12345qiupeng
/
gtsam
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge pull request #400 from borglab/fix/discreteBT 

Fix discrete BayesTree
release/4.3a0
Fan Jiang 2020-07-13 00:46:00 -04:00 committed by GitHub
parent 05ad89355d 80b42dcbef
commit 0a1a7510f9
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
9 changed files with 327 additions and 320 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
gtsam/discrete/DiscreteBayesNet.h
View File

@ -20,13 +20,14 @@
#include <vector>
#include <map>
#include <boost/shared_ptr.hpp>
#include <gtsam/inference/BayesNet.h>
#include <gtsam/inference/FactorGraph.h>
#include <gtsam/discrete/DiscreteConditional.h>
namespace gtsam {
/** A Bayes net made from linear-Discrete densities */
class GTSAM_EXPORT DiscreteBayesNet: public FactorGraph<DiscreteConditional>
class GTSAM_EXPORT DiscreteBayesNet: public BayesNet<DiscreteConditional>
{
public:

23
gtsam/discrete/DiscreteBayesTree.cpp
View File

@ -29,13 +29,32 @@ namespace gtsam {
template class BayesTreeCliqueBase<DiscreteBayesTreeClique, DiscreteFactorGraph>;
template class BayesTree<DiscreteBayesTreeClique>;
/* ************************************************************************* */
double DiscreteBayesTreeClique::evaluate(
const DiscreteConditional::Values& values) const {
// evaluate all conditionals and multiply
double result = (*conditional_)(values);
for (const auto& child : children) {
result *= child->evaluate(values);
}
return result;
}
/* ************************************************************************* */
bool DiscreteBayesTree::equals(const This& other, double tol) const
{
bool DiscreteBayesTree::equals(const This& other, double tol) const {
return Base::equals(other, tol);
}
/* ************************************************************************* */
double DiscreteBayesTree::evaluate(
const DiscreteConditional::Values& values) const {
double result = 1.0;
for (const auto& root : roots_) {
result *= root->evaluate(values);
}
return result;
}
} // \namespace gtsam

84
gtsam/discrete/DiscreteBayesTree.h
View File

@ -11,7 +11,8 @@
/**
* @file DiscreteBayesTree.h
* @brief Discrete Bayes Tree, the result of eliminating a DiscreteJunctionTree
* @brief Discrete Bayes Tree, the result of eliminating a
* DiscreteJunctionTree
* @brief DiscreteBayesTree
* @author Frank Dellaert
* @author Richard Roberts
@ -22,45 +23,62 @@
#include <gtsam/discrete/DiscreteBayesNet.h>
#include <gtsam/discrete/DiscreteFactorGraph.h>
#include <gtsam/inference/BayesTree.h>
#include <gtsam/inference/Conditional.h>
#include <gtsam/inference/BayesTreeCliqueBase.h>
#include <string>
namespace gtsam {
// Forward declarations
class DiscreteConditional;
class VectorValues;
// Forward declarations
