gtsam/gtsam/linear/tests/testGaussianBayesTree.cpp

350 lines
15 KiB
C++

/* ----------------------------------------------------------------------------
* 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 testGaussianBayesTree.cpp
* @date Jul 8, 2010
* @author Kai Ni
*/
#include <CppUnitLite/TestHarness.h>
#include <gtsam/base/debug.h>
#include <gtsam/base/numericalDerivative.h>
#include <gtsam/inference/Symbol.h>
#include <gtsam/linear/GaussianBayesTree.h>
#include <gtsam/linear/GaussianConditional.h>
#include <gtsam/linear/GaussianJunctionTree.h>
#include <iostream>
#include <vector>
using namespace std::placeholders;
using namespace std;
using namespace gtsam;
using Pairs = std::vector<std::pair<Key, Matrix>>;
namespace {
const Key x1=1, x2=2, x3=3, x4=4;
const SharedDiagonal chainNoise = noiseModel::Isotropic::Sigma(1, 0.5);
const GaussianFactorGraph chain = {
std::make_shared<JacobianFactor>(
x2, I_1x1, x1, I_1x1, (Vector(1) << 1.).finished(), chainNoise),
std::make_shared<JacobianFactor>(
x2, I_1x1, x3, I_1x1, (Vector(1) << 1.).finished(), chainNoise),
std::make_shared<JacobianFactor>(
x3, I_1x1, x4, I_1x1, (Vector(1) << 1.).finished(), chainNoise),
std::make_shared<JacobianFactor>(
x4, I_1x1, (Vector(1) << 1.).finished(), chainNoise)};
const Ordering chainOrdering {x2, x1, x3, x4};
/* ************************************************************************* */
// Helper functions for below
GaussianBayesTreeClique::shared_ptr LeafClique(const GaussianConditional& conditional)
{
return std::make_shared<GaussianBayesTreeClique>(
std::make_shared<GaussianConditional>(conditional));
}
typedef std::vector<GaussianBayesTreeClique::shared_ptr> Children;
GaussianBayesTreeClique::shared_ptr MakeClique(
const GaussianConditional& conditional, const Children& children)
{
auto clique = LeafClique(conditional);
clique->children.assign(children.begin(), children.end());
for(Children::const_iterator child = children.begin(); child != children.end(); ++child)
(*child)->parent_ = clique;
return clique;
}
}
/* ************************************************************************* */
/**
* x1 - x2 - x3 - x4
* x3 x4
* x2 x1 : x3
*
* x2 x1 x3 x4 b
* 1 1 1
* 1 1 1
* 1 1 1
* 1 1
*
* 1 0 0 1
*/
TEST( GaussianBayesTree, eliminate )
{
GaussianBayesTree bt = *chain.eliminateMultifrontal(chainOrdering);
Scatter scatter(chain);
EXPECT_LONGS_EQUAL(4, scatter.size());
EXPECT_LONGS_EQUAL(1, scatter.at(0).key);
EXPECT_LONGS_EQUAL(2, scatter.at(1).key);
EXPECT_LONGS_EQUAL(3, scatter.at(2).key);
EXPECT_LONGS_EQUAL(4, scatter.at(3).key);
const Matrix two = (Matrix(1, 1) << 2.).finished();
const Matrix one = I_1x1;
const GaussianConditional gc1(
std::map<Key, Matrix>{
{x3, (Matrix21() << 2., 0.).finished()},
{x4, (Matrix21() << 2., 2.).finished()},
},
2, Vector2(2., 2.));
const GaussianConditional gc2(
Pairs{
{x2, (Matrix21() << -2. * sqrt(2.), 0.).finished()},
{x1, (Matrix21() << -sqrt(2.), -sqrt(2.)).finished()},
{x3, (Matrix21() << -sqrt(2.), sqrt(2.)).finished()},
},
2, (Vector(2) << -2. * sqrt(2.), 0.).finished());
GaussianBayesTree bayesTree_expected;
bayesTree_expected.insertRoot(MakeClique(gc1, Children{LeafClique(gc2)}));
EXPECT(assert_equal(bayesTree_expected, bt));
}
