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

Merge pull request #116 from borglab/fix/huber-noise-model 

Fix Huber mEstimator
release/4.3a0
Fan Jiang 2019-09-19 21:28:31 +08:00 committed by GitHub
parent a614e5efa1 813f169d4a
commit 19315cc3f3
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 11 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
gtsam/linear/NoiseModel.h
View File

@ -629,7 +629,7 @@ namespace gtsam {
/** /**
* The mEstimator name space contains all robust error functions. * The mEstimator name space contains all robust error functions.
* It mirrors the exposition at * It mirrors the exposition at
* http://research.microsoft.com/en-us/um/people/zhang/INRIA/Publis/Tutorial-Estim/node24.html * https://members.loria.fr/MOBerger/Enseignement/Master2/Documents/ZhangIVC-97-01.pdf
* which talks about minimizing \sum \rho(r_i), where \rho is a residual function of choice. * which talks about minimizing \sum \rho(r_i), where \rho is a residual function of choice.
* *
* To illustrate, let's consider the least-squares (L2), L1, and Huber estimators as examples: * To illustrate, let's consider the least-squares (L2), L1, and Huber estimators as examples:
@ -681,7 +681,7 @@ namespace gtsam {
/* /*
* This method is responsible for returning the weight function for a given amount of error. * This method is responsible for returning the weight function for a given amount of error.
* The weight function is related to the analytic derivative of the residual function. See * The weight function is related to the analytic derivative of the residual function. See
* http://research.microsoft.com/en-us/um/people/zhang/INRIA/Publis/Tutorial-Estim/node24.html * https://members.loria.fr/MOBerger/Enseignement/Master2/Documents/ZhangIVC-97-01.pdf
* for details. This method is required when optimizing cost functions with robust penalties * for details. This method is required when optimizing cost functions with robust penalties
* using iteratively re-weighted least squares. * using iteratively re-weighted least squares.
*/ */
@ -776,7 +776,8 @@ namespace gtsam {
Huber(double k = 1.345, const ReweightScheme reweight = Block); Huber(double k = 1.345, const ReweightScheme reweight = Block);
double weight(double error) const { double weight(double error) const {
return (error < k_) ? (1.0) : (k_ / fabs(error)); double absError = std::abs(error);
return (absError < k_) ? (1.0) : (k_ / absError);
} }
void print(const std::string &s) const; void print(const std::string &s) const;
bool equals(const Base& expected, double tol=1e-8) const; bool equals(const Base& expected, double tol=1e-8) const;

13
gtsam/linear/tests/testNoiseModel.cpp
View File

@ -452,19 +452,20 @@ TEST(NoiseModel, WhitenInPlace)
/* /*
* These tests are responsible for testing the weight functions for the m-estimators in GTSAM. * These tests are responsible for testing the weight functions for the m-estimators in GTSAM.
* The weight function is related to the analytic derivative of the residual function. See * The weight function is related to the analytic derivative of the residual function. See
* http://research.microsoft.com/en-us/um/people/zhang/INRIA/Publis/Tutorial-Estim/node24.html * https://members.loria.fr/MOBerger/Enseignement/Master2/Documents/ZhangIVC-97-01.pdf
* for details. This weight function is required when optimizing cost functions with robust * for details. This weight function is required when optimizing cost functions with robust
* penalties using iteratively re-weighted least squares. * penalties using iteratively re-weighted least squares.
*/ */
TEST(NoiseModel, robustFunctionHuber) TEST(NoiseModel, robustFunctionHuber)
{ {
const double k = 5.0, error1 = 1.0, error2 = 10.0; const double k = 5.0, error1 = 1.0, error2 = 10.0, error3 = -10.0, error4 = -1.0;
const mEstimator::Huber::shared_ptr huber = mEstimator::Huber::Create(k); const mEstimator::Huber::shared_ptr huber = mEstimator::Huber::Create(k);
const double weight1 = huber->weight(error1), DOUBLES_EQUAL(1.0, huber->weight(error1), 1e-8);
weight2 = huber->weight(error2); DOUBLES_EQUAL(0.5, huber->weight(error2), 1e-8);
DOUBLES_EQUAL(1.0, weight1, 1e-8); // Test negative value to ensure we take absolute value of error.
DOUBLES_EQUAL(0.5, weight2, 1e-8); DOUBLES_EQUAL(0.5, huber->weight(error3), 1e-8);
DOUBLES_EQUAL(1.0, huber->weight(error4), 1e-8);
} }
TEST(NoiseModel, robustFunctionGemanMcClure) TEST(NoiseModel, robustFunctionGemanMcClure)