class DiscreteConditional;
class VectorValues;
/* ************************************************************************* */
/** A clique in a DiscreteBayesTree */
class GTSAM_EXPORT DiscreteBayesTreeClique :
public BayesTreeCliqueBase<DiscreteBayesTreeClique, DiscreteFactorGraph>
{
public:
typedef DiscreteBayesTreeClique This;
typedef BayesTreeCliqueBase<DiscreteBayesTreeClique, DiscreteFactorGraph> Base;
typedef boost::shared_ptr<This> shared_ptr;
typedef boost::weak_ptr<This> weak_ptr;
DiscreteBayesTreeClique() {}
DiscreteBayesTreeClique(const boost::shared_ptr<DiscreteConditional>& conditional) : Base(conditional) {}
};
/* ************************************************************************* */
/** A clique in a DiscreteBayesTree */
class GTSAM_EXPORT DiscreteBayesTreeClique
: public BayesTreeCliqueBase<DiscreteBayesTreeClique, DiscreteFactorGraph> {
public:
typedef DiscreteBayesTreeClique This;
typedef BayesTreeCliqueBase<DiscreteBayesTreeClique, DiscreteFactorGraph>
Base;
typedef boost::shared_ptr<This> shared_ptr;
typedef boost::weak_ptr<This> weak_ptr;
DiscreteBayesTreeClique() {}
DiscreteBayesTreeClique(
const boost::shared_ptr<DiscreteConditional>& conditional)
: Base(conditional) {}
/* ************************************************************************* */
/** A Bayes tree representing a Discrete density */
class GTSAM_EXPORT DiscreteBayesTree :
public BayesTree<DiscreteBayesTreeClique>
{
private:
typedef BayesTree<DiscreteBayesTreeClique> Base;
/// print index signature only
void printSignature(
const std::string& s = "Clique: ",
const KeyFormatter& formatter = DefaultKeyFormatter) const {
conditional_->printSignature(s, formatter);
}
public:
typedef DiscreteBayesTree This;
typedef boost::shared_ptr<This> shared_ptr;
//** evaluate conditional probability of subtree for given Values */
double evaluate(const DiscreteConditional::Values& values) const;
};
/** Default constructor, creates an empty Bayes tree */
DiscreteBayesTree() {}
/* ************************************************************************* */
/** A Bayes tree representing a Discrete density */
class GTSAM_EXPORT DiscreteBayesTree
: public BayesTree<DiscreteBayesTreeClique> {
private:
typedef BayesTree<DiscreteBayesTreeClique> Base;
/** Check equality */
bool equals(const This& other, double tol = 1e-9) const;
};
public:
typedef DiscreteBayesTree This;
typedef boost::shared_ptr<This> shared_ptr;
}
/** Default constructor, creates an empty Bayes tree */
DiscreteBayesTree() {}
/** Check equality */
bool equals(const This& other, double tol = 1e-9) const;
//** evaluate probability for given Values */
double evaluate(const DiscreteConditional::Values& values) const;
};
} // namespace gtsam

