update up TwoStateModel test and remove DifferentMeans and DifferentCovariances for later
Varun Agrawal
parent
bb77b0cabb
commit
28f30a232d
|
|
@ -428,7 +428,7 @@ TEST(GaussianMixtureFactor, TwoStateModel) {
|
||||||
|
|
||||||
VectorValues given;
|
VectorValues given;
|
||||||
given.insert(z0, Vector1(0.5));
|
given.insert(z0, Vector1(0.5));
|
||||||
// The motion model says we didn't move
|
// The motion model measurement says we didn't move
|
||||||
given.insert(f01, Vector1(0.0));
|
given.insert(f01, Vector1(0.0));
|
||||||
|
|
||||||
{
|
{
|
||||||
|
|
@ -467,258 +467,56 @@ TEST(GaussianMixtureFactor, TwoStateModel) {
|
||||||
* P(x0)ϕ(x0)ϕ(x1,x0,m1)ϕ(x1)P(m1)
|
* P(x0)ϕ(x0)ϕ(x1,x0,m1)ϕ(x1)P(m1)
|
||||||
*
|
*
|
||||||
* If we only have a measurement on z0, then
|
* If we only have a measurement on z0, then
|
||||||
* the probability of x1 should be the ratio of covariances.
|
* the P(m1) should be 0.5/0.5.
|
||||||
* Getting a measurement on z1 gives use more information.
|
* Getting a measurement on z1 gives use more information.
|
||||||
*/
|
*/
|
||||||
TEST(GaussianMixtureFactor, TwoStateModel2) {
|
TEST(GaussianMixtureFactor, TwoStateModel2) {
|
||||||
|
using namespace test_two_state_estimation;
|
||||||
|
|
||||||
double mu0 = 1.0, mu1 = 3.0;
|
double mu0 = 1.0, mu1 = 3.0;
|
||||||
double sigma0 = 6.0, sigma1 = 4.0;
|
double sigma0 = 6.0, sigma1 = 4.0;
|
||||||
auto model0 = noiseModel::Isotropic::Sigma(1, sigma0);
|
auto model0 = noiseModel::Isotropic::Sigma(1, sigma0);
|
||||||
auto model1 = noiseModel::Isotropic::Sigma(1, sigma1);
|
auto model1 = noiseModel::Isotropic::Sigma(1, sigma1);
|
||||||
|
|
||||||
DiscreteKey m1(M(1), 2);
|
DiscreteKey m1(M(1), 2);
|
||||||
Key z0 = Z(0), z1 = Z(1), x0 = X(0), x1 = X(1);
|
Key z0 = Z(0), z1 = Z(1), f01 = F(0);
|
||||||
|
|
||||||
auto c0 = make_shared<GaussianConditional>(x1, Vector1(mu0), I_1x1, model0),
|
// Start with no measurement on x1, only on x0
|
||||||
c1 = make_shared<GaussianConditional>(x1, Vector1(mu1), I_1x1, model1);
|
HybridBayesNet hbn = CreateBayesNet(mu0, mu1, sigma0, sigma1, false);
|
||||||
|
|
||||||
auto p_x0 = new GaussianConditional(x0, Vector1(0.0), I_1x1,
|
|
||||||
noiseModel::Isotropic::Sigma(1, 1.0));
|
|
||||||
auto p_z0x0 = new GaussianConditional(z0, Vector1(0.0), I_1x1, x0, -I_1x1,
|
|
||||||
noiseModel::Isotropic::Sigma(1, 1.0));
|
|
||||||
auto p_x1m1 = new GaussianMixture({x1}, {}, {m1}, {c0, c1});
|
|
||||||
auto p_z1x1 = new GaussianConditional(z1, Vector1(0.0), I_1x1, x1, -I_1x1,
|
|
||||||
noiseModel::Isotropic::Sigma(1, 3.0));
|
|
||||||
auto p_m1 = new DiscreteConditional(m1, "0.5/0.5");
|
|
||||||
|
|
||||||
HybridBayesNet hbn;
|
|
||||||
hbn.emplace_back(p_x0);
|
|
||||||
hbn.emplace_back(p_z0x0);
|
|
||||||
hbn.emplace_back(p_x1m1);
|
|
||||||
hbn.emplace_back(p_m1);
|
|
||||||
|
|
||||||
VectorValues given;
|
VectorValues given;
|
||||||
given.insert(z0, Vector1(0.5));
|
given.insert(z0, Vector1(0.5));
|
||||||
|
// The motion model measurement says we didn't move
|
||||||
|
given.insert(f01, Vector1(0.0));
|
||||||
|
|
||||||
{
|
{
|
||||||
// Start with no measurement on x1, only on x0
|
// Start with no measurement on x1, only on x0
|
||||||
HybridGaussianFactorGraph gfg = hbn.toFactorGraph(given);
|
HybridGaussianFactorGraph gfg = hbn.toFactorGraph(given);
|
||||||
HybridBayesNet::shared_ptr bn = gfg.eliminateSequential();
|
HybridBayesNet::shared_ptr bn = gfg.eliminateSequential();
|
||||||
|
|
||||||
// Since no measurement on x1, we get the ratio of covariances.
