#include <quala/decl/lbfgs.hpp>

Detailed Description

Limited memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) algorithm.

Definition at line 72 of file decl/lbfgs.hpp.

Collaboration diagram for LBFGS:

Public Types
enum class	Sign { Positive , Negative }
	The sign of the vectors \( p \) passed to the LBFGS::update method. More...

using	Params = LBFGSParams

Public Member Functions
	LBFGS (Params params)

	LBFGS (Params params, length_t n)

template<class VecS , class VecY >
bool	update_sy (const anymat< VecS > &s, const anymat< VecY > &y, real_t pₙₑₓₜᵀpₙₑₓₜ, bool forced=false)
	Update the inverse Hessian approximation using the new vectors sₖ = xₖ₊₁ - xₖ and yₖ = pₖ₊₁ - pₖ. More...

bool	update (crvec xₖ, crvec xₙₑₓₜ, crvec pₖ, crvec pₙₑₓₜ, Sign sign=Sign::Positive, bool forced=false)
	Update the inverse Hessian approximation using the new vectors xₖ₊₁ and pₖ₊₁. More...

bool	apply (rvec q, real_t γ=-1)
	Apply the inverse Hessian approximation to the given vector q. More...

template<class IndexVec >
bool	apply (rvec q, real_t γ, const IndexVec &J)
	Apply the inverse Hessian approximation to the given vector q, applying only the columns and rows of the Hessian in the index set J. More...

void	reset ()
	Throw away the approximation and all previous vectors s and y. More...

void	resize (length_t n)
	Re-allocate storage for a problem with a different size. More...

void	scale_y (real_t factor)
	Scale the stored y vectors by the given factor. More...

const Params &	get_params () const
	Get the parameters. More...

length_t	n () const
	Get the size of the s and y vectors in the buffer. More...

length_t	history () const
	Get the number of previous vectors s and y stored in the buffer. More...

index_t	succ (index_t i) const
	Get the next index in the circular buffer of previous s and y vectors. More...

index_t	pred (index_t i) const
	Get the previous index in the circular buffer of s and y vectors. More...

length_t	current_history () const
	Get the number of previous s and y vectors currently stored in the buffer. More...

auto	s (index_t i)

auto	s (index_t i) const

auto	y (index_t i)

auto	y (index_t i) const

real_t &	ρ (index_t i)

const real_t &	ρ (index_t i) const

real_t &	α (index_t i)

const real_t &	α (index_t i) const

template<class F >
void	foreach_fwd (const F &fun) const
	Iterate over the indices in the history buffer, oldest first. More...

template<class F >
void	foreach_rev (const F &fun) const
	Iterate over the indices in the history buffer, newest first. More...

Static Public Member Functions
static bool	update_valid (const LBFGSParams &params, real_t yᵀs, real_t sᵀs, real_t pᵀp)
	Check if the new vectors s and y allow for a valid BFGS update that preserves the positive definiteness of the Hessian approximation. More...

Private Attributes
LBFGSStorage	sto

index_t	idx = 0

bool	full = false

Params	params

Member Typedef Documentation

◆ Params

using Params = LBFGSParams

Definition at line 74 of file decl/lbfgs.hpp.

Member Enumeration Documentation

◆ Sign

enum class Sign

strong

The sign of the vectors \( p \) passed to the LBFGS::update method.

Enumerator
Positive	\( p \sim \nabla \psi(x) \)
Negative	\( p \sim -\nabla \psi(x) \)

Definition at line 78 of file decl/lbfgs.hpp.

Constructor & Destructor Documentation

◆ LBFGS() [1/2]

LBFGS ( Params params )

inline

Definition at line 83 of file decl/lbfgs.hpp.

◆ LBFGS() [2/2]

LBFGS	(	Params	params,
		length_t	n
	)

inline

Definition at line 84 of file decl/lbfgs.hpp.

Here is the call graph for this function:

Member Function Documentation

◆ update_valid()

bool update_valid	(	const LBFGSParams &	params,
		real_t	yᵀs,
		real_t	sᵀs,
		real_t	pᵀp
	)

inlinestatic

Check if the new vectors s and y allow for a valid BFGS update that preserves the positive definiteness of the Hessian approximation.

Definition at line 9 of file lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ update_sy()

bool update_sy	(	const anymat< VecS > &	s,
		const anymat< VecY > &	y,
		real_t	pₙₑₓₜᵀpₙₑₓₜ,
		bool	forced = `false`
	)

inline

Update the inverse Hessian approximation using the new vectors sₖ = xₖ₊₁ - xₖ and yₖ = pₖ₊₁ - pₖ.

Definition at line 34 of file lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ update()

bool update	(	crvec	xₖ,
		crvec	xₙₑₓₜ,
		crvec	pₖ,
		crvec	pₙₑₓₜ,
		Sign	sign = `Sign::Positive`,
		bool	forced = `false`
	)

inline

Update the inverse Hessian approximation using the new vectors xₖ₊₁ and pₖ₊₁.