9
gtsam/discrete/DiscreteConditional.h
View File

@ -24,6 +24,8 @@
#include <boost/shared_ptr.hpp>
#include <boost/make_shared.hpp>
#include <string>
namespace gtsam {
/**
@ -92,6 +94,13 @@ public:
/// @name Standard Interface
/// @{
/// print index signature only
void printSignature(
const std::string& s = "Discrete Conditional: ",
const KeyFormatter& formatter = DefaultKeyFormatter) const {
static_cast<const BaseConditional*>(this)->print(s, formatter);
}
/// Evaluate, just look up in AlgebraicDecisonTree
virtual double operator()(const Values& values) const {
return Potentials::operator()(values);

470
gtsam/discrete/tests/testDiscreteBayesTree.cpp
View File

@ -1,261 +1,216 @@
///* ----------------------------------------------------------------------------
//
// * GTSAM Copyright 2010, Georgia Tech Research Corporation,
// * Atlanta, Georgia 30332-0415
// * All Rights Reserved
// * Authors: Frank Dellaert, et al. (see THANKS for the full author list)
//
// * See LICENSE for the license information
//
// * -------------------------------------------------------------------------- */
//
///*
// * @file testDiscreteBayesTree.cpp
// * @date sept 15, 2012
// * @author Frank Dellaert
// */
//
//#include <gtsam/discrete/DiscreteBayesNet.h>
//#include <gtsam/discrete/DiscreteBayesTree.h>
//#include <gtsam/discrete/DiscreteFactorGraph.h>
//
//#include <boost/assign/std/vector.hpp>
//using namespace boost::assign;
//
/* ----------------------------------------------------------------------------
* GTSAM Copyright 2010-2020, Georgia Tech Research Corporation,
* Atlanta, Georgia 30332-0415
* All Rights Reserved
* Authors: Frank Dellaert, et al. (see THANKS for the full author list)
* See LICENSE for the license information
* -------------------------------------------------------------------------- */
/*
* @file testDiscreteBayesTree.cpp
* @date sept 15, 2012
* @author Frank Dellaert
*/
#include <gtsam/base/Vector.h>
#include <gtsam/discrete/DiscreteBayesNet.h>
#include <gtsam/discrete/DiscreteBayesTree.h>
#include <gtsam/discrete/DiscreteFactorGraph.h>
#include <gtsam/inference/BayesNet-inst.h>
#include <boost/assign/std/vector.hpp>
using namespace boost::assign;
#include <CppUnitLite/TestHarness.h>
//
//using namespace std;
//using namespace gtsam;
//
//static bool debug = false;
//
///**
// * Custom clique class to debug shortcuts
// */
////class Clique: public BayesTreeCliqueBaseOrdered<Clique, DiscreteConditional> {
////
////protected:
////
////public:
////
//// typedef BayesTreeCliqueBaseOrdered<Clique, DiscreteConditional> Base;
//// typedef boost::shared_ptr<Clique> shared_ptr;
////
//// // Constructors
//// Clique() {
//// }
//// Clique(const DiscreteConditional::shared_ptr& conditional) :
//// Base(conditional) {
//// }
//// Clique(
//// const std::pair<DiscreteConditional::shared_ptr,
//// DiscreteConditional::FactorType::shared_ptr>& result) :
//// Base(result) {
//// }
////
//// /// print index signature only
//// void printSignature(const std::string& s = "Clique: ",
//// const KeyFormatter& indexFormatter = DefaultKeyFormatter) const {
//// ((IndexConditionalOrdered::shared_ptr) conditional_)->print(s, indexFormatter);
//// }
////
//// /// evaluate value of sub-tree
//// double evaluate(const DiscreteConditional::Values & values) {