/* ************************************************************************* */
TEST( GaussianBayesTree, optimizeMultiFrontal )
{
VectorValues expected = {{x1, (Vector(1) << 0.).finished()},
{x2, (Vector(1) << 1.).finished()},
{x3, (Vector(1) << 0.).finished()},
{x4, (Vector(1) << 1.).finished()}};
VectorValues actual = chain.eliminateMultifrontal(chainOrdering)->optimize();
EXPECT(assert_equal(expected,actual));
}
/* ************************************************************************* */
TEST(GaussianBayesTree, complicatedMarginal) {
// Create the conditionals to go in the BayesTree
const GaussianConditional gc1(
Pairs{{11, (Matrix(3, 1) << 0.0971, 0, 0).finished()},
{12, (Matrix(3, 2) << 0.3171, 0.4387, 0.9502, 0.3816, 0, 0.7655)
.finished()}},
2, Vector3(0.2638, 0.1455, 0.1361));
const GaussianConditional gc2(
Pairs{
{9, (Matrix(3, 1) << 0.7952, 0, 0).finished()},
{10, (Matrix(3, 2) << 0.4456, 0.7547, 0.6463, 0.2760, 0, 0.6797)
.finished()},
{11, (Matrix(3, 1) << 0.6551, 0.1626, 0.1190).finished()},
{12, (Matrix(3, 2) << 0.4984, 0.5853, 0.9597, 0.2238, 0.3404, 0.7513)
.finished()}},
2, Vector3(0.4314, 0.9106, 0.1818));
const GaussianConditional gc3(
Pairs{{7, (Matrix(3, 1) << 0.2551, 0, 0).finished()},
{8, (Matrix(3, 2) << 0.8909, 0.1386, 0.9593, 0.1493, 0, 0.2575)
.finished()},
{11, (Matrix(3, 1) << 0.8407, 0.2543, 0.8143).finished()}},
2, Vector3(0.3998, 0.2599, 0.8001));
const GaussianConditional gc4(
Pairs{{5, (Matrix(3, 1) << 0.2435, 0, 0).finished()},
{6, (Matrix(3, 2) << 0.4733, 0.1966, 0.3517, 0.2511, 0.8308, 0.0)
.finished()},
// NOTE the non-upper-triangular form
// here since this test was written when we had column permutations
// from LDL. The code still works currently (does not enfore
// upper-triangularity in this case) but this test will need to be
// redone if this stops working in the future
{7, (Matrix(3, 1) << 0.5853, 0.5497, 0.9172).finished()},
{8, (Matrix(3, 2) << 0.2858, 0.3804, 0.7572, 0.5678, 0.7537, 0.0759)
.finished()}},
2, Vector3(0.8173, 0.8687, 0.0844));
const GaussianConditional gc5(
Pairs{{3, (Matrix(3, 1) << 0.0540, 0, 0).finished()},
{4, (Matrix(3, 2) << 0.9340, 0.4694, 0.1299, 0.0119, 0, 0.3371)
.finished()},
{6, (Matrix(3, 2) << 0.1622, 0.5285, 0.7943, 0.1656, 0.3112, 0.6020)
.finished()}},
2, Vector3(0.9619, 0.0046, 0.7749));
const GaussianConditional gc6(
Pairs{{1, (Matrix(3, 1) << 0.2630, 0, 0).finished()},
{2, (Matrix(3, 2) << 0.7482, 0.2290, 0.4505, 0.9133, 0, 0.1524)
.finished()},
{5, (Matrix(3, 1) << 0.8258, 0.5383, 0.9961).finished()}},
2, Vector3(0.0782, 0.4427, 0.1067));
// Create the bayes tree:
GaussianBayesTree bt;
bt.insertRoot(MakeClique(
gc1, Children{LeafClique(gc2),
MakeClique(gc3, Children{MakeClique(
gc4, Children{LeafClique(gc5),
LeafClique(gc6)})})}));
// Marginal on 5
Matrix expectedCov = (Matrix(1,1) << 236.5166).finished();
//GaussianConditional actualJacobianChol = *bt.marginalFactor(5, EliminateCholesky);
GaussianConditional actualJacobianQR = *bt.marginalFactor(5, EliminateQR);
//EXPECT(assert_equal(actualJacobianChol, actualJacobianQR)); // Check that Chol and QR obtained marginals are the same
LONGS_EQUAL(1, (long)actualJacobianQR.rows());
LONGS_EQUAL(1, (long)actualJacobianQR.size());
LONGS_EQUAL(5, (long)actualJacobianQR.keys()[0]);