|
// Since no measurement on x1, we a 50/50 probability
|
||||||
DiscreteConditional expected(m1, "0.6/0.4");
|
auto p_m = bn->at(2)->asDiscrete();
|
||||||
|
EXPECT_DOUBLES_EQUAL(0.5, p_m->operator()(DiscreteValues{{m1.first, 0}}),
|
||||||
EXPECT(assert_equal(expected, *(bn->at(2)->asDiscrete())));
|
1e-9);
|
||||||
|
EXPECT_DOUBLES_EQUAL(0.5, p_m->operator()(DiscreteValues{{m1.first, 1}}),
|
||||||
|
1e-9);
|
||||||
}
|
}
|
||||||
|
|
||||||
{
|
{
|
||||||
// Now we add a measurement z1 on x1
|
// Now we add a measurement z1 on x1
|
||||||
hbn.emplace_back(p_z1x1);
|
hbn = CreateBayesNet(mu0, mu1, sigma0, sigma1, true);
|
||||||
|
|
||||||
given.insert(z1, Vector1(2.2));
|
given.insert(z1, Vector1(2.2));
|
||||||
HybridGaussianFactorGraph gfg = hbn.toFactorGraph(given);
|
HybridGaussianFactorGraph gfg = hbn.toFactorGraph(given);
|
||||||
HybridBayesNet::shared_ptr bn = gfg.eliminateSequential();
|
HybridBayesNet::shared_ptr bn = gfg.eliminateSequential();
|
||||||
|
|
||||||
// Since we have a measurement on z2, we get a definite result
|
// Since we have a measurement on z2, we get a definite result
|
||||||
DiscreteConditional expected(m1, "0.52706646/0.47293354");
|
DiscreteConditional expected(m1, "0.4262682/0.5737318");
|
||||||
|
|
||||||
EXPECT(assert_equal(expected, *(bn->at(2)->asDiscrete()), 1e-6));
|
EXPECT(assert_equal(expected, *(bn->at(2)->asDiscrete()), 1e-6));
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
/**
|
|
||||||
* @brief Helper function to specify a Hybrid Bayes Net
|
|
||||||
* {P(X1) P(Z1 | X1, X2, M1)} and convert it to a Hybrid Factor Graph
|
|
||||||
* {P(X1)L(X1, X2, M1; Z1)} by converting to likelihoods given Z1.
|
|
||||||
*
|
|
||||||
* We can specify either different means or different sigmas,
|
|
||||||
* or both for each hybrid factor component.
|
|
||||||
*
|
|
||||||
* @param values Initial values for linearization.
|
|
||||||
* @param means The mean values for the conditional components.
|
|
||||||
* @param sigmas Noise model sigma values (standard deviation).
|
|
||||||
* @param m1 The discrete mode key.
|
|
||||||
* @param z1 The measurement value.
|
|
||||||
* @return HybridGaussianFactorGraph
|
|
||||||
*/
|
|
||||||
HybridGaussianFactorGraph GetFactorGraphFromBayesNet(
|
|
||||||
const gtsam::Values &values, const std::vector<double> &means,
|
|
||||||
const std::vector<double> &sigmas, DiscreteKey &m1, double z1 = 0.0) {
|
|
||||||
// Noise models
|
|
||||||
auto model0 = noiseModel::Isotropic::Sigma(1, sigmas[0]);
|
|
||||||
auto model1 = noiseModel::Isotropic::Sigma(1, sigmas[1]);
|
|
||||||
auto prior_noise = noiseModel::Isotropic::Sigma(1, 1e-3);
|
|
||||||
|
|
||||||
// GaussianMixtureFactor component factors
|
|
||||||
auto f0 =
|
|
||||||
std::make_shared<BetweenFactor<double>>(X(1), X(2), means[0], model0);
|
|
||||||
auto f1 =
|
|
||||||
std::make_shared<BetweenFactor<double>>(X(1), X(2), means[1], model1);
|
|
||||||
std::vector<NonlinearFactor::shared_ptr> factors{f0, f1};
|
|
||||||
|
|
||||||
/// Get terms for each p^m(z1 | x1, x2)
|
|
||||||
Matrix H0_1, H0_2, H1_1, H1_2;
|
|
||||||
double x1 = values.at<double>(X(1)), x2 = values.at<double>(X(2));
|
|
||||||
Vector d0 = f0->evaluateError(x1, x2, &H0_1, &H0_2);
|
|
||||||
std::vector<std::pair<Key, Matrix>> terms0 = {{Z(1), gtsam::I_1x1 /*Rx*/},
|
|
||||||
//
|