//// double result = (*(this->conditional_))(values);
//// // evaluate all children and multiply into result
//// for(boost::shared_ptr<Clique> c: children_)
//// result *= c->evaluate(values);
//// return result;
//// }
////
////};
//
////typedef BayesTreeOrdered<DiscreteConditional, Clique> DiscreteBayesTree;
////
/////* ************************************************************************* */
////double evaluate(const DiscreteBayesTree& tree,
//// const DiscreteConditional::Values & values) {
//// return tree.root()->evaluate(values);
////}
//
///* ************************************************************************* */
//
//TEST_UNSAFE( DiscreteBayesTree, thinTree ) {
//
// const int nrNodes = 15;
// const size_t nrStates = 2;
//
// // define variables
// vector<DiscreteKey> key;
// for (int i = 0; i < nrNodes; i++) {
// DiscreteKey key_i(i, nrStates);
// key.push_back(key_i);
// }
//
// // create a thin-tree Bayesnet, a la Jean-Guillaume
// DiscreteBayesNet bayesNet;
// bayesNet.add(key[14] % "1/3");
//
// bayesNet.add(key[13] | key[14] = "1/3 3/1");
// bayesNet.add(key[12] | key[14] = "3/1 3/1");
//
// bayesNet.add((key[11] | key[13], key[14]) = "1/4 2/3 3/2 4/1");
// bayesNet.add((key[10] | key[13], key[14]) = "1/4 3/2 2/3 4/1");
// bayesNet.add((key[9] | key[12], key[14]) = "4/1 2/3 F 1/4");
// bayesNet.add((key[8] | key[12], key[14]) = "T 1/4 3/2 4/1");
//
// bayesNet.add((key[7] | key[11], key[13]) = "1/4 2/3 3/2 4/1");
// bayesNet.add((key[6] | key[11], key[13]) = "1/4 3/2 2/3 4/1");
// bayesNet.add((key[5] | key[10], key[13]) = "4/1 2/3 3/2 1/4");
// bayesNet.add((key[4] | key[10], key[13]) = "2/3 1/4 3/2 4/1");
//
// bayesNet.add((key[3] | key[9], key[12]) = "1/4 2/3 3/2 4/1");
// bayesNet.add((key[2] | key[9], key[12]) = "1/4 8/2 2/3 4/1");
// bayesNet.add((key[1] | key[8], key[12]) = "4/1 2/3 3/2 1/4");
// bayesNet.add((key[0] | key[8], key[12]) = "2/3 1/4 3/2 4/1");
//
//// if (debug) {
//// GTSAM_PRINT(bayesNet);
//// bayesNet.saveGraph("/tmp/discreteBayesNet.dot");
//// }
//
// // create a BayesTree out of a Bayes net
// DiscreteBayesTree bayesTree(bayesNet);
// if (debug) {
// GTSAM_PRINT(bayesTree);
// bayesTree.saveGraph("/tmp/discreteBayesTree.dot");
// }
//
// // Check whether BN and BT give the same answer on all configurations
// // Also calculate all some marginals
// Vector marginals = zero(15);
// double joint_12_14 = 0, joint_9_12_14 = 0, joint_8_12_14 = 0, joint_8_12 = 0,
// joint82 = 0, joint12 = 0, joint24 = 0, joint45 = 0, joint46 = 0,
// joint_4_11 = 0;
// vector<DiscreteFactor::Values> allPosbValues = cartesianProduct(
// key[0] & key[1] & key[2] & key[3] & key[4] & key[5] & key[6] & key[7]
// & key[8] & key[9] & key[10] & key[11] & key[12] & key[13] & key[14]);
// for (size_t i = 0; i < allPosbValues.size(); ++i) {
// DiscreteFactor::Values x = allPosbValues[i];
// double expected = evaluate(bayesNet, x);
// double actual = evaluate(bayesTree, x);
// DOUBLES_EQUAL(expected, actual, 1e-9);
// // collect marginals
// for (size_t i = 0; i < 15; i++)
// if (x[i])
// marginals[i] += actual;
// // calculate shortcut 8 and 0
// if (x[12] && x[14])
// joint_12_14 += actual;
// if (x[9] && x[12] & x[14])
// joint_9_12_14 += actual;
// if (x[8] && x[12] & x[14])
// joint_8_12_14 += actual;
// if (x[8] && x[12])
// joint_8_12 += actual;
// if (x[8] && x[2])
