Merge pull request #1893 from borglab/fix-pruning

gtsam/hybrid/HybridBayesNet.cpp

@@ -197,6 +197,30 @@ AlgebraicDecisionTree<Key> HybridBayesNet::errorTree(
   return result;
 }
 
+/* ************************************************************************* */
+double HybridBayesNet::negLogConstant(
+    const std::optional<DiscreteValues> &discrete) const {
+  double negLogNormConst = 0.0;
+  // Iterate over each conditional.
+  for (auto &&conditional : *this) {
+    if (discrete.has_value()) {
+      if (auto gm = conditional->asHybrid()) {
+        negLogNormConst += gm->choose(*discrete)->negLogConstant();
+      } else if (auto gc = conditional->asGaussian()) {
+        negLogNormConst += gc->negLogConstant();
+      } else if (auto dc = conditional->asDiscrete()) {
+        negLogNormConst += dc->choose(*discrete)->negLogConstant();
+      } else {
+        throw std::runtime_error(
+            "Unknown conditional type when computing negLogConstant");
+      }
+    } else {
+      negLogNormConst += conditional->negLogConstant();
+    }
+  }
+  return negLogNormConst;
+}
+
 /* ************************************************************************* */
 AlgebraicDecisionTree<Key> HybridBayesNet::discretePosterior(
     const VectorValues &continuousValues) const {

gtsam/hybrid/HybridBayesNet.h

@@ -237,6 +237,16 @@ class GTSAM_EXPORT HybridBayesNet : public BayesNet<HybridConditional> {
 
   using BayesNet::logProbability;  // expose HybridValues version
 
+  /**
+   * @brief Get the negative log of the normalization constant
+   * corresponding to the joint density represented by this Bayes net.
+   * Optionally index by `discrete`.
+   *
+   * @param discrete Optional DiscreteValues
+   * @return double
+   */
+  double negLogConstant(const std::optional<DiscreteValues> &discrete) const;
+
   /**
    * @brief Compute normalized posterior P(M|X=x) and return as a tree.
    *

gtsam/hybrid/HybridGaussianConditional.cpp

@@ -322,8 +322,11 @@ HybridGaussianConditional::shared_ptr HybridGaussianConditional::prune(
           const GaussianFactorValuePair &pair) -> GaussianFactorValuePair {
     if (max->evaluate(choices) == 0.0)
       return {nullptr, std::numeric_limits<double>::infinity()};
-    else
-      return pair;
+    else {
+      // Add negLogConstant_ back so that the minimum negLogConstant in the
+      // HybridGaussianConditional is set correctly.
+      return {pair.first, pair.second + negLogConstant_};
+    }
   };
 
   FactorValuePairs prunedConditionals = factors().apply(pruner);

gtsam/hybrid/HybridGaussianFactorGraph.cpp

@@ -59,10 +59,11 @@ using OrphanWrapper = BayesTreeOrphanWrapper<HybridBayesTree::Clique>;
 
 /// Result from elimination.
 struct Result {
+  // Gaussian conditional resulting from elimination.
   GaussianConditional::shared_ptr conditional;
-  double negLogK;
-  GaussianFactor::shared_ptr factor;
-  double scalar;
+  double negLogK;  // Negative log of the normalization constant K.
+  GaussianFactor::shared_ptr factor;  // Leftover factor 𝜏.
+  double scalar;                      // Scalar value associated with factor 𝜏.
 
   bool operator==(const Result &other) const {
     return conditional == other.conditional && negLogK == other.negLogK &&

gtsam/hybrid/tests/testHybridBayesNet.cpp

@@ -363,10 +363,6 @@ TEST(HybridBayesNet, Pruning) {
   AlgebraicDecisionTree<Key> expected(s.modes, leaves);
   EXPECT(assert_equal(expected, discretePosterior, 1e-6));
 
