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

- fixed stopping conditions 

- handled degenerate case in mu initialization
- set TLS as default
- added more unit tests
release/4.3a0
lcarlone 2020-12-23 22:08:44 -05:00
parent c57174436f
commit dc5c769e7c
2 changed files with 233 additions and 50 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

128
gtsam/nonlinear/GncOptimizer.h
View File

@ -62,11 +62,12 @@ public:
/// GNC parameters
BaseOptimizerParameters baseOptimizerParams; /*optimization parameters used to solve the weighted least squares problem at each GNC iteration*/
/// any other specific GNC parameters:
RobustLossType lossType = GM; /* default loss*/
RobustLossType lossType = TLS; /* default loss*/
size_t maxIterations = 100; /* maximum number of iterations*/
double barcSq = 1.0; /* a factor is considered an inlier if factor.error() < barcSq. Note that factor.error() whitens by the covariance*/
double muStep = 1.4; /* multiplicative factor to reduce/increase the mu in gnc */
double relativeCostTol = 1e-5; ///< The maximum relative cost change to stop iterating
double relativeCostTol = 1e-5; ///< if relative cost change if below this threshold, stop iterating
double weightsTol = 1e-4; ///< if the weights are within weightsTol from being binary, stop iterating (only for TLS)
VerbosityGNC verbosityGNC = SILENT; /* verbosity level */
std::vector<size_t> knownInliers = std::vector<size_t>(); /* slots in the factor graph corresponding to measurements that we know are inliers */
@ -97,6 +98,9 @@ public:
/// Set the maximum relative difference in mu values to stop iterating
void setRelativeCostTol(double value) { relativeCostTol = value;
}
/// Set the maximum difference between the weights and their rounding in {0,1} to stop iterating
void setWeightsTol(double value) { weightsTol = value;
}
/// Set the verbosity level
void setVerbosityGNC(const VerbosityGNC verbosity) {
verbosityGNC = verbosity;
@ -136,6 +140,8 @@ public:
std::cout << "maxIterations: " << maxIterations << "\n";
std::cout << "barcSq: " << barcSq << "\n";
std::cout << "muStep: " << muStep << "\n";
std::cout << "relativeCostTol: " << relativeCostTol << "\n";
std::cout << "weightsTol: " << weightsTol << "\n";
std::cout << "verbosityGNC: " << verbosityGNC << "\n";
for (size_t i = 0; i < knownInliers.size(); i++)
std::cout << "knownInliers: " << knownInliers[i] << "\n";
@ -205,8 +211,8 @@ public:
BaseOptimizer baseOptimizer(nfg_, state_);
Values result = baseOptimizer.optimize();
double mu = initializeMu();
double cost = calculateWeightedCost();
double prev_cost = cost;
double prev_cost = nfg_.error(result);
double cost = 0.0; // this will be updated in the main loop
// handle the degenerate case that corresponds to small
// maximum residual errors at initialization
@ -215,16 +221,21 @@ public:
if (mu <= 0) {
if (params_.verbosityGNC >= GncParameters::VerbosityGNC::SUMMARY) {
std::cout << "GNC Optimizer stopped because maximum residual at "
"initialization is small." << std::endl;
"initialization is small." << std::endl;
}
if (params_.verbosityGNC >= GncParameters::VerbosityGNC::VALUES) {
result.print("result\n");
std::cout << "mu: " << mu << std::endl;
}
return result;
}
for (size_t iter = 0; iter < params_.maxIterations; iter++) {
size_t iter;
for (iter = 0; iter < params_.maxIterations; iter++) {
// display info
if (params_.verbosityGNC >= GncParameters::VerbosityGNC::VALUES) {
std::cout << "iter: " << iter << std::endl;
result.print("result\n");
std::cout << "mu: " << mu << std::endl;
std::cout << "weights: " << weights_ << std::endl;
@ -232,28 +243,34 @@ public:
// weights update
weights_ = calculateWeights(result, mu);
// update cost
prev_cost = cost;
cost = calculateWeightedCost();
// variable/values update
NonlinearFactorGraph graph_iter = this->makeWeightedGraph(weights_);
BaseOptimizer baseOptimizer_iter(graph_iter, state_);
result = baseOptimizer_iter.optimize();
// stopping condition
if (checkConvergence(mu, cost, prev_cost)) {
// display info
if (params_.verbosityGNC >= GncParameters::VerbosityGNC::SUMMARY) {
std::cout << "final iterations: " << iter << std::endl;
std::cout << "final mu: " << mu << std::endl;
std::cout << "final weights: " << weights_ << std::endl;
}
break;
}
cost = graph_iter.error(result);
if (checkConvergence(mu, weights_, cost, prev_cost)) { break; }
// update mu
mu = updateMu(mu);
// get ready for next iteration
prev_cost = cost;
// display info
if (params_.verbosityGNC >= GncParameters::VerbosityGNC::VALUES) {
std::cout << "previous cost: " << prev_cost << std::endl;
std::cout << "current cost: " << cost << std::endl;
}
}
// display info
if (params_.verbosityGNC >= GncParameters::VerbosityGNC::SUMMARY) {
std::cout << "final iterations: " << iter << std::endl;
std::cout << "final mu: " << mu << std::endl;
std::cout << "final weights: " << weights_ << std::endl;
std::cout << "previous cost: " << prev_cost << std::endl;
std::cout << "current cost: " << cost << std::endl;
}
return result;
}
@ -276,7 +293,9 @@ public:
// initialize mu to the value specified in Remark 5 in GNC paper.
