gtsam/gtsam/linear/tests/testGaussianBayesNet.cpp

399 lines
14 KiB
C++

/* ----------------------------------------------------------------------------
* GTSAM Copyright 2010-2022, 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 testGaussianBayesNet.cpp
* @brief Unit tests for GaussianBayesNet
* @author Frank Dellaert
*/
#include <gtsam/linear/GaussianBayesNet.h>
#include <gtsam/linear/GaussianDensity.h>
#include <gtsam/linear/JacobianFactor.h>
#include <gtsam/linear/GaussianFactorGraph.h>
#include <gtsam/base/Testable.h>
#include <gtsam/base/numericalDerivative.h>
#include <gtsam/inference/Symbol.h>
#include <CppUnitLite/TestHarness.h>
// STL/C++
#include <iostream>
#include <sstream>
using namespace std::placeholders;
using namespace std;
using namespace gtsam;
using symbol_shorthand::X;
static const Key _x_ = 11, _y_ = 22, _z_ = 33;
static GaussianBayesNet smallBayesNet = {
std::make_shared<GaussianConditional>(_x_, Vector1::Constant(9), I_1x1, _y_, I_1x1),
std::make_shared<GaussianConditional>(_y_, Vector1::Constant(5), I_1x1)};
static GaussianBayesNet noisyBayesNet = {
std::make_shared<GaussianConditional>(_x_, Vector1::Constant(9), I_1x1, _y_, I_1x1,
noiseModel::Isotropic::Sigma(1, 2.0)),
std::make_shared<GaussianConditional>(_y_, Vector1::Constant(5), I_1x1,
noiseModel::Isotropic::Sigma(1, 3.0))};
/* ************************************************************************* */
TEST( GaussianBayesNet, Matrix )
{
const auto [R, d] = smallBayesNet.matrix(); // find matrix and RHS
Matrix R1 = (Matrix2() <<
1.0, 1.0,
0.0, 1.0
).finished();
Vector d1 = Vector2(9.0, 5.0);
EXPECT(assert_equal(R,R1));
EXPECT(assert_equal(d,d1));
}
/* ************************************************************************* */
// Check that the evaluate function matches direct calculation with R.
TEST(GaussianBayesNet, Evaluate1) {
// Let's evaluate at the mean
const VectorValues mean = smallBayesNet.optimize();
// We get the matrix, which has noise model applied!
const Matrix R = smallBayesNet.matrix().first;
const Matrix invSigma = R.transpose() * R;
// The Bayes net is a Gaussian density ~ exp (-0.5*(Rx-d)'*(Rx-d))
// which at the mean is 1.0! So, the only thing we need to calculate is
// the normalization constant 1.0/sqrt((2*pi*Sigma).det()).
// The covariance matrix inv(Sigma) = R'*R, so the determinant is
const double constant = sqrt((invSigma / (2 * M_PI)).determinant());
EXPECT_DOUBLES_EQUAL(-log(constant),
smallBayesNet.at(0)->negLogConstant() +
smallBayesNet.at(1)->negLogConstant(),
1e-9);
EXPECT_DOUBLES_EQUAL(-log(constant), smallBayesNet.negLogConstant(), 1e-9);
const double actual = smallBayesNet.evaluate(mean);
EXPECT_DOUBLES_EQUAL(constant, actual, 1e-9);
}
// Check the evaluate with non-unit noise.
TEST(GaussianBayesNet, Evaluate2) {
// See comments in test above.
const VectorValues mean = noisyBayesNet.optimize();
const Matrix R = noisyBayesNet.matrix().first;
const Matrix invSigma = R.transpose() * R;
const double constant = sqrt((invSigma / (2 * M_PI)).determinant());
const double actual = noisyBayesNet.evaluate(mean);
EXPECT_DOUBLES_EQUAL(constant, actual, 1e-9);
}
/* ************************************************************************* */
TEST( GaussianBayesNet, NoisyMatrix )
{
const auto [R, d] = noisyBayesNet.matrix(); // find matrix and RHS
Matrix R1 = (Matrix2() <<
0.5, 0.5,
0.0, 1./3.
