Sparse optimizer that manages active vertices and edges for graph-based optimization. More...

Inheritance diagram for SparseOptimizer:

Collaboration diagram for SparseOptimizer:

Public Types
enum
	Action types for optimization callbacks.
Public Types inherited from OptimizableGraph
enum	ActionType { AT_PREITERATION , AT_POSTITERATION , AT_NUM_ELEMENTS }
	Types of actions that can be registered for optimization callbacks. More...
typedef Buffer< OptimizableGraph::OGEdge * >	EdgeContainer
	vector container for edges
typedef Buffer< OptimizableGraph::OGVertex * >	VertexContainer
	vector container for vertices

Public Member Functions
g_type	activeChi2 () const
	Returns the total chi-squared error of all active edges.
const EdgeContainer &	activeEdges () const
	the edges active in the current optimization
g_type	activeRobustChi2 () const
	Returns the total robustified chi-squared error of all active edges.
const VertexContainer &	activeVertices () const
	the vertices active in the current optimization
bool	buildIndexMapping (SparseOptimizer::VertexContainer &vlist)
	Builds the index mapping from active vertices to Hessian positions.
void	clear () final override
	Clears all internal structures and the base graph.
void	clearIndexMapping ()
	Clears the index mapping, resetting all vertex Hessian indices to -1.
void	computeActiveErrors ()
	computes the error vectors of all edges in the activeSet, and caches them
void	discardTop () final override
	discard the last backup of the estimate for all variables by removing it from the stack
OptimizableGraph::OGEdge const	findActiveEdge (const OptimizableGraph::OGEdge *e) const
	Finds an active edge by pointer using binary search.
OGVertex const	findActiveVertex (const OGVertex *v) const
	Finds an active vertex by pointer using binary search.
OGVertex *	findGauge ()
	Finds a gauge vertex (first vertex with maximum dimension) for fixing.
bool *	forceStopFlag () const
	Returns the external force-stop flag pointer.
bool	gaugeFreedom ()
	Returns true if the graph has gauge freedom (no fixed vertex of maximum dimension).
const VertexContainer &	indexMapping () const
	the index mapping of the vertices
bool	initializeOptimization (int level=0)
	Initializes the structures for optimizing a portion of the graph specified by a subset of vertices.
bool	initializeOptimization (Set< HyperGraph::HGVertex * > &eset)
	Initializes optimization from a specific edge set.
bool	isReadyToUpdate () const
	Returns true if the index mapping has been built and the optimizer is ready.
void	linearizeSystem ()
	Linearizes the system by computing the Jacobians for the nodes and edges in the graph.
void	pop () final override
	pop (restore) the estimate of all variables from the stack
void	pop (Set< HyperGraph::HGVertex * > &vlist) final override
	pop (restore) the estimate a subset of the variables from the stack
void	push () final override
	Remove a vertex.
void	push (Set< HyperGraph::HGVertex * > &vlist) final override
	push the estimate of a subset of the variables onto a stack
void	setForceStopFlag (bool *flag)
	sets a variable checked at every iteration to force a user stop.
bool	terminate ()
	if external stop flag is given, return its state. False otherwise
void	update (const g_type *update)
	update the estimate of the active vertices
Public Member Functions inherited from OptimizableGraph
	OptimizableGraph ()
	empty constructor
virtual bool	addEdge (HyperGraph::HGEdge *e_) final override
	adds a new edge.
virtual bool	addVertex (HyperGraph::HGVertex &v) final override
	adds a new vertex.
g_type	chi2 () const
	returns the chi2 of the current configuration
Set< sint04 >	dimensions () const
	iterates over all vertices and returns a set of all the vertex dimensions in the graph
bool	isSolverSuitable (const OptimizationAlgorithmProperty &solverProperty, const Set< sint04 > &vertDims_=Set< int >()) const
	test whether a solver is suitable for optimizing this graph.
JacobianWorkspace &	jacobianWorkspace ()
	verify that all the information of the edges are semi positive definite, i.e., all Eigenvalues are >= 0.
const JacobianWorkspace &	jacobianWorkspace () const
	Returns a const reference to the Jacobian workspace.
sint04	maxDimension () const
	return the maximum dimension of all vertices in the graph
virtual void	setFixed (Set< HyperGraph::HGVertex * > &vset, bool fixed)
	fixes/releases a set of vertices
Public Member Functions inherited from HyperGraph
	HyperGraph ()
	constructs an empty hyper graph
virtual	~HyperGraph ()
	destroys the hyper-graph and all the vertices of the graph
Buffer< HGEdge * > &	edges ()
const Buffer< HGEdge * > &	edges () const
HGVertex *	vertex (int id)
	returns a vertex id in the hyper-graph, or 0 if the vertex id is not present
const HGVertex *	vertex (int id) const
	returns a vertex id in the hyper-graph, or 0 if the vertex id is not present
Dictionary< int, HGVertex * > &	vertices ()
const Dictionary< int, HGVertex * > &	vertices () const

Protected Member Functions
void	sortVectorContainers ()
	Sorts active vertices and edges by their ID comparators for deterministic ordering.

