Detailed Description

template<int ErrorDim, int... BlockDims>
class gtsam::BatchJacobianFactor< ErrorDim, BlockDims >

Fixed-dimension row-sparse batch Jacobian factor.

This factor stores many small Jacobian row groups in compact form. A row group corresponds to one original nonlinear factor and has ErrorDim scalar rows. For each row group we store:

the local key slot used by each factor variable,
one fixed-size Jacobian block for each factor variable, and
one fixed-size right-hand side vector.

The factor's key list is the union of all keys appearing in the batch. Unlike a dense JacobianFactor, a row group only stores blocks for the keys touched by that original factor. For example, point-first SFM batches avoid storing zero camera blocks for cameras that did not observe the point.

The template parameters require fixed dimensions so the hot loops can use Eigen fixed-size products. Dynamic-dimension factors should use the ordinary JacobianFactor fallback in BatchFactor.

Noise-model convention: if model_ is null or unit, blocks and right-hand sides are treated as already whitened. If model_ is a non-unit diagonal model, updateHessian() applies the per-row diagonal weights directly. Constrained models are intentionally rejected by updateHessian(), matching JacobianFactor's Hessian assembly behavior.

Inheritance diagram for gtsam::BatchJacobianFactor< ErrorDim, BlockDims >:

Public Member Functions
	BatchJacobianFactor (const KeyVector &keys, std::vector< size_t > keyDims, const SharedDiagonal &model=SharedDiagonal())
	Construct an empty compact batch factor with known key dimensions.
GaussianFactor::shared_ptr	clone () const override
	Return a deep copy as a GaussianFactor.
void	reserve (size_t rowCount)
	Reserve storage for row groups before repeated addRow() calls.
void	addRow (const SlotIndices &slots, const std::vector< Matrix > &blocks, const Vector &rhs)
	Add one row group corresponding to one original nonlinear factor.
void	addUnaryRow (DenseIndex keySlot, const typename std::tuple_element< 0, Blocks >::type::value_type &block, const RhsVector &rhs)
	Add one row to a unary compact batch without dynamic block containers.
size_t	rows () const override
	Return the number of scalar rows represented by all row groups.
const SharedDiagonal &	get_model () const override
	Return the optional diagonal model on the stored rows.
DenseIndex	getDim (const_iterator variable) const override
	Return the dimension of the variable at the given key iterator.
const std::vector< SlotIndices > &	rowSlots () const
	Return the compact key slot used by each row group and factor slot.
template<size_t Slot>
const std::tuple_element< Slot, Blocks >::type::value_type &	block (size_t rowIndex) const
	Return one fixed-size Jacobian block from a compact row group.
const RhsVector &	rowRhs (size_t rowIndex) const
	Return the right-hand side for one compact row group.
double	deltaError (const VectorValues &values, double oldError=nullptr, double newError=nullptr) const override
	Evaluate the linear-model error change directly from compact rows.
void	hessianDiagonalAdd (VectorValues &diagonal) const override
	Accumulate the Hessian scalar diagonal directly from compact rows.
void	multiplyHessianAdd (double alpha, const std::vector< size_t > &scalarOffsets, const double x, double y) const override
	Add alphaA.transpose()WAx directly to a flat output vector.
void	gradientAtZeroAdd (const std::vector< size_t > &scalarOffsets, double *gradient) const override
	Add -A.transpose()Wb directly to a flat gradient vector.
void	hessianBlockDiagonalAdd (const std::vector< size_t > &blockSlots, std::vector< Matrix > *diagonalBlocks) const override
	Add compact A.transpose()WA diagonal blocks without keyed maps.
JacobianFactor	toJacobianFactor () const override
	Convert compact row-block storage into a conventional JacobianFactor.
void	updateHessian (const KeyVector &infoKeys, SymmetricBlockMatrix *info) const override
	Add this factor's augmented information matrix to info.
void	updateHessian (const KeyVector &infoKeys, SymmetricBlockMatrix *info, DenseIndex beginCol, DenseIndex endCol) const override