).finished();
Vector d1 = Vector2(9./2., 5./3.);
EXPECT(assert_equal(R,R1));
EXPECT(assert_equal(d,d1));
}
/* ************************************************************************* */
TEST(GaussianBayesNet, Optimize) {
const VectorValues expected{{_x_, Vector1::Constant(4)},
{_y_, Vector1::Constant(5)}};
const VectorValues actual = smallBayesNet.optimize();
EXPECT(assert_equal(expected, actual));
}
/* ************************************************************************* */
TEST(GaussianBayesNet, NoisyOptimize) {
const auto [R, d] = noisyBayesNet.matrix(); // find matrix and RHS
const Vector x = R.inverse() * d;
const VectorValues expected{{_x_, x.head(1)}, {_y_, x.tail(1)}};
VectorValues actual = noisyBayesNet.optimize();
EXPECT(assert_equal(expected, actual));
}
/* ************************************************************************* */
TEST( GaussianBayesNet, optimizeIncomplete )
{
static GaussianBayesNet incompleteBayesNet;
incompleteBayesNet.emplace_shared<GaussianConditional>(
_x_, Vector1::Constant(9), I_1x1, _y_, I_1x1);
VectorValues solutionForMissing { {_y_, Vector1::Constant(5)} };
VectorValues actual = incompleteBayesNet.optimize(solutionForMissing);
VectorValues expected{{_x_, Vector1::Constant(4)},
{_y_, Vector1::Constant(5)}};
EXPECT(assert_equal(expected,actual));
}
/* ************************************************************************* */
TEST( GaussianBayesNet, optimize3 )
{
// y = R*x, x=inv(R)*y
// 4 = 1 1 -1
// 5 1 5
// NOTE: we are supplying a new RHS here
VectorValues expected { {_x_, Vector1::Constant(-1)},
{_y_, Vector1::Constant(5)} };
// Test different RHS version
VectorValues gx{{_x_, Vector1::Constant(4)}, {_y_, Vector1::Constant(5)}};
VectorValues actual = smallBayesNet.backSubstitute(gx);
EXPECT(assert_equal(expected, actual));
}
/* ************************************************************************* */
namespace sampling {
static Matrix A1 = (Matrix(2, 2) << 1., 2., 3., 4.).finished();
static const Vector2 mean(20, 40), b(10, 10);
static const double sigma = 0.01;
static const GaussianBayesNet gbn = {
GaussianConditional::sharedMeanAndStddev(X(0), A1, X(1), b, sigma),
GaussianDensity::sharedMeanAndStddev(X(1), mean, sigma)};
} // namespace sampling
/* ************************************************************************* */
TEST(GaussianBayesNet, sample) {
using namespace sampling;
auto actual = gbn.sample();
EXPECT_LONGS_EQUAL(2, actual.size());
EXPECT(assert_equal(mean, actual[X(1)], 50 * sigma));
EXPECT(assert_equal(A1 * mean + b, actual[X(0)], 50 * sigma));
// Use a specific random generator
std::mt19937_64 rng(4242);
auto actual3 = gbn.sample(&rng);
EXPECT_LONGS_EQUAL(2, actual.size());
// regressions
#ifdef __APPLE__
EXPECT(assert_equal(Vector2(20.0129382, 40.0039798), actual[X(1)], 1e-5));
EXPECT(assert_equal(Vector2(110.032083, 230.039811), actual[X(0)], 1e-5));
#elif __linux__
EXPECT(assert_equal(Vector2(20.0070499, 39.9942591), actual[X(1)], 1e-5));
EXPECT(assert_equal(Vector2(109.976501, 229.990945), actual[X(0)], 1e-5));
#elif _WIN32
EXPECT(assert_equal(Vector2(20.0129382, 40.0039798), actual[X(1)], 1e-5));
EXPECT(assert_equal(Vector2(110.032083, 230.039811), actual[X(0)], 1e-5));
#endif
}
/* ************************************************************************* */
// Do Monte Carlo integration of square deviation, should be equal to 9.0.