-  // Prune and get probabilities
-  auto prunedBayesNet = posterior->prune(2);
-  auto prunedTree = prunedBayesNet.discretePosterior(delta.continuous());
-
   // Verify logProbability computation and check specific logProbability value
   const DiscreteValues discrete_values{{M(0), 1}, {M(1), 1}};
   const HybridValues hybridValues{delta.continuous(), discrete_values};
@@ -381,10 +377,21 @@ TEST(HybridBayesNet, Pruning) {
   EXPECT_DOUBLES_EQUAL(logProbability, posterior->logProbability(hybridValues),
                        1e-9);
 
+  double negLogConstant = posterior->negLogConstant(discrete_values);
+
+  // The sum of all the mode densities
+  double normalizer =
+      AlgebraicDecisionTree<Key>(posterior->errorTree(delta.continuous()),
+                                 [](double error) { return exp(-error); })
+          .sum();
+
   // Check agreement with discrete posterior
-  // double density = exp(logProbability);
-  // FAILS: EXPECT_DOUBLES_EQUAL(density, discretePosterior(discrete_values),
-  // 1e-6);
+  double density = exp(logProbability + negLogConstant) / normalizer;
+  EXPECT_DOUBLES_EQUAL(density, discretePosterior(discrete_values), 1e-6);
+
+  // Prune and get probabilities
+  auto prunedBayesNet = posterior->prune(2);
+  auto prunedTree = prunedBayesNet.discretePosterior(delta.continuous());
 
   // Regression test on pruned logProbability tree
   std::vector<double> pruned_leaves = {0.0, 0.50758422, 0.0, 0.49241578};
@@ -392,7 +399,26 @@ TEST(HybridBayesNet, Pruning) {
   EXPECT(assert_equal(expected_pruned, prunedTree, 1e-6));
 
   // Regression
-  // FAILS: EXPECT_DOUBLES_EQUAL(density, prunedTree(discrete_values), 1e-9);
+  double pruned_logProbability = 0;
+  pruned_logProbability +=
+      prunedBayesNet.at(0)->asDiscrete()->logProbability(hybridValues);
+  pruned_logProbability +=
+      prunedBayesNet.at(1)->asHybrid()->logProbability(hybridValues);
+  pruned_logProbability +=
+      prunedBayesNet.at(2)->asHybrid()->logProbability(hybridValues);
+  pruned_logProbability +=
+      prunedBayesNet.at(3)->asHybrid()->logProbability(hybridValues);
+
+  double pruned_negLogConstant = prunedBayesNet.negLogConstant(discrete_values);
+
+  // The sum of all the mode densities
+  double pruned_normalizer =
+      AlgebraicDecisionTree<Key>(prunedBayesNet.errorTree(delta.continuous()),
+                                 [](double error) { return exp(-error); })
+          .sum();
+  double pruned_density =
+      exp(pruned_logProbability + pruned_negLogConstant) / pruned_normalizer;
+  EXPECT_DOUBLES_EQUAL(pruned_density, prunedTree(discrete_values), 1e-9);
 }
 
 /* ****************************************************************************/

gtsam/hybrid/tests/testHybridGaussianConditional.cpp

@@ -275,6 +275,11 @@ TEST(HybridGaussianConditional, Prune) {
 
     // Check that the pruned HybridGaussianConditional has 2 conditionals
     EXPECT_LONGS_EQUAL(2, pruned->nrComponents());
+
+    // Check that the minimum negLogConstant is set correctly
+    EXPECT_DOUBLES_EQUAL(
+        hgc.conditionals()({{M(1), 0}, {M(2), 1}})->negLogConstant(),
+        pruned->negLogConstant(), 1e-9);
   }
   {
     const std::vector<double> potentials{0.2, 0, 0.3, 0,  //
@@ -285,6 +290,9 @@ TEST(HybridGaussianConditional, Prune) {
 
     // Check that the pruned HybridGaussianConditional has 3 conditionals
     EXPECT_LONGS_EQUAL(3, pruned->nrComponents());
+
+    // Check that the minimum negLogConstant is correct
+    EXPECT_DOUBLES_EQUAL(hgc.negLogConstant(), pruned->negLogConstant(), 1e-9);
   }
 }