ConicBundle
Public Member Functions | Private Attributes | List of all members

represents a polyhedral feasible ground set by affine constraints as in Linear Programming with special support for simple box constraints More...

#include <LPGroundset.hxx>

Inheritance diagram for ConicBundle::LPGroundset:
ConicBundle::Groundset ConicBundle::VariableMetricModel ConicBundle::CBout

Public Member Functions

virtual void clear (CH_Matrix_Classes::Integer indim=0, CH_Matrix_Classes::Integer in_groundset_id=0)
 resets all values as described in Groundset::clear()
 
 LPGroundset (CBout *cb=0)
 calls clear() with the same parameters
 
 LPGroundset (CH_Matrix_Classes::Integer dim, const CH_Matrix_Classes::Matrix *lbyp=0, const CH_Matrix_Classes::Matrix *ubyp=0, const CH_Matrix_Classes::Sparsemat *Gp=0, const CH_Matrix_Classes::Matrix *rhslbp=0, const CH_Matrix_Classes::Matrix *rhsubp=0, const CH_Matrix_Classes::Matrix *start_val=0, const CH_Matrix_Classes::Matrix *costs=0, const CH_Matrix_Classes::Real offset=0., CH_Matrix_Classes::Integer in_groundset_id=0, CBout *cb=0)
 allows to specify the groundset in the constructor, zero is allowed everywhere
 
CH_Matrix_Classes::Integer get_groundset_id () const
 returns the current groundset_id, increased values indicate changes in the ground set
 
void set_groundset_id (CH_Matrix_Classes::Integer gsid)
 sets the groundset_id to the desired value, increasing it is safer here because this is used to indicate changes
 
virtual CH_Matrix_Classes::Integer get_dim () const
 returns the dimension of the ground set, i.e., the length of the variables vector y
 
int set_qpsolver (QPSolverParametersObject *qpparams, QPSolverObject *qpsolver=0)
 Set the qp solver's parameters to qpparams (if not null); if the second argument qpsolver is also given, the old solver is first discarded and replaced by this new solver and then the parameters are set (if given). Any object passed here will be owned and deleted by *this. For correct continuaton a new qpsolver needs to have the same feasible set as the current solver but this must be ensured by the caller.
 
bool constrained () const
 returns false if the feasible set is the entire space (unconstrained optimization), true otherwise.
 
const CH_Matrix_Classes::Matrixget_lby () const
 number of linear constraints More...
 
const CH_Matrix_Classes::Matrixget_uby () const
 returns the upper bounds vector on y if it exists
 
virtual bool is_feasible (CH_Matrix_Classes::Integer &in_groundset_id, const CH_Matrix_Classes::Matrix &y, CH_Matrix_Classes::Real relprec=1e-10)
 returns the indices of variable lower bounds > ConicBundle::CB_minus_infinity More...
 
int ensure_feasibility (CH_Matrix_Classes::Integer &in_groundset_id, CH_Matrix_Classes::Matrix &y, bool &ychanged, BundleProxObject *Hp, CH_Matrix_Classes::Real relprec=1e-10)
 makes y feasible if not so, see Groundset::ensure_feasibility()
 
virtual QPSolverObjectget_qp_solver (bool &solves_model_without_gs, BundleProxObject *Hp)
 returns a pointer to an internal QPSolverObject that is able to solve bundle suproblems efficiently for this kind of groundset and scaling; if solves_model_without_gs == true the qp solver does not include the groundset and the groundset has to be dealt with by the Gauss Seidel approach
 
int set_qp_solver_parameters (QPSolverParametersObject *)
 set parameters for the QP_Solver
 
virtual const CH_Matrix_Classes::Matrixget_starting_point () const
 returns a stored starting point, note: this need not be feasible; if generated automatically, its dimension is correct.
 
