>(f)) {
-        result = addGaussian(result, gf->inner());
-      }
-      if (auto cg = boost::dynamic_pointer_cast<HybridConditional>(f)) {
-        result = addGaussian(result, cg->asGaussian());
-      }
-
-    } else if (f->isDiscrete()) {
       // Don't do anything for discrete-only factors
       // since we want to eliminate continuous values only.
       continue;
-
+    } else if (auto orphan = dynamic_pointer_cast<
-    } else {
+                   BayesTreeOrphanWrapper<HybridBayesTree::Clique>>(f)) {
       // We need to handle the case where the object is actually an
       // BayesTreeOrphanWrapper!
-      auto orphan = boost::dynamic_pointer_cast<
+      throw std::invalid_argument(
-          BayesTreeOrphanWrapper<HybridBayesTree::Clique>>(f);
+          "gtsam::assembleGraphTree: BayesTreeOrphanWrapper is not implemented "
-      if (!orphan) {
+          "yet.");
-        auto &fr = *f;
+    } else {
-        throw std::invalid_argument(
+      auto &fr = *f;
-            std::string("factor is discrete in continuous elimination ") +
+      throw std::invalid_argument(
-            demangle(typeid(fr).name()));
+          std::string("gtsam::assembleGraphTree: factor type not handled: ") +
-      }
+          demangle(typeid(fr).name()));
     }
   }
 
@@ -377,8 +376,8 @@ EliminateHybrid(const HybridGaussianFactorGraph &factors,
   // Build a map from keys to DiscreteKeys
   std::unordered_map<Key, DiscreteKey> mapFromKeyToDiscreteKey;
   for (auto &&factor : factors) {
-    if (!factor->isContinuous()) {
+    if (auto p = boost::dynamic_pointer_cast<HybridFactor>(factor)) {
-      for (auto &k : factor->discreteKeys()) {
+      for (auto &k : p->discreteKeys()) {
         mapFromKeyToDiscreteKey[k.first] = k;
       }
     }
@@ -451,12 +450,6 @@ void HybridGaussianFactorGraph::add(DecisionTreeFactor::shared_ptr factor) {
 /* ************************************************************************ */
 const Ordering HybridGaussianFactorGraph::getHybridOrdering() const {
   KeySet discrete_keys = discreteKeys();
-  for (auto &factor : factors_) {
-    for (const DiscreteKey &k : factor->discreteKeys()) {
-      discrete_keys.insert(k.first);
-    }
-  }
-
   const VariableIndex index(factors_);
   Ordering ordering = Ordering::ColamdConstrainedLast(
       index, KeyVector(discrete_keys.begin(), discrete_keys.end()), true);
@@ -466,25 +459,23 @@ const Ordering HybridGaussianFactorGraph::getHybridOrdering() const {
 /* ************************************************************************ */
 AlgebraicDecisionTree<Key> HybridGaussianFactorGraph::error(
     const VectorValues &continuousValues) const {
+  using boost::dynamic_pointer_cast;
+
   AlgebraicDecisionTree<Key> error_tree(0.0);
 
   // Iterate over each factor.
-  for (size_t idx = 0; idx < size(); idx++) {
+  for (auto &f : factors_) {
     // TODO(dellaert): just use a virtual method defined in HybridFactor.
     AlgebraicDecisionTree<Key> factor_error;
 
-    if (factors_.at(idx)->isHybrid()) {
+    if (auto gaussianMixture = dynamic_pointer_cast<GaussianMixtureFactor>(f)) {
-      // If factor is hybrid, select based on assignment.
-      GaussianMixtureFactor::shared_ptr gaussianMixture =
-          boost::static_pointer_cast<GaussianMixtureFactor>(factors_.at(idx));
       // Compute factor error and add it.
       error_tree = error_tree + gaussianMixture->error(continuousValues);
 
-    } else if (factors_.at(idx)->isContinuous()) {
+    } else if (auto hybridGaussianFactor =
+                   dynamic_pointer_cast<HybridGaussianFactor>(f)) {
       // If continuous only, get the (double) error
       // and add it to the error_tree
-      auto hybridGaussianFactor =
-          boost::static_pointer_cast<HybridGaussianFactor>(factors_.at(idx));
       GaussianFactor::shared_ptr gaussian = hybridGaussianFactor->inner();
 
       // Compute the error of the gaussian factor.
@@ -493,9 +484,16 @@ AlgebraicDecisionTree<Key> HybridGaussianFactorGraph::error(
       error_tree = error_tree.apply(
           [error](double leaf_value) { return leaf_value + error; });
 
-    } else if (factors_.at(idx)->isDiscrete()) {
+    } else if (dynamic_pointer_cast<DiscreteFactor>(f) ||
+               dynamic_pointer_cast<HybridDiscreteFactor>(f)) {
       // If factor at `idx` is discrete-only, we skip.
       continue;
+    } else {
+      auto &fr = *f;
+      throw std::invalid_argument(
+          std::string(
+              "HybridGaussianFactorGraph::error: factor type not handled: ") +
+          demangle(typeid(fr).name()));
     }
   }
 
@@ -506,7 +504,9 @@ AlgebraicDecisionTree<Key> HybridGaussianFactorGraph::error(
 double HybridGaussianFactorGraph::error(const HybridValues &values) const {
   double error = 0.0;
   for (auto &factor : factors_) {
-    error += factor->error(values);
+    if (auto p = boost::dynamic_pointer_cast<HybridFactor>(factor)) {
+      error += p->error(values);
+    }
   }
   return error;
 }

gtsam/hybrid/HybridJunctionTree.cpp

@@ -61,9 +61,11 @@ struct HybridConstructorTraversalData {
     parentData.junctionTreeNode->addChild(data.junctionTreeNode);
 