// joint82 += actual;
// if (x[1] && x[2])
// joint12 += actual;
// if (x[2] && x[4])
// joint24 += actual;
// if (x[4] && x[5])
// joint45 += actual;
// if (x[4] && x[6])
// joint46 += actual;
// if (x[4] && x[11])
// joint_4_11 += actual;
// }
// DiscreteFactor::Values all1 = allPosbValues.back();
//
// Clique::shared_ptr R = bayesTree.root();
//
// // check separator marginal P(S0)
// Clique::shared_ptr c = bayesTree[0];
// DiscreteFactorGraph separatorMarginal0 = c->separatorMarginal(R,
// EliminateDiscrete);
// EXPECT_DOUBLES_EQUAL(joint_8_12, separatorMarginal0(all1), 1e-9);
//
// // check separator marginal P(S9), should be P(14)
// c = bayesTree[9];
// DiscreteFactorGraph separatorMarginal9 = c->separatorMarginal(R,
// EliminateDiscrete);
// EXPECT_DOUBLES_EQUAL(marginals[14], separatorMarginal9(all1), 1e-9);
//
// // check separator marginal of root, should be empty
// c = bayesTree[11];
// DiscreteFactorGraph separatorMarginal11 = c->separatorMarginal(R,
// EliminateDiscrete);
// EXPECT_LONGS_EQUAL(0, separatorMarginal11.size());
//
// // check shortcut P(S9||R) to root
// c = bayesTree[9];
// DiscreteBayesNet shortcut = c->shortcut(R, EliminateDiscrete);
// EXPECT_LONGS_EQUAL(0, shortcut.size());
//
// // check shortcut P(S8||R) to root
// c = bayesTree[8];
// shortcut = c->shortcut(R, EliminateDiscrete);
// EXPECT_DOUBLES_EQUAL(joint_12_14/marginals[14], evaluate(shortcut,all1),
// 1e-9);
//
// // check shortcut P(S2||R) to root
// c = bayesTree[2];
// shortcut = c->shortcut(R, EliminateDiscrete);
// EXPECT_DOUBLES_EQUAL(joint_9_12_14/marginals[14], evaluate(shortcut,all1),
// 1e-9);
//
// // check shortcut P(S0||R) to root
// c = bayesTree[0];
// shortcut = c->shortcut(R, EliminateDiscrete);
// EXPECT_DOUBLES_EQUAL(joint_8_12_14/marginals[14], evaluate(shortcut,all1),
// 1e-9);
//
// // calculate all shortcuts to root
// DiscreteBayesTree::Nodes cliques = bayesTree.nodes();
// for(Clique::shared_ptr c: cliques) {
// DiscreteBayesNet shortcut = c->shortcut(R, EliminateDiscrete);
// if (debug) {
// c->printSignature();
// shortcut.print("shortcut:");
// }
// }
//
// // Check all marginals
// DiscreteFactor::shared_ptr marginalFactor;
// for (size_t i = 0; i < 15; i++) {
// marginalFactor = bayesTree.marginalFactor(i, EliminateDiscrete);
// double actual = (*marginalFactor)(all1);
// EXPECT_DOUBLES_EQUAL(marginals[i], actual, 1e-9);
// }
//
// DiscreteBayesNet::shared_ptr actualJoint;
//
// // Check joint P(8,2) TODO: not disjoint !
//// actualJoint = bayesTree.jointBayesNet(8, 2, EliminateDiscrete);
//// EXPECT_DOUBLES_EQUAL(joint82, evaluate(*actualJoint,all1), 1e-9);
//
// // Check joint P(1,2) TODO: not disjoint !
//// actualJoint = bayesTree.jointBayesNet(1, 2, EliminateDiscrete);
//// EXPECT_DOUBLES_EQUAL(joint12, evaluate(*actualJoint,all1), 1e-9);
//
// // Check joint P(2,4)
// actualJoint = bayesTree.jointBayesNet(2, 4, EliminateDiscrete);
// EXPECT_DOUBLES_EQUAL(joint24, evaluate(*actualJoint,all1), 1e-9);
//
// // Check joint P(4,5) TODO: not disjoint !
//// actualJoint = bayesTree.jointBayesNet(4, 5, EliminateDiscrete);
//// EXPECT_DOUBLES_EQUAL(joint46, evaluate(*actualJoint,all1), 1e-9);
//
// // Check joint P(4,6) TODO: not disjoint !
//// actualJoint = bayesTree.jointBayesNet(4, 6, EliminateDiscrete);
//// EXPECT_DOUBLES_EQUAL(joint46, evaluate(*actualJoint,all1), 1e-9);