	Add this factor's augmented information matrix over a block-column range.
Public Member Functions inherited from gtsam::BatchJacobianFactorBase
void	print (const std::string &s="", const KeyFormatter &formatter=DefaultKeyFormatter) const override
	Print this compact factor by converting to a JacobianFactor.
bool	equals (const GaussianFactor &factor, double tol=1e-9) const override
	Test equality by comparing dense JacobianFactor representations.
double	error (const VectorValues &values) const override
	Compute the factor error by delegating to the dense compatibility factor.
double	deltaError (const VectorValues &values, double oldError=nullptr, double newError=nullptr) const override
	Compute the error change by delegating to the dense compatibility factor.
Matrix	augmentedJacobian () const override
	Return the dense augmented Jacobian matrix for compatibility callers.
std::pair< Matrix, Vector >	jacobian () const override
	Return the dense Jacobian matrix and right-hand side vector.
Matrix	augmentedInformation () const override
	Return the dense augmented information matrix for compatibility callers.
Matrix	information () const override
	Return the dense information matrix for compatibility callers.
void	hessianDiagonalAdd (VectorValues &diagonal) const override
	Add this factor's Hessian diagonal into the given VectorValues.
void	hessianDiagonal (double *diagonal) const override
	Add this factor's Hessian diagonal into a raw scalar buffer.
std::map< Key, Matrix >	hessianBlockDiagonal () const override
	Return the Hessian diagonal blocks keyed by variable.
GaussianFactor::shared_ptr	negate () const override
	Return a factor representing the negated linear system.
void	updateHessian (const KeyVector &keys, SymmetricBlockMatrix *info) const override
	Update the augmented Hessian using key lookup through the dense fallback.
void	updateHessian (const KeyVector &keys, SymmetricBlockMatrix *info, DenseIndex beginCol, DenseIndex endCol) const override
	Update a column range of the augmented Hessian using the dense fallback.
void	multiplyHessianAdd (double alpha, const VectorValues &x, VectorValues &y) const override
	Add the Hessian-vector product into y using the dense fallback.
VectorValues	gradientAtZero () const override
	Return the full gradient evaluated at zero.
void	gradientAtZero (double *d) const override
	Add the gradient evaluated at zero into a raw scalar buffer.
Vector	gradient (Key key, const VectorValues &x) const override
	Return the gradient block for one key at the given linearization point.
	GaussianFactor ()
	Inherit GaussianFactor constructors.
template<typename CONTAINER>
	GaussianFactor (const CONTAINER &keys)
	Inherit GaussianFactor constructors.
Public Member Functions inherited from gtsam::GaussianFactor
	GaussianFactor ()
	Default constructor creates empty factor.
template<typename CONTAINER>
	GaussianFactor (const CONTAINER &keys)
	Construct from container of keys.
double	error (const HybridValues &c) const override
	All factor types need to implement an error function.
VectorValues	hessianDiagonal () const
	Return the diagonal of the Hessian for this factor.
virtual bool	isJacobian () const
	Fast check for JacobianFactor-based types.
Public Member Functions inherited from gtsam::Factor
virtual	~Factor ()=default
	Default destructor.
bool	empty () const
	Whether the factor is empty (involves zero variables).
Key	front () const
	First key.
Key	back () const
	Last key.
const_iterator	find (Key key) const
	find
const KeyVector &	keys () const
	Access the factor's involved variable keys.
const_iterator	begin () const
	Iterator at beginning of involved variable keys.
const_iterator	end () const
	Iterator at end of involved variable keys.
size_t	size () const
virtual void	printKeys (const std::string &s="Factor", const KeyFormatter &formatter=DefaultKeyFormatter) const
	print only keys
bool	equals (const This &other, double tol=1e-9) const
	check equality
KeyVector &	keys ()
iterator	begin ()
	Iterator at beginning of involved variable keys.
iterator	end ()
	Iterator at end of involved variable keys.