Matrix actualA = actualJacobianQR.getA(actualJacobianQR.begin());
Matrix actualCov = (actualA.transpose() * actualA).inverse();
EXPECT(assert_equal(expectedCov, actualCov, 1e-1));
// Marginal on 6
// expectedCov = (Matrix(2,2) <<
// 8471.2, 2886.2,
// 2886.2, 1015.8);
expectedCov = (Matrix(2,2) <<
1015.8, 2886.2,
2886.2, 8471.2).finished();
//actualJacobianChol = bt.marginalFactor(6, EliminateCholesky);
actualJacobianQR = *bt.marginalFactor(6, EliminateQR);
//EXPECT(assert_equal(actualJacobianChol, actualJacobianQR)); // Check that Chol and QR obtained marginals are the same
LONGS_EQUAL(2, (long)actualJacobianQR.rows());
LONGS_EQUAL(1, (long)actualJacobianQR.size());
LONGS_EQUAL(6, (long)actualJacobianQR.keys()[0]);
actualA = actualJacobianQR.getA(actualJacobianQR.begin());
actualCov = (actualA.transpose() * actualA).inverse();
EXPECT(assert_equal(expectedCov, actualCov, 1e1));
}
/* ************************************************************************* */
namespace {
double computeError(const GaussianBayesTree& gbt, const Vector10& values) {
pair<Matrix, Vector> Rd = GaussianFactorGraph(gbt).jacobian();
return 0.5 * (Rd.first * values - Rd.second).squaredNorm();
}
}
/* ************************************************************************* */
TEST(GaussianBayesTree, ComputeSteepestDescentPointBT) {
const GaussianConditional gc1(
Pairs{{2, (Matrix(6, 2) << 31.0, 32.0, 0.0, 34.0, 0.0, 0.0, 0.0, 0.0, 0.0,
0.0, 0.0, 0.0)
.finished()},
{3, (Matrix(6, 2) << 35.0, 36.0, 37.0, 38.0, 41.0, 42.0, 0.0, 44.0,
0.0, 0.0, 0.0, 0.0)
.finished()},
{4, (Matrix(6, 2) << 0.0, 0.0, 0.0, 0.0, 45.0, 46.0, 47.0, 48.0,
51.0, 52.0, 0.0, 54.0)
.finished()}},
3, (Vector(6) << 29.0, 30.0, 39.0, 40.0, 49.0, 50.0).finished()),
gc2(Pairs{{0, (Matrix(4, 2) << 3.0, 4.0, 0.0, 6.0, 0.0, 0.0, 0.0, 0.0)
.finished()},
{1, (Matrix(4, 2) << 0.0, 0.0, 0.0, 0.0, 17.0, 18.0, 0.0, 20.0)
.finished()},
{2, (Matrix(4, 2) << 0.0, 0.0, 0.0, 0.0, 21.0, 22.0, 23.0, 24.0)
.finished()},
{3, (Matrix(4, 2) << 7.0, 8.0, 9.0, 10.0, 0.0, 0.0, 0.0, 0.0)
.finished()},
{4, (Matrix(4, 2) << 11.0, 12.0, 13.0, 14.0, 25.0, 26.0, 27.0,
28.0)
.finished()}},
2, (Vector(4) << 1.0, 2.0, 15.0, 16.0).finished());
// Create an arbitrary Bayes Tree
GaussianBayesTree bt;
bt.insertRoot(MakeClique(gc1, Children{LeafClique(gc2)}));
// Compute the Hessian numerically
Matrix hessian = numericalHessian<Vector10>(
std::bind(&computeError, bt, std::placeholders::_1), Vector10::Zero());
// Compute the gradient numerically
Vector gradient = numericalGradient<Vector10>(
std::bind(&computeError, bt, std::placeholders::_1), Vector10::Zero());
// Compute the gradient using dense matrices
Matrix augmentedHessian = GaussianFactorGraph(bt).augmentedHessian();
LONGS_EQUAL(11, (long)augmentedHessian.cols());
Vector denseMatrixGradient = -augmentedHessian.col(10).segment(0,10);
EXPECT(assert_equal(gradient, denseMatrixGradient, 1e-5));
// Compute the steepest descent point
double step = -gradient.squaredNorm() / (gradient.transpose() * hessian * gradient)(0);
Vector expected = gradient * step;
// Known steepest descent point from Bayes' net version
VectorValues expectedFromBN{{0, Vector2(0.000129034, 0.000688183)},
{1, Vector2(0.0109679, 0.0253767)},
{2, Vector2(0.0680441, 0.114496)},