virtual int set_starting_point (const CH_Matrix_Classes::Matrix &vec)
 stores the a new starting point irrespective of whether it is feasible or not and returns 0 if it feasible, 1 if it is infeasible
 
virtual int candidate (CH_Matrix_Classes::Integer &gs_id, CH_Matrix_Classes::Matrix &newy, CH_Matrix_Classes::Real &cand_gs_val, CH_Matrix_Classes::Real &linval, CH_Matrix_Classes::Real &augval_lb, CH_Matrix_Classes::Real &augval_ub, CH_Matrix_Classes::Real &subgnorm2, const CH_Matrix_Classes::Matrix &center_y, CH_Matrix_Classes::Real center_value, const MinorantPointer &model_minorant, BundleProxObject *Hp, MinorantPointer *delta_groundset_minorant=0, CH_Matrix_Classes::Indexmatrix *delta_index=0, CH_Matrix_Classes::Real relprec=1e-2)
 computes the next ground set minorant and candidate, see Groundset::candidate()
 
virtual const MinorantPointerget_gs_aggregate () const
 returns the groundset aggregate computed in candidate()
 
const MinorantPointerget_gs_minorant () const
 returns the linear minorant valid on the entire ground set (e.g. a linear cost funciton)
 
virtual const CH_Matrix_Classes::Indexmatrixget_yfixed () const
 if not NULL (iff get_use_yfixing()==false) it returns the vector yfixed with yfixed(i)=0 if not fixed, =1 is fixed already, =2 if newly fixed
 
virtual CH_Matrix_Classes::Indexmatrixset_yfixed ()
 if not NULL (iff get_use_yfixing()==false) returns the vector yfixed with yfixed(i)=0 if not fixed, =1 is fixed already, =2 if newly fixed
 
bool get_use_yfixing () const
 true if the cooridinate fixing heuristic is switched on (only constrained cases)
 
void set_use_yfixing (bool uyf)
 set to true to switch on the cooridinate fixing heuristic (only constrained cases)
 
int set_variable_metric_selection (VariableMetricSelection *vms=0)
 delete old selector and set a new one (0 is allowed resulting in no local selector)
 
VariableMetricSelectionget_variable_metric_selection () const
 delete old selector and set a new one (0 is allowed resulting in no local selector)
 
int add_variable_metric (VariableMetric &H, CH_Matrix_Classes::Integer y_id, const CH_Matrix_Classes::Matrix &y, bool descent_step, CH_Matrix_Classes::Real weightu, CH_Matrix_Classes::Real model_maxviol, const CH_Matrix_Classes::Indexmatrix *indices=0)
 see VariableMetric
 
GroundsetModificationstart_modification ()
 propagates the call to QPSolverObject::QPstart_modification() of the current qpsolver
 
int apply_modification (const GroundsetModification &mdf)
 change the groundset description as specified by the argument
 
int mfile_data (std::ostream &out) const
 m-file output routine for debugging or testing in Matlab (not yet working)
 
void set_cbout (const CBout *cb, int incr=-1)
 output settings
 
- Public Member Functions inherited from ConicBundle::VariableMetricModel
 VariableMetricModel (CBout *cb=0, int cbincr=-1)
 constructor for passing on ouptut information
 
virtual ~VariableMetricModel ()
 virtual destructor
 
virtual VariableMetricModelvariable_metric_transform ()
 Overload this in order apply transformations in between. More...
 
- Public Member Functions inherited from ConicBundle::CBout
virtual void set_out (std::ostream *out=0, int print_level=1)
 Specifies the output level (out==NULL: no output at all, out!=NULL and level=0: errors and warnings, level>0 increasingly detailed information) More...
 
void clear_cbout ()
 reset to default settings (out=0,print_level=1)
 
 CBout (const CBout *cb=0, int incr=-1)
 calls set_cbout
 
 CBout (std::ostream *outp, int pl=1)
 initialize correspondingly
 
 CBout (const CBout &cb, int incr=0)
 copy constructor
 
virtual bool cb_out (int pl=-1) const
 Returns true if out!=0 and (pl<print_level), pl<0 should be used for WARNINGS and ERRORS only, pl==0 for usual output.
 
std::ostream & get_out () const
 If cb_out() returned true, this returns the output stream, but it will abort if called with out==0.
 