     // Add all the discrete keys in the hybrid factors to the current data
-    for (HybridFactor::shared_ptr& f : node->factors) {
+    for (const auto& f : node->factors) {
-      for (auto& k : f->discreteKeys()) {
+      if (auto p = boost::dynamic_pointer_cast<HybridFactor>(f)) {
-        data.discreteKeys.insert(k.first);
+        for (auto& k : p->discreteKeys()) {
+          data.discreteKeys.insert(k.first);
+        }
       }
     }
 

gtsam/hybrid/HybridNonlinearFactorGraph.cpp

@@ -50,47 +50,42 @@ void HybridNonlinearFactorGraph::print(const std::string& s,
 /* ************************************************************************* */
 HybridGaussianFactorGraph::shared_ptr HybridNonlinearFactorGraph::linearize(
     const Values& continuousValues) const {
+  using boost::dynamic_pointer_cast;
+
   // create an empty linear FG
   auto linearFG = boost::make_shared<HybridGaussianFactorGraph>();
 
   linearFG->reserve(size());
 
   // linearize all hybrid factors
-  for (auto&& factor : factors_) {
+  for (auto& f : factors_) {
     // First check if it is a valid factor
-    if (factor) {
+    if (!f) {
-      // Check if the factor is a hybrid factor.
+      // TODO(dellaert): why?
-      // It can be either a nonlinear MixtureFactor or a linear
-      // GaussianMixtureFactor.
-      if (factor->isHybrid()) {
-        // Check if it is a nonlinear mixture factor
-        if (auto nlmf = boost::dynamic_pointer_cast<MixtureFactor>(factor)) {
-          linearFG->push_back(nlmf->linearize(continuousValues));
-        } else {
-          linearFG->push_back(factor);
-        }
-
-        // Now check if the factor is a continuous only factor.
-      } else if (factor->isContinuous()) {
-        // In this case, we check if factor->inner() is nonlinear since
-        // HybridFactors wrap over continuous factors.
-        auto nlhf = boost::dynamic_pointer_cast<HybridNonlinearFactor>(factor);
-        if (auto nlf =
-                boost::dynamic_pointer_cast<NonlinearFactor>(nlhf->inner())) {
-          auto hgf = boost::make_shared<HybridGaussianFactor>(
-              nlf->linearize(continuousValues));
-          linearFG->push_back(hgf);
-        } else {
-          linearFG->push_back(factor);
-        }
-        // Finally if nothing else, we are discrete-only which doesn't need
-        // lineariztion.
-      } else {
-        linearFG->push_back(factor);
-      }
-
-    } else {
       linearFG->push_back(GaussianFactor::shared_ptr());
+      continue;
+    }
+    // Check if it is a nonlinear mixture factor
+    if (auto nlmf = dynamic_pointer_cast<MixtureFactor>(f)) {
+      const GaussianMixtureFactor::shared_ptr& gmf =
+          nlmf->linearize(continuousValues);
+      linearFG->push_back(gmf);
+    } else if (auto nlhf = dynamic_pointer_cast<HybridNonlinearFactor>(f)) {
+      // Nonlinear wrapper case:
+      const GaussianFactor::shared_ptr& gf =
+          nlhf->inner()->linearize(continuousValues);
+      const auto hgf = boost::make_shared<HybridGaussianFactor>(gf);
+      linearFG->push_back(hgf);
+    } else if (dynamic_pointer_cast<DiscreteFactor>(f) ||
+               dynamic_pointer_cast<HybridDiscreteFactor>(f)) {
+      // If discrete-only: doesn't need linearization.
+      linearFG->push_back(f);
+    } else {
+      auto& fr = *f;
+      throw std::invalid_argument(
+          std::string("HybridNonlinearFactorGraph::linearize: factor type "
+                      "not handled: ") +
+          demangle(typeid(fr).name()));
     }
   }
   return linearFG;

gtsam/hybrid/HybridNonlinearFactorGraph.h

@@ -47,11 +47,6 @@ class GTSAM_EXPORT HybridNonlinearFactorGraph : public HybridFactorGraph {
   using HasDerivedValueType = typename std::enable_if<
       std::is_base_of<HybridFactor, typename T::value_type>::value>::type;
 
-  /// Check if T has a pointer type derived from FactorType.
-  template <typename T>
-  using HasDerivedElementType = typename std::enable_if<std::is_base_of<
-      HybridFactor, typename T::value_type::element_type>::value>::type;
-
  public:
   using Base = HybridFactorGraph;
   using This = HybridNonlinearFactorGraph;     ///< this class
@@ -124,7 +119,7 @@ class GTSAM_EXPORT HybridNonlinearFactorGraph : public HybridFactorGraph {
    * copied)
    */
   template <typename CONTAINER>
-  HasDerivedElementType<CONTAINER> push_back(const CONTAINER& container) {
+  void push_back(const CONTAINER& container) {
     Base::push_back(container.begin(), container.end());
   }
 

gtsam/hybrid/tests/testHybridNonlinearFactorGraph.cpp

@@ -435,7 +435,7 @@ TEST(HybridFactorGraph, Full_Elimination) {
 
     DiscreteFactorGraph discrete_fg;
     // TODO(Varun) Make this a function of HybridGaussianFactorGraph?
-    for (HybridFactor::shared_ptr& factor : (*remainingFactorGraph_partial)) {
+    for (auto& factor : (*remainingFactorGraph_partial)) {
       auto df = dynamic_pointer_cast<HybridDiscreteFactor>(factor);
       discrete_fg.push_back(df->inner());
     }