Protected Attributes
EdgeContainer	_activeEdges
	Active edges, sorted by EdgeIDCompare.
VertexContainer	_activeVertices
	Active vertices, sorted by VertexIDCompare.
bool *	_forceStopFlag
	External flag to force stopping the optimization.
VertexContainer	_ivMap
	Index-to-vertex mapping for the Hessian.
Protected Attributes inherited from OptimizableGraph
JacobianWorkspace	_jacobianWorkspace
	Workspace for computing Jacobians.
int	_nextEdgeId = 0
	Counter for assigning internal edge IDs.
Protected Attributes inherited from HyperGraph
Buffer< HGEdge * >	_edges
	Collection of all edges in the graph.
Dictionary< int, HGVertex * >	_vertices
	Map from vertex id to vertex pointer.

Detailed Description

Sparse optimizer that manages active vertices and edges for graph-based optimization.

Definition at line 13 of file sparse_optimizer.h.

Member Function Documentation

◆ buildIndexMapping()

bool SparseOptimizer::buildIndexMapping ( SparseOptimizer::VertexContainer & vlist )

inline

Builds the index mapping from active vertices to Hessian positions.

Parameters

[in] vlist The vertex container to build the mapping from.

Returns: True on success, false if the list is empty.

Definition at line 176 of file sparse_optimizer.h.

References _ivMap, OptimizableGraph::OGVertex::fixed(), OptimizableGraph::OGVertex::marginalized(), and OptimizableGraph::OGVertex::setHessianIndex().

Referenced by initializeOptimization(), and initializeOptimization().

◆ findActiveEdge()

OptimizableGraph::OGEdge *const * SparseOptimizer::findActiveEdge ( const OptimizableGraph::OGEdge * e ) const

inline

Finds an active edge by pointer using binary search.

Parameters

[in] e The edge to find.

Returns: Pointer to the found element, or end() if not found.

Definition at line 336 of file sparse_optimizer.h.

References _activeEdges.

◆ findActiveVertex()

OGVertex *const * SparseOptimizer::findActiveVertex ( const OGVertex * v ) const

inline

Finds an active vertex by pointer using binary search.

Parameters

[in] v The vertex to find.

Returns: Pointer to the found element, or end() if not found.

Definition at line 323 of file sparse_optimizer.h.

References _activeVertices.

◆ initializeOptimization() [1/2]

bool SparseOptimizer::initializeOptimization ( int level = 0 )

inline

Initializes the structures for optimizing a portion of the graph specified by a subset of vertices.

Before calling it be sure to invoke marginalized() and fixed() to the vertices you want to include in the schur complement or to set as fixed during the optimization.

Parameters

vset	the subgraph to be optimized.
level	is the level (in multilevel optimization)

Returns: false if somethings goes wrong

Initializes optimization structures from all graph vertices at the given level.

Parameters

[in] level The multi-level optimization level (-1 for all levels).

Returns: True on success.

Definition at line 221 of file sparse_optimizer.h.

References _activeEdges, _activeVertices, OptimizableGraph::_jacobianWorkspace, OptimizableGraph::OGEdge::allVerticesFixed(), buildIndexMapping(), cast(), clearIndexMapping(), HyperGraph::edges(), HyperGraph::HGVertex::edges(), HyperGraph::HGVertex::id(), Set< t_value >::size(), sortVectorContainers(), HyperGraph::HGEdge::vertex(), HyperGraph::HGEdge::vertexCount(), and HyperGraph::vertices().

◆ initializeOptimization() [2/2]

bool SparseOptimizer::initializeOptimization ( Set< HyperGraph::HGVertex * > & eset )

inline

Initializes optimization from a specific edge set.

Parameters

[in] eset The set of edges to optimize.

Returns: True on success.

Definition at line 280 of file sparse_optimizer.h.

References _activeEdges, _activeVertices, OptimizableGraph::_jacobianWorkspace, buildIndexMapping(), cast(), clearIndexMapping(), Set< t_value >::size(), sortVectorContainers(), HyperGraph::HGEdge::vertex(), and HyperGraph::HGEdge::vertexCount().

◆ push()

void SparseOptimizer::push ( )

inlinefinaloverridevirtual

Remove a vertex.

If the vertex is contained in the currently active set of vertices, then the internal temporary structures are cleaned, e.g., the index mapping is erased. In case you need the index mapping for manipulating the graph, you have to store it in your own copy.

Reimplemented from OptimizableGraph.

Definition at line 413 of file sparse_optimizer.h.

References _activeVertices, and push().

Referenced by push().

◆ setForceStopFlag()

void SparseOptimizer::setForceStopFlag ( bool * flag )

inline

sets a variable checked at every iteration to force a user stop.

The iteration exits when the variable is true;

Definition at line 390 of file sparse_optimizer.h.

References _forceStopFlag.

◆ update()

void SparseOptimizer::update ( const g_type * update )

inline

update the estimate of the active vertices

Parameters

update the g_type vector containing the stacked elements of the increments on the vertices.

Definition at line 61 of file sparse_optimizer.h.

References _ivMap, OptimizableGraph::OGVertex::dimension(), OptimizableGraph::OGVertex::oplus(), and update().

Referenced by OptimizationAlgorithmGaussNewton< t_type >::solve(), OptimizationAlgorithmLevenberg< t_type >::solve(), and update().

The documentation for this class was generated from the following file:

GraphOptimization/Headers/sparse_optimizer.h

Public Types

Public Member Functions

Protected Member Functions

Protected Attributes

Detailed Description

Member Function Documentation

◆ buildIndexMapping()

◆ findActiveEdge()

◆ findActiveVertex()

◆ initializeOptimization() [1/2]

◆ initializeOptimization() [2/2]

◆ push()

◆ setForceStopFlag()

◆ update()