|
||||||
{X(1), H0_1 /*Sp1*/},
|
|
||||||
{X(2), H0_2 /*Tp2*/}};
|
|
||||||
|
|
||||||
Vector d1 = f1->evaluateError(x1, x2, &H1_1, &H1_2);
|
|
||||||
std::vector<std::pair<Key, Matrix>> terms1 = {{Z(1), gtsam::I_1x1 /*Rx*/},
|
|
||||||
//
|
|
||||||
{X(1), H1_1 /*Sp1*/},
|
|
||||||
{X(2), H1_2 /*Tp2*/}};
|
|
||||||
// Create conditional P(Z1 | X1, X2, M1)
|
|
||||||
auto gm = new gtsam::GaussianMixture(
|
|
||||||
{Z(1)}, {X(1), X(2)}, {m1},
|
|
||||||
{std::make_shared<GaussianConditional>(terms0, 1, -d0, model0),
|
|
||||||
std::make_shared<GaussianConditional>(terms1, 1, -d1, model1)});
|
|
||||||
gtsam::HybridBayesNet bn;
|
|
||||||
bn.emplace_back(gm);
|
|
||||||
// bn.print();
|
|
||||||
|
|
||||||
// Create FG via toFactorGraph
|
|
||||||
gtsam::VectorValues measurements;
|
|
||||||
measurements.insert(Z(1), gtsam::I_1x1 * z1); // Set Z1 = 0
|
|
||||||
HybridGaussianFactorGraph mixture_fg = bn.toFactorGraph(measurements);
|
|
||||||
|
|
||||||
// Linearized prior factor on X1
|
|
||||||
auto prior = PriorFactor<double>(X(1), x1, prior_noise).linearize(values);
|
|
||||||
mixture_fg.push_back(prior);
|
|
||||||
|
|
||||||
return mixture_fg;
|
|
||||||
}
|
|
||||||
|
|
||||||
/* ************************************************************************* */
|
|
||||||
/**
|
|
||||||
* @brief Test components with differing means.
|
|
||||||
*
|
|
||||||
* We specify a hybrid Bayes network P(Z | X, M) =p(X1)p(Z1 | X1, X2, M1),
|
|
||||||
* which is then converted to a factor graph by specifying Z1.
|
|
||||||
* This is a different case since now we have a hybrid factor
|
|
||||||
* with 2 continuous variables ϕ(x1, x2, m1).
|
|
||||||
* p(Z1 | X1, X2, M1) has 2 factors each for the binary
|
|
||||||
* mode m1, with only the means being different.
|
|
||||||
*/
|
|
||||||
TEST(GaussianMixtureFactor, DifferentMeans) {
|
|
||||||
DiscreteKey m1(M(1), 2);
|
|
||||||
|
|
||||||
Values values;
|
|
||||||
double x1 = 0.0, x2 = 1.75;
|
|
||||||
values.insert(X(1), x1);
|
|
||||||
values.insert(X(2), x2);
|
|
||||||
|
|
||||||
// Different means, same sigma
|
|
||||||
std::vector<double> means{0.0, 2.0}, sigmas{1e-0, 1e-0};
|
|
||||||
|
|
||||||
HybridGaussianFactorGraph hfg =
|
|
||||||
GetFactorGraphFromBayesNet(values, means, sigmas, m1, 0.0);
|
|
||||||
|
|
||||||
{
|
|
||||||
// With no measurement on X2, each mode should be equally likely
|
|
||||||
auto bn = hfg.eliminateSequential();
|
|
||||||
HybridValues actual = bn->optimize();
|
|
||||||
|
|
||||||
HybridValues expected(
|
|
||||||
VectorValues{{X(1), Vector1(0.0)}, {X(2), Vector1(-1.75)}},
|
|
||||||
DiscreteValues{{M(1), 0}});
|
|
||||||
|
|
||||||
EXPECT(assert_equal(expected, actual));
|
|
||||||
|
|
||||||
{
|
|
||||||
DiscreteValues dv{{M(1), 0}};
|
|
||||||
VectorValues cont = bn->optimize(dv);
|
|
||||||
double error = bn->error(HybridValues(cont, dv));
|
|
||||||
// regression
|
|
||||||
EXPECT_DOUBLES_EQUAL(0.69314718056, error, 1e-9);
|
|
||||||
}
|
|
||||||
{
|
|
||||||
DiscreteValues dv{{M(1), 1}};
|
|
||||||
VectorValues cont = bn->optimize(dv);
|
|
||||||
double error = bn->error(HybridValues(cont, dv));
|
|
||||||
// regression
|
|
||||||
EXPECT_DOUBLES_EQUAL(0.69314718056, error, 1e-9);
|
|
||||||
}
|
|
||||||
}
|
|
||||||
{
|
|
||||||
// If we add a measurement on X2, we have more information to work with.