// surrogate cost is convex for mu close to zero
// degenerate case: 2 * rmax_sq - params_.barcSq < 0 (handled in the main loop)
return params_.barcSq / (2 * rmax_sq - params_.barcSq);
// according to remark mu = params_.barcSq / (2 * rmax_sq - params_.barcSq) = params_.barcSq/ excessResidual
// however, if the denominator is 0 or negative, we return mu = -1 which leads to termination of the main GNC loop
return (2 * rmax_sq - params_.barcSq) > 0 ? params_.barcSq / (2 * rmax_sq - params_.barcSq) : -1;
default:
throw std::runtime_error(
"GncOptimizer::initializeMu: called with unknown loss type.");
@ -298,53 +317,68 @@ public:
}
}
/// calculated sum of weighted squared residuals
double calculateWeightedCost() const {
double cost = 0;
for (size_t i = 0; i < nfg_.size(); i++) {
cost += weights_[i] * nfg_[i]->error(state_);
}
return cost;
}
/// check if we have reached the value of mu for which the surrogate loss matches the original loss
bool checkMuConvergence(const double mu) const {
bool muConverged = false;
switch (params_.lossType) {
case GncParameters::GM:
return std::fabs(mu - 1.0) < 1e-9; // mu=1 recovers the original GM function
muConverged = std::fabs(mu - 1.0) < 1e-9; // mu=1 recovers the original GM function
break;
case GncParameters::TLS:
muConverged = false; // for TLS there is no stopping condition on mu (it must tend to infinity)
break;
default:
throw std::runtime_error(
"GncOptimizer::checkMuConvergence: called with unknown loss type.");
}
if (muConverged && params_.verbosityGNC >= GncParameters::VerbosityGNC::SUMMARY)
std::cout << "muConverged = true " << std::endl;
return muConverged;
}
/// check convergence of relative cost differences
bool checkCostConvergence(const double cost, const double prev_cost) const {
switch (params_.lossType) {
case GncParameters::TLS:
return std::fabs(cost - prev_cost) < params_.relativeCostTol;
default:
throw std::runtime_error(
"GncOptimizer::checkMuConvergence: called with unknown loss type.");
}
bool costConverged = std::fabs(cost - prev_cost) / std::max(prev_cost,1e-7) < params_.relativeCostTol;
if (costConverged && params_.verbosityGNC >= GncParameters::VerbosityGNC::SUMMARY)
std::cout << "checkCostConvergence = true " << std::endl;
return costConverged;
}
/// check convergence of weights to binary values
bool checkWeightsConvergence(const Vector& weights) const {
bool weightsConverged = false;
switch (params_.lossType) {
case GncParameters::GM:
weightsConverged = false; // for GM, there is no clear binary convergence for the weights
break;
case GncParameters::TLS:
weightsConverged = true;
for(size_t i=0; i<weights.size(); i++){
if( std::fabs ( weights[i] - std::round(weights[i]) ) > params_.weightsTol ){
weightsConverged = false;
break;
}
}
break;
default:
throw std::runtime_error(
"GncOptimizer::checkWeightsConvergence: called with unknown loss type.");
}
if (weightsConverged && params_.verbosityGNC >= GncParameters::VerbosityGNC::SUMMARY)
std::cout << "weightsConverged = true " << std::endl;
return weightsConverged;
}
/// check for convergence between consecutive GNC iterations
bool checkConvergence(const double mu,
const Vector& weights,
const double cost,
const double prev_cost) const {
switch (params_.lossType) {
case GncParameters::GM:
return checkMuConvergence(mu);
case GncParameters::TLS:
return checkCostConvergence(cost, prev_cost);
default:
throw std::runtime_error(
"GncOptimizer::checkMuConvergence: called with unknown loss type.");
}
return checkCostConvergence(cost,prev_cost) ||
checkWeightsConvergence(weights) ||
checkMuConvergence(mu);
}
/// create a graph where each factor is weighted by the gnc weights
NonlinearFactorGraph makeWeightedGraph(const Vector& weights) const {
// make sure all noiseModels are Gaussian or convert to Gaussian

155
tests/testGncOptimizer.cpp
View File

@ -66,7 +66,7 @@ TEST(GncOptimizer, gncParamsConstructor) {
// change something at the gncParams level
GncParams<GaussNewtonParams> gncParams2c(gncParams2b);
gncParams2c.setLossType(GncParams<GaussNewtonParams>::RobustLossType::TLS);
gncParams2c.setLossType(GncParams<GaussNewtonParams>::RobustLossType::GM);
CHECK(!gncParams2c.equals(gncParams2b.baseOptimizerParams));
}
@ -186,7 +186,7 @@ TEST(GncOptimizer, updateMuTLS) {
}
/* ************************************************************************* */
TEST(GncOptimizer, checkMuConvergenceGM) {
TEST(GncOptimizer, checkMuConvergence) {
// has to have Gaussian noise models !