Static Public Attributes
static constexpr size_t	NumSlots = sizeof...(BlockDims)

Public Types
using	Base = BatchJacobianFactorBase
using	This = BatchJacobianFactor<ErrorDim, BlockDims...>
using	shared_ptr = std::shared_ptr<This>
using	SlotIndices = std::array<DenseIndex, NumSlots>
using	HessianSlots = std::array<DenseIndex, NumSlots + 1>
using	RhsVector = Eigen::Matrix<double, ErrorDim, 1>
template<int BlockDim>
using	BlockMatrix = Eigen::Matrix<double, ErrorDim, BlockDim>
template<int BlockDim>
using	BlockVector
using	Blocks = std::tuple<BlockVector<BlockDims>...>
Public Types inherited from gtsam::GaussianFactor
typedef GaussianFactor	This
	This class.
typedef std::shared_ptr< This >	shared_ptr
	shared_ptr to this class
typedef Factor	Base
	Our base class.
Public Types inherited from gtsam::Factor
typedef KeyVector::iterator	iterator
	Iterator over keys.
typedef KeyVector::const_iterator	const_iterator
	Const iterator over keys.

Additional Inherited Members
template<typename CONTAINER>
static DenseIndex	Slot (const CONTAINER &keys, Key key)
	Factor ()
	Default constructor for I/O.
template<typename CONTAINER>
	Factor (const CONTAINER &keys)
	Construct factor from container of keys.
template<typename ITERATOR>
	Factor (ITERATOR first, ITERATOR last)
	Construct factor from iterator keys.
template<typename CONTAINER>
static Factor	FromKeys (const CONTAINER &keys)
	Construct factor from container of keys.
template<typename ITERATOR>
static Factor	FromIterators (ITERATOR first, ITERATOR last)
	Construct factor from iterator keys.
Protected Attributes inherited from gtsam::Factor
KeyVector	keys_
	The keys involved in this factor.

Member Typedef Documentation

◆ BlockVector

template<int ErrorDim, int... BlockDims>

template<int BlockDim>

using gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::BlockVector

Initial value:

 
std::vector<BlockMatrix<BlockDim>,
            Eigen::aligned_allocator<BlockMatrix<BlockDim>>>

Constructor & Destructor Documentation

◆ BatchJacobianFactor()

template<int ErrorDim, int... BlockDims>

gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::BatchJacobianFactor	(	const KeyVector &	keys,
		std::vector< size_t >	keyDims,
		const SharedDiagonal &	model = SharedDiagonal() )

inline

Construct an empty compact batch factor with known key dimensions.

Parameters

keys	Union of all keys represented by this batch factor.
keyDims	Dimensions corresponding to keys, in the same order.
model	Optional diagonal model on the stored rows.

Member Function Documentation

◆ addRow()

template<int ErrorDim, int... BlockDims>

void gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::addRow	(	const SlotIndices &	slots,
		const std::vector< Matrix > &	blocks,
		const Vector &	rhs )

inline

Add one row group corresponding to one original nonlinear factor.

Parameters

slots	Local key slots touched by this row group. Each entry indexes into this factor's keys() vector.
blocks	Fixed-size Jacobian blocks, one for each factor variable.
rhs	Right-hand side vector for this row group.

◆ clone()

template<int ErrorDim, int... BlockDims>

GaussianFactor::shared_ptr gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::clone ( ) const

inlineoverridevirtual

Return a deep copy as a GaussianFactor.

Implements gtsam::GaussianFactor.

◆ deltaError()

template<int ErrorDim, int... BlockDims>

double gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::deltaError	(	const VectorValues &	values,
		double *	oldError = nullptr,
		double *	newError = nullptr ) const

inlineoverridevirtual

Evaluate the linear-model error change directly from compact rows.

Reimplemented from gtsam::GaussianFactor.

◆ get_model()

template<int ErrorDim, int... BlockDims>

const SharedDiagonal & gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::get_model ( ) const

inlineoverridevirtual

Return the optional diagonal model on the stored rows.

Implements gtsam::BatchJacobianFactorBase.

◆ getDim()

template<int ErrorDim, int... BlockDims>

DenseIndex gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::getDim ( const_iterator variable ) const

inlineoverridevirtual

Return the dimension of the variable at the given key iterator.

Implements gtsam::GaussianFactor.