|
|
||||||
// Add a measurement on X2
|
|
||||||
auto prior_noise = noiseModel::Isotropic::Sigma(1, 1e-3);
|
|
||||||
GaussianConditional meas_z2(Z(2), Vector1(2.0), I_1x1, X(2), I_1x1,
|
|
||||||
prior_noise);
|
|
||||||
auto prior_x2 = meas_z2.likelihood(Vector1(x2));
|
|
||||||
|
|
||||||
hfg.push_back(prior_x2);
|
|
||||||
|
|
||||||
auto bn = hfg.eliminateSequential();
|
|
||||||
HybridValues actual = bn->optimize();
|
|
||||||
|
|
||||||
HybridValues expected(
|
|
||||||
VectorValues{{X(1), Vector1(0.0)}, {X(2), Vector1(0.25)}},
|
|
||||||
DiscreteValues{{M(1), 1}});
|
|
||||||
|
|
||||||
EXPECT(assert_equal(expected, actual));
|
|
||||||
|
|
||||||
{
|
|
||||||
DiscreteValues dv{{M(1), 0}};
|
|
||||||
VectorValues cont = bn->optimize(dv);
|
|
||||||
double error = bn->error(HybridValues(cont, dv));
|
|
||||||
// regression
|
|
||||||
EXPECT_DOUBLES_EQUAL(2.12692448787, error, 1e-9);
|
|
||||||
}
|
|
||||||
{
|
|
||||||
DiscreteValues dv{{M(1), 1}};
|
|
||||||
VectorValues cont = bn->optimize(dv);
|
|
||||||
double error = bn->error(HybridValues(cont, dv));
|
|
||||||
// regression
|
|
||||||
EXPECT_DOUBLES_EQUAL(0.126928487854, error, 1e-9);
|
|
||||||
}
|
|
||||||
}
|
|
||||||
}
|
|
||||||
|
|
||||||
/* ************************************************************************* */
|
|
||||||
/**
|
|
||||||
* @brief Test components with differing covariances
|
|
||||||
* but with a Bayes net P(Z|X, M) converted to a FG.
|
|
||||||
* Same as the DifferentMeans example but in this case,
|
|
||||||
* we keep the means the same and vary the covariances.
|
|
||||||
*/
|
|
||||||
TEST(GaussianMixtureFactor, DifferentCovariances) {
|
|
||||||
DiscreteKey m1(M(1), 2);
|
|
||||||
|
|
||||||
Values values;
|
|
||||||
double x1 = 1.0, x2 = 1.0;
|
|
||||||
values.insert(X(1), x1);
|
|
||||||
values.insert(X(2), x2);
|
|
||||||
|
|
||||||
std::vector<double> means{0.0, 0.0}, sigmas{1e2, 1e-2};
|
|
||||||
HybridGaussianFactorGraph mixture_fg =
|
|
||||||
GetFactorGraphFromBayesNet(values, means, sigmas, m1);
|
|
||||||
|
|
||||||
auto hbn = mixture_fg.eliminateSequential();
|
|
||||||
|
|
||||||
VectorValues cv;
|
|
||||||
cv.insert(X(1), Vector1(0.0));
|
|
||||||
cv.insert(X(2), Vector1(0.0));
|
|
||||||
|
|
||||||
// Check that the error values at the MLE point μ.
|
|
||||||
AlgebraicDecisionTree<Key> errorTree = hbn->errorTree(cv);
|
|
||||||
|
|
||||||
DiscreteValues dv0{{M(1), 0}};
|
|
||||||
DiscreteValues dv1{{M(1), 1}};
|
|
||||||
|
|
||||||
// regression
|
|
||||||
EXPECT_DOUBLES_EQUAL(9.90348755254, errorTree(dv0), 1e-9);
|
|
||||||
EXPECT_DOUBLES_EQUAL(0.69314718056, errorTree(dv1), 1e-9);
|
|
||||||
|
|
||||||
DiscreteConditional expected_m1(m1, "0.5/0.5");
|
|
||||||
DiscreteConditional actual_m1 = *(hbn->at(2)->asDiscrete());
|
|
||||||
|
|
||||||
EXPECT(assert_equal(expected_m1, actual_m1));
|
|
||||||
}
|
|
||||||
|
|
||||||
/* ************************************************************************* */
|
/* ************************************************************************* */
|
||||||
int main() {
|
int main() {
|
||||||
TestResult tr;
|
TestResult tr;
|
||||||
|
|
|
||||||
Loading…
Reference in New Issue