Detailed Description

CPU SFM optimizer supporting either a joint solve or Schur elimination of Point3 and Unit3 variables.

MULTIFRONTAL_SOLVER performs Schur elimination in one fused factorization; other solvers receive an explicitly reduced graph containing every other variable type.

Inheritance diagram for gtsam::SfmLevenbergMarquardtOptimizer:

Public Member Functions
	SfmLevenbergMarquardtOptimizer (const NonlinearFactorGraph &graph, const Values &initialValues, const SfmLevenbergMarquardtParams &params=SfmLevenbergMarquardtParams())
const SfmLevenbergMarquardtParams &	sfmParams () const
Public Member Functions inherited from gtsam::LevenbergMarquardtOptimizer
	LevenbergMarquardtOptimizer (const NonlinearFactorGraph &graph, const Values &initialValues, const LevenbergMarquardtParams &params=LevenbergMarquardtParams())
	Standard constructor, requires a nonlinear factor graph, initial variable assignments, and optimization parameters.
	LevenbergMarquardtOptimizer (const NonlinearFactorGraph &graph, const Values &initialValues, const Ordering &ordering, const LevenbergMarquardtParams &params=LevenbergMarquardtParams())
	Standard constructor, requires a nonlinear factor graph, initial variable assignments, and optimization parameters.
	~LevenbergMarquardtOptimizer () override
	Virtual destructor.
double	lambda () const
	Access the current damping value.
int	getInnerIterations () const
	Access the current number of inner iterations.
void	print (const std::string &str="") const
	print
GaussianFactorGraph::shared_ptr	iterate () override
	Perform a single iteration, returning GaussianFactorGraph corresponding to the linearized factor graph.
const LevenbergMarquardtParams &	params () const
	Read-only access the parameters.
void	writeLogFile (double currentError)
virtual GaussianFactorGraph::shared_ptr	linearize () const
	linearize, can be overwritten
virtual GaussianFactorGraph	buildDampedSystem (const GaussianFactorGraph &linear, const VectorValues &sqrtHessianDiagonal) const
	Build the damped linear system for the current lambda.
bool	tryLambda (const GaussianFactorGraph &linear, const VectorValues &sqrtHessianDiagonal)
	Inner loop, changes state, returns true if successful or giving up.
Public Member Functions inherited from gtsam::NonlinearOptimizer
virtual const Values &	optimize ()
	Optimize for the maximum-likelihood estimate, returning a the optimized variable assignments.
const Values &	optimizeSafely ()
	Optimize, but return empty result if any uncaught exception is thrown Intended for MATLAB.
double	error () const
	return error in current optimizer state
size_t	iterations () const
	return number of iterations in current optimizer state
const Values &	values () const
	return values in current optimizer state
const NonlinearFactorGraph &	graph () const
	return the graph with nonlinear factors
virtual	~NonlinearOptimizer ()
	Virtual destructor.

Static Public Member Functions
static Ordering	CreateReducedOrdering (const NonlinearFactorGraph &graph, const Values &initialValues)
	Create a fill-reducing ordering for the reduced system.
static Ordering	CreateSchurOrdering (const NonlinearFactorGraph &graph, const Ordering &reducedOrdering)
	Create a Schur ordering by prefixing a reduced-system ordering with all other active graph keys in natural key order.

Protected Member Functions
VectorValues	solve (const GaussianFactorGraph &graph, const NonlinearOptimizerParams &params) const override
	Default function to do linear solve, i.e.
bool	ensureMultifrontalSolver (const NonlinearOptimizerParams &params, const Values &values) const override
	Ensure that the nonlinearMultifrontalSolver_ is populated if (and only if) the params request the multifrontal Cholesky solver type (e.g., MULTIFRONTAL_SOLVER).
Protected Member Functions inherited from gtsam::LevenbergMarquardtOptimizer
void	initTime ()
virtual double	linearDeltaError (const GaussianFactorGraph &linear, const VectorValues &delta, double oldError, double newError) const
	Evaluate the linear-model error change used for LM step quality.
const NonlinearOptimizerParams &	_params () const override
	Access the parameters (base class version).
Protected Member Functions inherited from gtsam::NonlinearOptimizer
void	defaultOptimize ()
	A default implementation of the optimization loop, which calls iterate() until checkConvergence returns true.
	NonlinearOptimizer (const NonlinearFactorGraph &graph, std::unique_ptr< internal::NonlinearOptimizerState > state)
	Constructor for initial construction of base classes.

Additional Inherited Members
Public Types inherited from gtsam::LevenbergMarquardtOptimizer
typedef std::shared_ptr< LevenbergMarquardtOptimizer >	shared_ptr
Public Types inherited from gtsam::NonlinearOptimizer
using	shared_ptr = std::shared_ptr<const NonlinearOptimizer>
	A shared pointer to this class.
Protected Attributes inherited from gtsam::LevenbergMarquardtOptimizer
const LevenbergMarquardtParams	params_
	LM parameters.
std::chrono::time_point< std::chrono::high_resolution_clock >	startTime_
	time when optimization started
std::chrono::time_point< std::chrono::high_resolution_clock >	iterationStart_
double	lastLinearizeTime_ {0.0}
double	lastSolveTime_ {0.0}
Protected Attributes inherited from gtsam::NonlinearOptimizer
std::shared_ptr< const NonlinearFactorGraph >	graph_
	The graph with nonlinear factors.
std::unique_ptr< internal::NonlinearOptimizerState >	state_
	PIMPL'd state.
std::unique_ptr< NonlinearMultifrontalSolver >	nonlinearMultifrontalSolver_
	Solver for multifrontal Cholesky, lazily created.
std::unique_ptr< internal::CholmodSolver >	cholmodSolver_
	Optional reusable CHOLMOD numerical and symbolic session.

Member Function Documentation

◆ CreateReducedOrdering()

Ordering gtsam::SfmLevenbergMarquardtOptimizer::CreateReducedOrdering	(	const NonlinearFactorGraph &	graph,
		const Values &	initialValues )

static

Create a fill-reducing ordering for the reduced system.

Active Point3 and Unit3 variables are symbolically eliminated in natural key order before METIS orders every remaining variable. This is the ordering accepted by Schur-mode parameters.

◆ CreateSchurOrdering()

Ordering gtsam::SfmLevenbergMarquardtOptimizer::CreateSchurOrdering	(	const NonlinearFactorGraph &	graph,
		const Ordering &	reducedOrdering )

static

Create a Schur ordering by prefixing a reduced-system ordering with all other active graph keys in natural key order.

This complete ordering is used by Full mode and by the fused Schur multifrontal path.

◆ ensureMultifrontalSolver()

bool gtsam::SfmLevenbergMarquardtOptimizer::ensureMultifrontalSolver	(	const NonlinearOptimizerParams &	params,
		const Values &	values ) const

overrideprotectedvirtual

Ensure that the nonlinearMultifrontalSolver_ is populated if (and only if) the params request the multifrontal Cholesky solver type (e.g., MULTIFRONTAL_SOLVER).

Returns true if the multifrontal solver is available and ready. If a different solver type is requested in params, this function returns false without modifying any solver state. If constraints are present or multifrontal initialization fails, it also returns false (and callers should fall back to the legacy linear solver path).

Reimplemented from gtsam::NonlinearOptimizer.