Definition at line 56 of file lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ apply() [1/2]

bool apply	(	rvec	q,
		real_t	γ = `-1`
	)

inline

Apply the inverse Hessian approximation to the given vector q.

Initial inverse Hessian approximation is set to \( H_0 = \gamma I \). If γ is negative, \( H_0 = \frac{s^\top y}{y^\top y} I \).

Definition at line 64 of file lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ apply() [2/2]

bool apply	(	rvec	q,
		real_t	γ,
		const IndexVec &	J
	)

Apply the inverse Hessian approximation to the given vector q, applying only the columns and rows of the Hessian in the index set J.

Definition at line 93 of file lbfgs.hpp.

Here is the call graph for this function:

◆ reset()

void reset ( )

inline

Throw away the approximation and all previous vectors s and y.

Definition at line 180 of file lbfgs.hpp.

Here is the caller graph for this function:

◆ resize()

void resize ( length_t n )

inline

Re-allocate storage for a problem with a different size.

Causes a reset.

Definition at line 185 of file lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ scale_y()

void scale_y ( real_t factor )

inline

Scale the stored y vectors by the given factor.

Definition at line 196 of file lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ get_params()

const Params & get_params ( ) const

inline

Get the parameters.

Definition at line 122 of file decl/lbfgs.hpp.

Here is the caller graph for this function:

◆ n()

length_t n ( ) const

inline

Get the size of the s and y vectors in the buffer.

Definition at line 125 of file decl/lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ history()

length_t history ( ) const

inline

Get the number of previous vectors s and y stored in the buffer.

Definition at line 127 of file decl/lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ succ()

index_t succ ( index_t i ) const

inline

Get the next index in the circular buffer of previous s and y vectors.

Definition at line 129 of file decl/lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ pred()

index_t pred ( index_t i ) const

inline

Get the previous index in the circular buffer of s and y vectors.

Definition at line 131 of file decl/lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ current_history()

length_t current_history ( ) const

inline

Get the number of previous s and y vectors currently stored in the buffer.

Definition at line 134 of file decl/lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ s() [1/2]

auto s ( index_t i )

inline

Definition at line 136 of file decl/lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ s() [2/2]

auto s ( index_t i ) const

inline

Definition at line 137 of file decl/lbfgs.hpp.

Here is the call graph for this function:

◆ y() [1/2]

auto y ( index_t i )

inline

Definition at line 138 of file decl/lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ y() [2/2]

auto y ( index_t i ) const

inline

Definition at line 139 of file decl/lbfgs.hpp.

Here is the call graph for this function:

◆ ρ() [1/2]

real_t & ρ ( index_t i )

inline

Definition at line 140 of file decl/lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ ρ() [2/2]

const real_t & ρ ( index_t i ) const

inline

Definition at line 141 of file decl/lbfgs.hpp.

Here is the call graph for this function:

◆ α() [1/2]

real_t & α ( index_t i )

inline

Definition at line 142 of file decl/lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ α() [2/2]

const real_t & α ( index_t i ) const

inline

Definition at line 143 of file decl/lbfgs.hpp.

Here is the call graph for this function:

◆ foreach_fwd()

void foreach_fwd ( const F & fun ) const

inline

Iterate over the indices in the history buffer, oldest first.

Definition at line 147 of file decl/lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ foreach_rev()

void foreach_rev ( const F & fun ) const

inline

Iterate over the indices in the history buffer, newest first.

Definition at line 158 of file decl/lbfgs.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

Member Data Documentation

◆ sto

LBFGSStorage sto

private

Definition at line 168 of file decl/lbfgs.hpp.

◆ idx

index_t idx = 0

private

Definition at line 169 of file decl/lbfgs.hpp.

◆ full

bool full = false

private

Definition at line 170 of file decl/lbfgs.hpp.

◆ params

Params params

private

Definition at line 171 of file decl/lbfgs.hpp.

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

Detailed Description

Public Types

Public Member Functions

Static Public Member Functions

Private Attributes

Member Typedef Documentation

◆ Params

Member Enumeration Documentation

◆ Sign

Constructor & Destructor Documentation

◆ LBFGS() [1/2]

◆ LBFGS() [2/2]

Member Function Documentation

◆ update_valid()

◆ update_sy()

◆ update()

◆ apply() [1/2]

◆ apply() [2/2]

◆ reset()

◆ resize()

◆ scale_y()

◆ get_params()

◆ n()

◆ history()

◆ succ()

◆ pred()

◆ current_history()

◆ s() [1/2]

◆ s() [2/2]

◆ y() [1/2]

◆ y() [2/2]

◆ ρ() [1/2]

◆ ρ() [2/2]

◆ α() [1/2]

◆ α() [2/2]

◆ foreach_fwd()

◆ foreach_rev()

Member Data Documentation

◆ sto

◆ idx

◆ full

◆ params