{3, Vector2(0.16125, 0.241294)},
{4, Vector2(0.300134, 0.423233)}};
// Compute the steepest descent point with the dogleg function
VectorValues actual = bt.optimizeGradientSearch();
// Check that points agree
KeyVector keys {0, 1, 2, 3, 4};
EXPECT(assert_equal(expected, actual.vector(keys), 1e-5));
EXPECT(assert_equal(expectedFromBN, actual, 1e-5));
// Check that point causes a decrease in error
double origError = GaussianFactorGraph(bt).error(VectorValues::Zero(actual));
double newError = GaussianFactorGraph(bt).error(actual);
EXPECT(newError < origError);
}
/* ************************************************************************* */
TEST(GaussianBayesTree, determinant_and_smallestEigenvalue) {
// create small factor graph
GaussianFactorGraph fg;
Key x1 = 2, x2 = 0, l1 = 1;
SharedDiagonal unit2 = noiseModel::Unit::Create(2);
fg.emplace_shared<JacobianFactor>(x1, 10 * I_2x2, -1.0 * Vector::Ones(2), unit2);
fg.emplace_shared<JacobianFactor>(x2, 10 * I_2x2, x1, -10 * I_2x2, Vector2(2.0, -1.0), unit2);
fg.emplace_shared<JacobianFactor>(l1, 5 * I_2x2, x1, -5 * I_2x2, Vector2(0.0, 1.0), unit2);
fg.emplace_shared<JacobianFactor>(x2, -5 * I_2x2, l1, 5 * I_2x2, Vector2(-1.0, 1.5), unit2);
// create corresponding Bayes tree:
std::shared_ptr<gtsam::GaussianBayesTree> bt = fg.eliminateMultifrontal();
Matrix H = fg.hessian().first;
// test determinant
// NOTE: the hessian of the factor graph is H = R'R where R is the matrix encoded by the bayes tree,
// for this reason we have to take the sqrt
double expectedDeterminant = sqrt(H.determinant()); // determinant computed from full matrix
double actualDeterminant = bt->determinant();
EXPECT_DOUBLES_EQUAL(expectedDeterminant,actualDeterminant,expectedDeterminant*1e-6);// relative tolerance
}
/* ************************************************************************* */
/// Test to expose bug in GaussianBayesTree::logDeterminant.
TEST(GaussianBayesTree, LogDeterminant) {
using symbol_shorthand::L;
using symbol_shorthand::X;
// Create a factor graph that will result in
// a bayes tree with at least 2 nodes.
GaussianFactorGraph fg;
Key x1 = X(1), x2 = X(2), l1 = L(1);
SharedDiagonal unit2 = noiseModel::Unit::Create(2);
fg.emplace_shared<JacobianFactor>(x1, 10 * I_2x2, -1.0 * Vector2::Ones(), unit2);
fg.emplace_shared<JacobianFactor>(x2, 10 * I_2x2, x1, -10 * I_2x2,
Vector2(2.0, -1.0), unit2);
fg.emplace_shared<JacobianFactor>(l1, 5 * I_2x2, x1, -5 * I_2x2, Vector2(0.0, 1.0), unit2);
fg.emplace_shared<JacobianFactor>(x2, -5 * I_2x2, l1, 5 * I_2x2, Vector2(-1.0, 1.5), unit2);
fg.emplace_shared<JacobianFactor>(x3, 10 * I_2x2, x2, -10 * I_2x2,
Vector2(2.0, -1.0), unit2);
fg.emplace_shared<JacobianFactor>(x3, 10 * I_2x2, -1.0 * Vector2::Ones(), unit2);
// create corresponding Bayes net and Bayes tree:
std::shared_ptr<gtsam::GaussianBayesNet> bn = fg.eliminateSequential();
std::shared_ptr<gtsam::GaussianBayesTree> bt = fg.eliminateMultifrontal();
// Test logDeterminant
EXPECT_DOUBLES_EQUAL(bn->logDeterminant(), bt->logDeterminant(), 1e-9);
}
/* ************************************************************************* */
int main() { TestResult tr; return TestRegistry::runAllTests(tr);}
/* ************************************************************************* */