TEST(GaussianBayesNet, MonteCarloIntegration) {
GaussianBayesNet gbn;
gbn.push_back(noisyBayesNet.at(1));
double sum = 0.0;
constexpr size_t N = 1000;
// loop for N samples:
for (size_t i = 0; i < N; i++) {
const auto X_i = gbn.sample();
sum += pow(X_i[_y_].x() - 5.0, 2.0);
}
// Expected is variance = 3*3
EXPECT_DOUBLES_EQUAL(9.0, sum / N, 0.5); // Pretty high.
}
/* ************************************************************************* */
TEST(GaussianBayesNet, ordering)
{
const Ordering expected{_x_, _y_};
const auto actual = noisyBayesNet.ordering();
EXPECT(assert_equal(expected, actual));
}
/* ************************************************************************* */
TEST( GaussianBayesNet, MatrixStress )
{
GaussianBayesNet bn;
using GC = GaussianConditional;
bn.emplace_shared<GC>(_x_, Vector2(1, 2), 1 * I_2x2, _y_, 2 * I_2x2, _z_, 3 * I_2x2);
bn.emplace_shared<GC>(_y_, Vector2(3, 4), 4 * I_2x2, _z_, 5 * I_2x2);
bn.emplace_shared<GC>(_z_, Vector2(5, 6), 6 * I_2x2);
const VectorValues expected = bn.optimize();
for (const auto& keys :
{KeyVector({_x_, _y_, _z_}), KeyVector({_x_, _z_, _y_}),
KeyVector({_y_, _x_, _z_}), KeyVector({_y_, _z_, _x_}),
KeyVector({_z_, _x_, _y_}), KeyVector({_z_, _y_, _x_})}) {
const Ordering ordering(keys);
const auto [R, d] = bn.matrix(ordering);
EXPECT(assert_equal(expected.vector(ordering), R.inverse() * d));
}
}
/* ************************************************************************* */
TEST( GaussianBayesNet, backSubstituteTranspose )
{
// x=R'*y, expected=inv(R')*x
// 2 = 1 2
// 5 1 1 3
const VectorValues x{{_x_, Vector1::Constant(2)},
{_y_, Vector1::Constant(5)}},
expected{{_x_, Vector1::Constant(2)}, {_y_, Vector1::Constant(3)}};
VectorValues actual = smallBayesNet.backSubstituteTranspose(x);
EXPECT(assert_equal(expected, actual));
const auto ordering = noisyBayesNet.ordering();
const Matrix R = smallBayesNet.matrix(ordering).first;
const Vector expected_vector = R.transpose().inverse() * x.vector(ordering);
EXPECT(assert_equal(expected_vector, actual.vector(ordering)));
}
/* ************************************************************************* */
TEST( GaussianBayesNet, backSubstituteTransposeNoisy )
{
// x=R'*y, expected=inv(R')*x
// 2 = 1 2
// 5 1 1 3
VectorValues x{{_x_, Vector1::Constant(2)}, {_y_, Vector1::Constant(5)}},
expected{{_x_, Vector1::Constant(4)}, {_y_, Vector1::Constant(9)}};
VectorValues actual = noisyBayesNet.backSubstituteTranspose(x);
EXPECT(assert_equal(expected, actual));
const auto ordering = noisyBayesNet.ordering();
const Matrix R = noisyBayesNet.matrix(ordering).first;
const Vector expected_vector = R.transpose().inverse() * x.vector(ordering);
EXPECT(assert_equal(expected_vector, actual.vector(ordering)));
}
/* ************************************************************************* */
// Tests computing Determinant
TEST( GaussianBayesNet, DeterminantTest )
{
GaussianBayesNet cbn;
cbn.emplace_shared<GaussianConditional>(
0, Vector2(3.0, 4.0), (Matrix2() << 1.0, 3.0, 0.0, 4.0).finished(),
1, (Matrix2() << 2.0, 1.0, 2.0, 3.0).finished(), noiseModel::Isotropic::Sigma(2, 2.0));
cbn.emplace_shared<GaussianConditional>(
1, Vector2(5.0, 6.0), (Matrix2() << 1.0, 1.0, 0.0, 3.0).finished(),
2, (Matrix2() << 1.0, 0.0, 5.0, 2.0).finished(), noiseModel::Isotropic::Sigma(2, 2.0));
cbn.emplace_shared<GaussianConditional>(