//
// // Check joint P(4,11)
// actualJoint = bayesTree.jointBayesNet(4, 11, EliminateDiscrete);
// EXPECT_DOUBLES_EQUAL(joint_4_11, evaluate(*actualJoint,all1), 1e-9);
//
//}
#include <vector>
using namespace std;
using namespace gtsam;
static bool debug = false;
/* ************************************************************************* */
TEST_UNSAFE(DiscreteBayesTree, ThinTree) {
const int nrNodes = 15;
const size_t nrStates = 2;
// define variables
vector<DiscreteKey> key;
for (int i = 0; i < nrNodes; i++) {
DiscreteKey key_i(i, nrStates);
key.push_back(key_i);
}
// create a thin-tree Bayesnet, a la Jean-Guillaume
DiscreteBayesNet bayesNet;
bayesNet.add(key[14] % "1/3");
bayesNet.add(key[13] | key[14] = "1/3 3/1");
bayesNet.add(key[12] | key[14] = "3/1 3/1");
bayesNet.add((key[11] | key[13], key[14]) = "1/4 2/3 3/2 4/1");
bayesNet.add((key[10] | key[13], key[14]) = "1/4 3/2 2/3 4/1");
bayesNet.add((key[9] | key[12], key[14]) = "4/1 2/3 F 1/4");
bayesNet.add((key[8] | key[12], key[14]) = "T 1/4 3/2 4/1");
bayesNet.add((key[7] | key[11], key[13]) = "1/4 2/3 3/2 4/1");
bayesNet.add((key[6] | key[11], key[13]) = "1/4 3/2 2/3 4/1");
bayesNet.add((key[5] | key[10], key[13]) = "4/1 2/3 3/2 1/4");
bayesNet.add((key[4] | key[10], key[13]) = "2/3 1/4 3/2 4/1");
bayesNet.add((key[3] | key[9], key[12]) = "1/4 2/3 3/2 4/1");
bayesNet.add((key[2] | key[9], key[12]) = "1/4 8/2 2/3 4/1");
bayesNet.add((key[1] | key[8], key[12]) = "4/1 2/3 3/2 1/4");
bayesNet.add((key[0] | key[8], key[12]) = "2/3 1/4 3/2 4/1");
if (debug) {
GTSAM_PRINT(bayesNet);
bayesNet.saveGraph("/tmp/discreteBayesNet.dot");
}
// create a BayesTree out of a Bayes net
auto bayesTree = DiscreteFactorGraph(bayesNet).eliminateMultifrontal();
if (debug) {
GTSAM_PRINT(*bayesTree);
bayesTree->saveGraph("/tmp/discreteBayesTree.dot");
}
auto R = bayesTree->roots().front();
// Check whether BN and BT give the same answer on all configurations
vector<DiscreteFactor::Values> allPosbValues = cartesianProduct(
key[0] & key[1] & key[2] & key[3] & key[4] & key[5] & key[6] & key[7] &
key[8] & key[9] & key[10] & key[11] & key[12] & key[13] & key[14]);
for (size_t i = 0; i < allPosbValues.size(); ++i) {
DiscreteFactor::Values x = allPosbValues[i];
double expected = bayesNet.evaluate(x);
double actual = bayesTree->evaluate(x);
DOUBLES_EQUAL(expected, actual, 1e-9);
}
// Calculate all some marginals for Values==all1
Vector marginals = Vector::Zero(15);
double joint_12_14 = 0, joint_9_12_14 = 0, joint_8_12_14 = 0, joint_8_12 = 0,
joint82 = 0, joint12 = 0, joint24 = 0, joint45 = 0, joint46 = 0,
joint_4_11 = 0, joint_11_13 = 0, joint_11_13_14 = 0,
joint_11_12_13_14 = 0, joint_9_11_12_13 = 0, joint_8_11_12_13 = 0;
for (size_t i = 0; i < allPosbValues.size(); ++i) {
DiscreteFactor::Values x = allPosbValues[i];
double px = bayesTree->evaluate(x);
for (size_t i = 0; i < 15; i++)
if (x[i]) marginals[i] += px;
if (x[12] && x[14]) joint_12_14 += px;
if (x[9] && x[12] && x[14]) joint_9_12_14 += px;
if (x[8] && x[12] && x[14]) joint_8_12_14 += px;
if (x[8] && x[12]) joint_8_12 += px;
if (x[8] && x[2]) joint82 += px;
if (x[1] && x[2]) joint12 += px;
if (x[2] && x[4]) joint24 += px;
if (x[4] && x[5]) joint45 += px;
if (x[4] && x[6]) joint46 += px;
if (x[4] && x[11]) joint_4_11 += px;
if (x[11] && x[13]) {
