gtsam::SubgraphPreconditioner Class Reference

Detailed Description

Subgraph conditioner class, as explained in the RSS 2010 submission.

Starting with a graph A*x=b, we split it in two systems A1*x=b1 and A2*x=b2 We solve R1*x=c1, and make the substitution y=R1*x-c1. To use the class, give the Bayes Net R1*x=c1 and Graph A2*x=b2. Then solve for yhat using CG, and solve for xhat = system.x(yhat).

Inheritance diagram for gtsam::SubgraphPreconditioner:

Public Member Functions
	SubgraphPreconditioner (const SubgraphPreconditionerParameters &p=SubgraphPreconditionerParameters())
	SubgraphPreconditioner (const sharedFG &Ab2, const sharedBayesNet &Rc1, const sharedValues &xbar, const SubgraphPreconditionerParameters &p=SubgraphPreconditionerParameters())
	Constructor. More...
void	print (const std::string &s="SubgraphPreconditioner") const
	print the object
const sharedFG &	Ab2 () const
	Access Ab2.
const sharedBayesNet &	Rc1 () const
	Access Rc1.
const sharedErrors	b2bar () const
	Access b2bar.
VectorValues	x (const VectorValues &y) const
	Add zero-mean i.i.d. More...
VectorValues	zero () const
void	transposeMultiplyAdd2 (double alpha, Errors::const_iterator begin, Errors::const_iterator end, VectorValues &y) const
	Add constraint part of the error only y += alpha*inv(R1')*A2'*e2 Takes a range indicating e2 !!!!
double	error (const VectorValues &y) const
VectorValues	gradient (const VectorValues &y) const
	gradient = y + inv(R1')*A2'*(A2*inv(R1)*y-b2bar)
Errors	operator* (const VectorValues &y) const
	Apply operator A.
void	multiplyInPlace (const VectorValues &y, Errors &e) const
	Apply operator A in place: needs e allocated already.
VectorValues	operator^ (const Errors &e) const
	Apply operator A'.
void	transposeMultiplyAdd (double alpha, const Errors &e, VectorValues &y) const
	Add A'e to y y += alpha*A'[e1;e2] = [alpha*e1; alpha*inv(R1')*A2'*e2].
void	solve (const Vector &y, Vector &x) const override
	implement x = R^{-1} y More...
void	transposeSolve (const Vector &y, Vector &x) const override
	implement x = R^{-T} y More...
void	build (const GaussianFactorGraph &gfg, const KeyInfo &info, const std::map< Key, Vector > &lambda) override
	build/factorize the preconditioner More...
virtual void	solve (const Vector &y, Vector &x) const =0
	implement x = L^{-1} y More...
virtual void	transposeSolve (const Vector &y, Vector &x) const =0
	implement x = L^{-T} y More...
virtual void	build (const GaussianFactorGraph &gfg, const KeyInfo &info, const std::map< Key, Vector > &lambda)=0
	build/factorize the preconditioner More...

Public Types
typedef boost::shared_ptr< SubgraphPreconditioner >	shared_ptr
typedef boost::shared_ptr< const GaussianBayesNet >	sharedBayesNet
typedef boost::shared_ptr< const GaussianFactorGraph >	sharedFG
typedef boost::shared_ptr< const VectorValues >	sharedValues
typedef boost::shared_ptr< const Errors >	sharedErrors
Public Types inherited from gtsam::Preconditioner
typedef boost::shared_ptr< Preconditioner >	shared_ptr
typedef std::vector< size_t >	Dimensions

Constructor & Destructor Documentation

SubgraphPreconditioner()

gtsam::SubgraphPreconditioner::SubgraphPreconditioner	(	const sharedFG &	Ab2,
		const sharedBayesNet &	Rc1,
		const sharedValues &	xbar,
		const SubgraphPreconditionerParameters &	p = SubgraphPreconditionerParameters()
	)

Constructor.

Parameters

Ab2	the Graph A2*x=b2
Rc1	the Bayes Net R1*x=c1
xbar	the solution to R1*x=c1

Member Function Documentation

build()

void gtsam::SubgraphPreconditioner::build ( const GaussianFactorGraph &  gfg, const KeyInfo &  info, const std::map< Key, Vector > &  lambda  )

overridevirtual

build/factorize the preconditioner

Implements gtsam::Preconditioner.

solve()

void gtsam::SubgraphPreconditioner::solve ( const Vector &  y, Vector &  x  ) const

overridevirtual

implement x = R^{-1} y

Implements gtsam::Preconditioner.

transposeSolve()

void gtsam::SubgraphPreconditioner::transposeSolve ( const Vector &  y, Vector &  x  ) const

overridevirtual

implement x = R^{-T} y

Implements gtsam::Preconditioner.

x()

VectorValues gtsam::SubgraphPreconditioner::x

(

const VectorValues &

y

)

const

Add zero-mean i.i.d.

Gaussian prior terms to each variable

Parameters

sigma

Standard deviation of Gaussian

---

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

• /Users/dellaert/git/github/gtsam/linear/SubgraphPreconditioner.h
• /Users/dellaert/git/github/gtsam/linear/SubgraphPreconditioner.cpp