auto fg = example::createReallyNonlinearFactorGraph();
@ -194,6 +194,7 @@ TEST(GncOptimizer, checkMuConvergenceGM) {
Values initial;
initial.insert(X(1), p0);
{
LevenbergMarquardtParams lmParams;
GncParams<LevenbergMarquardtParams> gncParams(lmParams);
gncParams.setLossType(
@ -203,6 +204,112 @@ TEST(GncOptimizer, checkMuConvergenceGM) {
double mu = 1.0;
CHECK(gnc.checkMuConvergence(mu));
}
{
LevenbergMarquardtParams lmParams;
GncParams<LevenbergMarquardtParams> gncParams(lmParams);
gncParams.setLossType(
GncParams<LevenbergMarquardtParams>::RobustLossType::TLS);
auto gnc =
GncOptimizer<GncParams<LevenbergMarquardtParams>>(fg, initial, gncParams);
double mu = 1.0;
CHECK(!gnc.checkMuConvergence(mu)); //always false for TLS
}
}
/* ************************************************************************* */
TEST(GncOptimizer, checkCostConvergence) {
// has to have Gaussian noise models !
auto fg = example::createReallyNonlinearFactorGraph();
Point2 p0(3, 3);
Values initial;
initial.insert(X(1), p0);
{
LevenbergMarquardtParams lmParams;
GncParams<LevenbergMarquardtParams> gncParams(lmParams);
gncParams.setRelativeCostTol(0.49);
auto gnc =
GncOptimizer<GncParams<LevenbergMarquardtParams>>(fg, initial, gncParams);
double prev_cost = 1.0;
double cost = 0.5;
// relative cost reduction = 0.5 > 0.49, hence checkCostConvergence = false
CHECK(!gnc.checkCostConvergence(cost, prev_cost));
}
{
LevenbergMarquardtParams lmParams;
GncParams<LevenbergMarquardtParams> gncParams(lmParams);
gncParams.setRelativeCostTol(0.51);
auto gnc =
GncOptimizer<GncParams<LevenbergMarquardtParams>>(fg, initial, gncParams);
double prev_cost = 1.0;
double cost = 0.5;
// relative cost reduction = 0.5 < 0.51, hence checkCostConvergence = true
CHECK(gnc.checkCostConvergence(cost, prev_cost));
}
}
/* ************************************************************************* */
TEST(GncOptimizer, checkWeightsConvergence) {
// has to have Gaussian noise models !