joint_11_13 += px;
if (x[8] && x[12]) joint_8_11_12_13 += px;
if (x[9] && x[12]) joint_9_11_12_13 += px;
if (x[14]) {
joint_11_13_14 += px;
if (x[12]) {
joint_11_12_13_14 += px;
}
}
}
}
DiscreteFactor::Values all1 = allPosbValues.back();
// check separator marginal P(S0)
auto c = (*bayesTree)[0];
DiscreteFactorGraph separatorMarginal0 =
c->separatorMarginal(EliminateDiscrete);
DOUBLES_EQUAL(joint_8_12, separatorMarginal0(all1), 1e-9);
// check separator marginal P(S9), should be P(14)
c = (*bayesTree)[9];
DiscreteFactorGraph separatorMarginal9 =
c->separatorMarginal(EliminateDiscrete);
DOUBLES_EQUAL(marginals[14], separatorMarginal9(all1), 1e-9);
// check separator marginal of root, should be empty
c = (*bayesTree)[11];
DiscreteFactorGraph separatorMarginal11 =
c->separatorMarginal(EliminateDiscrete);
LONGS_EQUAL(0, separatorMarginal11.size());
// check shortcut P(S9||R) to root
c = (*bayesTree)[9];
DiscreteBayesNet shortcut = c->shortcut(R, EliminateDiscrete);
LONGS_EQUAL(1, shortcut.size());
DOUBLES_EQUAL(joint_11_13_14 / joint_11_13, shortcut.evaluate(all1), 1e-9);
// check shortcut P(S8||R) to root
c = (*bayesTree)[8];
shortcut = c->shortcut(R, EliminateDiscrete);
DOUBLES_EQUAL(joint_11_12_13_14 / joint_11_13, shortcut.evaluate(all1), 1e-9);
// check shortcut P(S2||R) to root
c = (*bayesTree)[2];
shortcut = c->shortcut(R, EliminateDiscrete);
DOUBLES_EQUAL(joint_9_11_12_13 / joint_11_13, shortcut.evaluate(all1), 1e-9);
// check shortcut P(S0||R) to root
c = (*bayesTree)[0];
shortcut = c->shortcut(R, EliminateDiscrete);
DOUBLES_EQUAL(joint_8_11_12_13 / joint_11_13, shortcut.evaluate(all1), 1e-9);
// calculate all shortcuts to root
DiscreteBayesTree::Nodes cliques = bayesTree->nodes();
for (auto c : cliques) {
DiscreteBayesNet shortcut = c.second->shortcut(R, EliminateDiscrete);
if (debug) {
c.second->conditional_->printSignature();
shortcut.print("shortcut:");
}
}
// Check all marginals
DiscreteFactor::shared_ptr marginalFactor;
for (size_t i = 0; i < 15; i++) {
marginalFactor = bayesTree->marginalFactor(i, EliminateDiscrete);
double actual = (*marginalFactor)(all1);
DOUBLES_EQUAL(marginals[i], actual, 1e-9);
}
DiscreteBayesNet::shared_ptr actualJoint;
// Check joint P(8, 2)
actualJoint = bayesTree->jointBayesNet(8, 2, EliminateDiscrete);
DOUBLES_EQUAL(joint82, actualJoint->evaluate(all1), 1e-9);
// Check joint P(1, 2)
actualJoint = bayesTree->jointBayesNet(1, 2, EliminateDiscrete);
DOUBLES_EQUAL(joint12, actualJoint->evaluate(all1), 1e-9);
// Check joint P(2, 4)
actualJoint = bayesTree->jointBayesNet(2, 4, EliminateDiscrete);
DOUBLES_EQUAL(joint24, actualJoint->evaluate(all1), 1e-9);
// Check joint P(4, 5)
actualJoint = bayesTree->jointBayesNet(4, 5, EliminateDiscrete);
DOUBLES_EQUAL(joint45, actualJoint->evaluate(all1), 1e-9);
// Check joint P(4, 6)
actualJoint = bayesTree->jointBayesNet(4, 6, EliminateDiscrete);
DOUBLES_EQUAL(joint46, actualJoint->evaluate(all1), 1e-9);
// Check joint P(4, 11)
actualJoint = bayesTree->jointBayesNet(4, 11, EliminateDiscrete);
DOUBLES_EQUAL(joint_4_11, actualJoint->evaluate(all1), 1e-9);
}
/* ************************************************************************* */
int main() {
@ -263,4 +218,3 @@ int main() {
return TestRegistry::runAllTests(tr);
}
/* ************************************************************************* */