3, Vector2(7.0, 8.0), (Matrix2() << 1.0, 1.0, 0.0, 5.0).finished(), noiseModel::Isotropic::Sigma(2, 2.0));
double expectedDeterminant = 60.0 / 64.0;
double actualDeterminant = cbn.determinant();
EXPECT_DOUBLES_EQUAL( expectedDeterminant, actualDeterminant, 1e-9);
}
/* ************************************************************************* */
namespace {
double computeError(const GaussianBayesNet& gbn, const Vector10& values)
{
pair<Matrix,Vector> Rd = GaussianFactorGraph(gbn).jacobian();
return 0.5 * (Rd.first * values - Rd.second).squaredNorm();
}
}
/* ************************************************************************* */
TEST(GaussianBayesNet, ComputeSteepestDescentPoint) {
// Create an arbitrary Bayes Net
GaussianBayesNet gbn;
gbn.emplace_shared<GaussianConditional>(
0, Vector2(1.0,2.0), (Matrix2() << 3.0,4.0,0.0,6.0).finished(),
3, (Matrix2() << 7.0,8.0,9.0,10.0).finished(),
4, (Matrix2() << 11.0,12.0,13.0,14.0).finished());
gbn.emplace_shared<GaussianConditional>(
1, Vector2(15.0,16.0), (Matrix2() << 17.0,18.0,0.0,20.0).finished(),
2, (Matrix2() << 21.0,22.0,23.0,24.0).finished(),
4, (Matrix2() << 25.0,26.0,27.0,28.0).finished());
gbn.emplace_shared<GaussianConditional>(
2, Vector2(29.0,30.0), (Matrix2() << 31.0,32.0,0.0,34.0).finished(),
3, (Matrix2() << 35.0,36.0,37.0,38.0).finished());
gbn.emplace_shared<GaussianConditional>(
3, Vector2(39.0,40.0), (Matrix2() << 41.0,42.0,0.0,44.0).finished(),
4, (Matrix2() << 45.0,46.0,47.0,48.0).finished());
gbn.emplace_shared<GaussianConditional>(
4, Vector2(49.0,50.0), (Matrix2() << 51.0,52.0,0.0,54.0).finished());
// Compute the Hessian numerically
Matrix hessian = numericalHessian<Vector10>(
std::bind(&computeError, gbn, std::placeholders::_1), Vector10::Zero());
// Compute the gradient numerically
Vector gradient = numericalGradient<Vector10>(
std::bind(&computeError, gbn, std::placeholders::_1), Vector10::Zero());
// Compute the gradient using dense matrices
Matrix augmentedHessian = GaussianFactorGraph(gbn).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;
// Compute the steepest descent point with the dogleg function
VectorValues actual = gbn.optimizeGradientSearch();
// Check that points agree
KeyVector keys {0, 1, 2, 3, 4};
Vector actualAsVector = actual.vector(keys);
EXPECT(assert_equal(expected, actualAsVector, 1e-5));
// Check that point causes a decrease in error
double origError = GaussianFactorGraph(gbn).error(VectorValues::Zero(actual));
double newError = GaussianFactorGraph(gbn).error(actual);
EXPECT(newError < origError);
}
/* ************************************************************************* */
TEST(GaussianBayesNet, Dot) {
GaussianBayesNet fragment;
DotWriter writer;
writer.variablePositions.emplace(_x_, Vector2(10, 20));
writer.variablePositions.emplace(_y_, Vector2(50, 20));
auto position = writer.variablePos(_x_);
CHECK(position);
EXPECT(assert_equal(Vector2(10, 20), *position, 1e-5));
string actual = noisyBayesNet.dot(DefaultKeyFormatter, writer);
EXPECT(actual ==
"digraph {\n"
" size=\"5,5\";\n"
"\n"
" var11[label=\"11\", pos=\"10,20!\"];\n"
" var22[label=\"22\", pos=\"50,20!\"];\n"
"\n"
" var22->var11\n"
"}");
}
/* ************************************************************************* */
int main() {
TestResult tr;
return TestRegistry::runAllTests(tr);
}
/* ************************************************************************* */