◆ gradientAtZeroAdd()

template<int ErrorDim, int... BlockDims>

void gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::gradientAtZeroAdd	(	const std::vector< size_t > &	scalarOffsets,
		double *	gradient ) const

inlineoverridevirtual

Add -A.transpose()*W*b directly to a flat gradient vector.

Reimplemented from gtsam::BatchJacobianFactorBase.

◆ hessianBlockDiagonalAdd()

template<int ErrorDim, int... BlockDims>

void gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::hessianBlockDiagonalAdd	(	const std::vector< size_t > &	blockSlots,
		std::vector< Matrix > *	diagonalBlocks ) const

inlineoverridevirtual

Add compact A.transpose()*W*A diagonal blocks without keyed maps.

Reimplemented from gtsam::BatchJacobianFactorBase.

◆ hessianDiagonalAdd()

template<int ErrorDim, int... BlockDims>

void gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::hessianDiagonalAdd ( VectorValues & diagonal ) const

inlineoverridevirtual

Accumulate the Hessian scalar diagonal directly from compact rows.

Implements gtsam::GaussianFactor.

◆ multiplyHessianAdd()

template<int ErrorDim, int... BlockDims>

void gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::multiplyHessianAdd	(	double	alpha,
		const std::vector< size_t > &	scalarOffsets,
		const double *	x,
		double *	y ) const

inlineoverridevirtual

Add alpha*A.transpose()*W*A*x directly to a flat output vector.

Reimplemented from gtsam::BatchJacobianFactorBase.

◆ rows()

template<int ErrorDim, int... BlockDims>

size_t gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::rows ( ) const

inlineoverridevirtual

Return the number of scalar rows represented by all row groups.

Implements gtsam::BatchJacobianFactorBase.

◆ toJacobianFactor()

template<int ErrorDim, int... BlockDims>

JacobianFactor gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::toJacobianFactor ( ) const

inlineoverridevirtual

Convert compact row-block storage into a conventional JacobianFactor.

This compatibility path allocates dense blocks for every key in the batch, including structural zeros. Performance-critical solvers use the internal compact-factor elimination adapter.

Implements gtsam::BatchJacobianFactorBase.

◆ updateHessian() [1/2]

template<int ErrorDim, int... BlockDims>

void gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::updateHessian	(	const KeyVector &	infoKeys,
		SymmetricBlockMatrix *	info ) const

inlineoverridevirtual

Add this factor's augmented information matrix to info.

This direct implementation is the legacy Cholesky fast path. It accumulates fixed-size row-group products into the target SymmetricBlockMatrix and avoids constructing a dense compatibility JacobianFactor.

Implements gtsam::GaussianFactor.

◆ updateHessian() [2/2]

template<int ErrorDim, int... BlockDims>

void gtsam::BatchJacobianFactor< ErrorDim, BlockDims >::updateHessian	(	const KeyVector &	infoKeys,
		SymmetricBlockMatrix *	info,
		DenseIndex	beginCol,
		DenseIndex	endCol ) const

inlineoverridevirtual

Add this factor's augmented information matrix over a block-column range.

This overload supports the partial-column Hessian assembly used by parallel Cholesky paths. It follows the same column ownership convention as JacobianFactor: a block (I,J) is updated when max(I,J) lies in [beginCol,endCol).

Implements gtsam::GaussianFactor.

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

/tmp/gtsam-4.3.0-doxygen.rsXPUS/source/gtsam/linear/BatchJacobianFactor.h

Detailed Description

Public Member Functions

Static Public Attributes

Public Types

Additional Inherited Members

Member Typedef Documentation

◆ BlockVector

Constructor & Destructor Documentation

◆ BatchJacobianFactor()

Member Function Documentation

◆ addRow()

◆ clone()

◆ deltaError()

◆ get_model()

◆ getDim()

◆ gradientAtZeroAdd()

◆ hessianBlockDiagonalAdd()

◆ hessianDiagonalAdd()

◆ multiplyHessianAdd()

◆ rows()

◆ toJacobianFactor()

◆ updateHessian() [1/2]

◆ updateHessian() [2/2]