BIN
gtsam/discrete/tests/testDiscreteBayesTree.pdf Normal file
View File

Binary file not shown.

1
gtsam/discrete/tests/testDiscreteFactorGraph.cpp
View File

@ -19,6 +19,7 @@
#include <gtsam/discrete/DiscreteFactorGraph.h>
#include <gtsam/discrete/DiscreteEliminationTree.h>
#include <gtsam/discrete/DiscreteBayesTree.h>
#include <gtsam/inference/BayesNet-inst.h>
#include <CppUnitLite/TestHarness.h>

54
gtsam/inference/BayesTreeCliqueBase-inst.h
View File

@ -136,57 +136,61 @@ namespace gtsam {
}
}
/* ************************************************************************* */
/* *********************************************************************** */
// separator marginal, uses separator marginal of parent recursively
// P(C) = P(F|S) P(S)
/* ************************************************************************* */
template<class DERIVED, class FACTORGRAPH>
/* *********************************************************************** */
template <class DERIVED, class FACTORGRAPH>
typename BayesTreeCliqueBase<DERIVED, FACTORGRAPH>::FactorGraphType
BayesTreeCliqueBase<DERIVED, FACTORGRAPH>::separatorMarginal(Eliminate function) const
{
BayesTreeCliqueBase<DERIVED, FACTORGRAPH>::separatorMarginal(
Eliminate function) const {
gttic(BayesTreeCliqueBase_separatorMarginal);
// Check if the Separator marginal was already calculated
if (!cachedSeparatorMarginal_)
{
if (!cachedSeparatorMarginal_) {
gttic(BayesTreeCliqueBase_separatorMarginal_cachemiss);
// If this is the root, there is no separator
if (parent_.expired() /*(if we're the root)*/)
{
if (parent_.expired() /*(if we're the root)*/) {
// we are root, return empty
FactorGraphType empty;
cachedSeparatorMarginal_ = empty;
}
else
{
} else {
// Flatten recursion in timing outline
gttoc(BayesTreeCliqueBase_separatorMarginal_cachemiss);
gttoc(BayesTreeCliqueBase_separatorMarginal);
// Obtain P(S) = \int P(Cp) = \int P(Fp|Sp) P(Sp)
// initialize P(Cp) with the parent separator marginal
derived_ptr parent(parent_.lock());
gttoc(BayesTreeCliqueBase_separatorMarginal_cachemiss); // Flatten recursion in timing outline
gttoc(BayesTreeCliqueBase_separatorMarginal);
FactorGraphType p_Cp(parent->separatorMarginal(function)); // P(Sp)
FactorGraphType p_Cp(parent->separatorMarginal(function)); // P(Sp)
gttic(BayesTreeCliqueBase_separatorMarginal);
gttic(BayesTreeCliqueBase_separatorMarginal_cachemiss);
// now add the parent conditional
p_Cp += parent->conditional_; // P(Fp|Sp)
p_Cp += parent->conditional_; // P(Fp|Sp)
// The variables we want to keepSet are exactly the ones in S
KeyVector indicesS(this->conditional()->beginParents(), this->conditional()->endParents());
cachedSeparatorMarginal_ = *p_Cp.marginalMultifrontalBayesNet(Ordering(indicesS), function);
KeyVector indicesS(this->conditional()->beginParents(),
this->conditional()->endParents());
auto separatorMarginal =
p_Cp.marginalMultifrontalBayesNet(Ordering(indicesS), function);
cachedSeparatorMarginal_.reset(*separatorMarginal);
}
}
// return the shortcut P(S||B)
return *cachedSeparatorMarginal_; // return the cached version
return *cachedSeparatorMarginal_; // return the cached version
}
/* ************************************************************************* */
// marginal2, uses separator marginal of parent recursively
/* *********************************************************************** */
// marginal2, uses separator marginal of parent
// P(C) = P(F|S) P(S)
/* ************************************************************************* */
template<class DERIVED, class FACTORGRAPH>
/* *********************************************************************** */
template <class DERIVED, class FACTORGRAPH>
typename BayesTreeCliqueBase<DERIVED, FACTORGRAPH>::FactorGraphType
BayesTreeCliqueBase<DERIVED, FACTORGRAPH>::marginal2(Eliminate function) const
{
BayesTreeCliqueBase<DERIVED, FACTORGRAPH>::marginal2(
Eliminate function) const {
gttic(BayesTreeCliqueBase_marginal2);
// initialize with separator marginal P(S)
FactorGraphType p_C = this->separatorMarginal(function);

3
gtsam/inference/Conditional.h
View File

@ -65,6 +65,8 @@ namespace gtsam {
Conditional(size_t nrFrontals) : nrFrontals_(nrFrontals) {}
/// @}
public:
/// @name Testable
/// @{
@ -76,7 +78,6 @@ namespace gtsam {
/// @}
public:
/// @name Standard Interface
/// @{