std::ostream * get_out_ptr () const
 returns the pointer to the output stream
 
int get_print_level () const
 returns the print_level
 

Private Attributes

CH_Matrix_Classes::Integer dim
 current dimension of the ground set
 
CH_Matrix_Classes::Integer groundset_id
 nonnegative update counter for recognizing changes in the groundset description
 
CH_Matrix_Classes::Matrix starting_point
 the starting point
 
MinorantPointer gs_aggregate
 the ground set aggregate, here giving the linear coefficients of the affine cost function
 
bool use_yfixing
 if set to true in constrained versions, y-variables that are at strongly active bounds should be fixed to this value by indicating so in yfixed
 
CH_Matrix_Classes::Indexmatrix yfixed
 in constrained versions, when use_yfixing is true, this indicates whether a coordinate should be considered as free (0), fixed (1), or newly fixed (2)
 
QPSolverObjectqpsolver
 general solver for all groundsets
 
UQPSolver uqpsolver
 special solver for easy groundsets
 
QPSolverObjectqp_solver
 the selected solver
 
bool solve_model_without_gs
 selected varaint of solving the subproblem
 
BundleProxObjectHp
 points to the quadratic cost matrix inside ensure_feasibility() and candidate(), otherwise ==NULL
 
CH_Matrix_Classes::Matrix c
 linear cost term for QP-subproblems
 
CH_Matrix_Classes::Real gamma
 constant offset for QP-subproblems
 
MinorantPointer gs_minorant
 affine cost function for the ground set
 
VariableMetricSelectionvm_selection
 specifieable routine for computing a metric contribution of the groundset, see Variable Metric
 
bool store_old_minorants
 false by default, set to true if variable metric is used
 
MinorantBundle old_minorants
 minorants collected for variable metric purpose
 
CH_Matrix_Classes::Integer max_minorants
 maximum number of minorants
 
CH_Matrix_Classes::Integer minorant_nexti
 used by the variable metric low rank heuristic
 
CH_Matrix_Classes::Matrix old_lowrank
 used by the variable metric low rank heuristic
 
CH_Matrix_Classes::Matrix old_diagonal
 used by the variable metric low rank heuristic
 
CH_Matrix_Classes::Symmatrix old_sym
 used by the variable metric heuristic
 

Detailed Description

represents a polyhedral feasible ground set by affine constraints as in Linear Programming with special support for simple box constraints

The polyhedral feasible set is described in the form

\[Y=\{y\in\mathbf{R}^m\colon l\le y\le u, r_l\le Ay \le r_u\} \]

where $l\le u$ are $m$-vectors, $A$ is an $m\times n$ matrix and $r_l\le r_u$ are $n$-vectors. $m$ and $n$ may be zero (empty), lower and upper bounds may take values ConicBundle::CB_minus_infinity and ConicBundle::CB_plus_infinity to implement appropriate inequalties.

For general polyhedral ground sets the two routines ensure_feasibility() and candidate() need to solve a general convex QP; this is currently done by interior point methods.

If there are only box constraints and a diagonal scaling matrix for the proximal term, then all computations are done directly witout the need to solve a full convex QP. In this case the routine also offers a scaling heuristic and a heuristic for fixing y variables at their bounds.

Member Function Documentation

◆ get_lby()

const CH_Matrix_Classes::Matrix* ConicBundle::LPGroundset::get_lby ( ) const

number of linear constraints

returns the lower bounds vector on y if it exists

Referenced by constrained().

◆ is_feasible()

virtual bool ConicBundle::LPGroundset::is_feasible ( CH_Matrix_Classes::Integer in_groundset_id,
const CH_Matrix_Classes::Matrix y,
CH_Matrix_Classes::Real  relprec = 1e-10 
)
virtual

returns the indices of variable lower bounds > ConicBundle::CB_minus_infinity

returns the indices of variable lower bounds < ConicBundle::CB_plus_infinity returns the constraint matrix of the feasible set returns the constraint lower bounds returns the constraint upper bounds returns the indices with constraint lower bound slacks returns the indices with constraint upper bound slacks returns true if still feasible, see Groundset::is_feasible()

Implements ConicBundle::Groundset.

Referenced by constrained(), and set_starting_point().


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