auto fg = example::createReallyNonlinearFactorGraph();
Point2 p0(3, 3);
Values initial;
initial.insert(X(1), p0);
{
LevenbergMarquardtParams lmParams;
GncParams<LevenbergMarquardtParams> gncParams(lmParams);
gncParams.setLossType(
GncParams<LevenbergMarquardtParams>::RobustLossType::GM);
auto gnc =
GncOptimizer<GncParams<LevenbergMarquardtParams>>(fg, initial, gncParams);
Vector weights = Vector::Ones(fg.size());
CHECK(!gnc.checkWeightsConvergence(weights)); //always false for GM
}
{
LevenbergMarquardtParams lmParams;
GncParams<LevenbergMarquardtParams> gncParams(lmParams);
gncParams.setLossType(
GncParams<LevenbergMarquardtParams>::RobustLossType::TLS);
auto gnc =
GncOptimizer<GncParams<LevenbergMarquardtParams>>(fg, initial, gncParams);
Vector weights = Vector::Ones(fg.size());
// weights are binary, so checkWeightsConvergence = true
CHECK(gnc.checkWeightsConvergence(weights));
}
{
LevenbergMarquardtParams lmParams;
GncParams<LevenbergMarquardtParams> gncParams(lmParams);
gncParams.setLossType(
GncParams<LevenbergMarquardtParams>::RobustLossType::TLS);
auto gnc =
GncOptimizer<GncParams<LevenbergMarquardtParams>>(fg, initial, gncParams);
Vector weights = Vector::Ones(fg.size());
weights[0] = 0.9; // more than weightsTol = 1e-4 from 1, hence checkWeightsConvergence = false
CHECK(!gnc.checkWeightsConvergence(weights));
}
{
LevenbergMarquardtParams lmParams;
GncParams<LevenbergMarquardtParams> gncParams(lmParams);
gncParams.setLossType(
GncParams<LevenbergMarquardtParams>::RobustLossType::TLS);
gncParams.setWeightsTol(0.1);
auto gnc =
GncOptimizer<GncParams<LevenbergMarquardtParams>>(fg, initial, gncParams);
Vector weights = Vector::Ones(fg.size());
weights[0] = 0.9; // exactly weightsTol = 0.1 from 1, hence checkWeightsConvergence = true
CHECK(gnc.checkWeightsConvergence(weights));
}
}
/* ************************************************************************* */
@ -455,9 +562,12 @@ TEST(GncOptimizer, optimizeWithKnownInliers) {
knownInliers.push_back(2);
// nonconvexity with known inliers
{
GncParams<GaussNewtonParams> gncParams;
gncParams.setKnownInliers(knownInliers);
// gncParams.setVerbosityGNC(GncParams<GaussNewtonParams>::VerbosityGNC::VALUES);
gncParams.setLossType(
GncParams<GaussNewtonParams>::RobustLossType::GM);
//gncParams.setVerbosityGNC(GncParams<GaussNewtonParams>::VerbosityGNC::SUMMARY);
auto gnc = GncOptimizer<GncParams<GaussNewtonParams>>(fg, initial, gncParams);
Values gnc_result = gnc.optimize();
@ -468,6 +578,45 @@ TEST(GncOptimizer, optimizeWithKnownInliers) {
DOUBLES_EQUAL(1.0, finalWeights[0], tol);
DOUBLES_EQUAL(1.0, finalWeights[1], tol);
DOUBLES_EQUAL(1.0, finalWeights[2], tol);
}
{
GncParams<GaussNewtonParams> gncParams;
gncParams.setKnownInliers(knownInliers);
gncParams.setLossType(
GncParams<GaussNewtonParams>::RobustLossType::TLS);
// gncParams.setVerbosityGNC(GncParams<GaussNewtonParams>::VerbosityGNC::SUMMARY);
auto gnc = GncOptimizer<GncParams<GaussNewtonParams>>(fg, initial, gncParams);
Values gnc_result = gnc.optimize();
CHECK(assert_equal(Point2(0.0, 0.0), gnc_result.at<Point2>(X(1)), 1e-3));
// check weights were actually fixed:
Vector finalWeights = gnc.getWeights();
DOUBLES_EQUAL(1.0, finalWeights[0], tol);
DOUBLES_EQUAL(1.0, finalWeights[1], tol);
DOUBLES_EQUAL(1.0, finalWeights[2], tol);
DOUBLES_EQUAL(0.0, finalWeights[3], tol);
}
{
// if we set the threshold large, they are all inliers
GncParams<GaussNewtonParams> gncParams;
gncParams.setKnownInliers(knownInliers);
gncParams.setLossType(
GncParams<GaussNewtonParams>::RobustLossType::TLS);
//gncParams.setVerbosityGNC(GncParams<GaussNewtonParams>::VerbosityGNC::VALUES);
gncParams.setInlierCostThreshold( 100.0 );
auto gnc = GncOptimizer<GncParams<GaussNewtonParams>>(fg, initial, gncParams);
Values gnc_result = gnc.optimize();
CHECK(assert_equal(Point2(0.25, 0.0), gnc_result.at<Point2>(X(1)), 1e-3));
// check weights were actually fixed:
Vector finalWeights = gnc.getWeights();
DOUBLES_EQUAL(1.0, finalWeights[0], tol);
DOUBLES_EQUAL(1.0, finalWeights[1], tol);
DOUBLES_EQUAL(1.0, finalWeights[2], tol);
DOUBLES_EQUAL(1.0, finalWeights[3], tol);
}
}
/* ************************************************************************* */