Cut Selectors

Detailed Description

methods and files provided by the default cut selectors of SCIP

A detailed description what a cut selector does and how to add a cut selector to SCIP can be found here.

Modules
	Inclusion methods
	methods to include specific cut selectors into SCIP

Functions
SCIP_RETCODE	SCIPselectCutsDynamic (SCIP *scip, SCIP_ROW **cuts, SCIP_ROW **forcedcuts, SCIP_RANDNUMGEN *randnumgen, char filtermode, SCIP_Real mingain, SCIP_Real maxparall, SCIP_Real dircutoffdistweight, SCIP_Real efficacyweight, SCIP_Real objparalweight, SCIP_Real intsupportweight, int ncuts, int nforcedcuts, int maxselectedcuts, int *nselectedcuts)
SCIP_RETCODE	SCIPselectCutsEnsemble (SCIP *scip, SCIP_ROW **cuts, SCIP_ROW **forcedcuts, SCIP_CUTSELDATA *cutseldata, SCIP_Bool root, int ncuts, int nforcedcuts, int maxselectedcuts, int *nselectedcuts)
SCIP_RETCODE	SCIPselectCutsHybrid (SCIP *scip, SCIP_ROW **cuts, SCIP_ROW **forcedcuts, SCIP_RANDNUMGEN *randnumgen, SCIP_Real goodscorefac, SCIP_Real badscorefac, SCIP_Real goodmaxparall, SCIP_Real maxparall, SCIP_Real dircutoffdistweight, SCIP_Real efficacyweight, SCIP_Real objparalweight, SCIP_Real intsupportweight, int ncuts, int nforcedcuts, int maxselectedcuts, int *nselectedcuts)

Files
file	cutsel_dynamic.h
	dynamic cut selector
file	cutsel_ensemble.h
	ensemble cut selector
file	cutsel_hybrid.h
	hybrid cut selector

Function Documentation

SCIPselectCutsDynamic()

SCIP_RETCODE SCIPselectCutsDynamic	(	SCIP *	scip,
		SCIP_ROW **	cuts,
		SCIP_ROW **	forcedcuts,
		SCIP_RANDNUMGEN *	randnumgen,
		char	filtermode,
		SCIP_Real	mingain,
		SCIP_Real	maxparall,
		SCIP_Real	dircutoffdistweight,
		SCIP_Real	efficacyweight,
		SCIP_Real	objparalweight,
		SCIP_Real	intsupportweight,
		int	ncuts,
		int	nforcedcuts,
		int	maxselectedcuts,
		int *	nselectedcuts
	)

perform a cut selection algorithm for the given array of cuts

This is an extension of the hybrid cutselector to employ a dynamic range when applying orthogonality filtering, dependent on the efficacy ratio between cuts.

The input cuts array should be resorted such that the selected cuts come first.

perform a cut selection algorithm for the given array of cuts

This is the selection method of the dynamic cut selector which implements the dynamic orthognality filtering based on the ratio of efficacies. The input cuts array gets resorted s.t the selected cuts come first and the remaining ones are the end.

Parameters

scip	SCIP data structure
cuts	array with cuts to perform selection algorithm
forcedcuts	array with forced cuts
randnumgen	random number generator for tie-breaking, or NULL
filtermode	filtering strategy during cut selection ( 'd'ynamic- and 'f'ull dynamic parallelism)
mingain	minimum efficacy gain in percentage to filter cuts
maxparall	maximal parallelism for all cuts that are not good
dircutoffdistweight	weight of directed cutoff distance in cut score calculation
efficacyweight	weight of efficacy in cut score calculation
objparalweight	weight of objective parallelism in cut score calculation
intsupportweight	weight of integral support in cut score calculation
ncuts	number of cuts in cuts array
nforcedcuts	number of forced cuts
maxselectedcuts	maximal number of cuts from cuts array to select
nselectedcuts	pointer to return number of selected cuts from cuts array

Definition at line 540 of file cutsel_dynamic.c.

References computeProjectionScore(), filterWithDynamicParallelism(), NULL, SCIP_CALL, SCIP_OKAY, SCIP_Real, SCIPallocBufferArray, SCIPdebugMsg, SCIPfreeBufferArray, SCIProwGetName(), scoring(), and selectBestCut().

Referenced by SCIPincludeCutselDynamic().

SCIPselectCutsEnsemble()

SCIP_RETCODE SCIPselectCutsEnsemble	(	SCIP *	scip,
		SCIP_ROW **	cuts,
		SCIP_ROW **	forcedcuts,
		SCIP_CUTSELDATA *	cutseldata,
		SCIP_Bool	root,
		int	ncuts,
		int	nforcedcuts,
		int	maxselectedcuts,
		int *	nselectedcuts
	)

perform a cut selection algorithm for the given array of cuts

This is the selection method of the ensemble cut selector. It uses a weighted sum of normalised efficacy, normalised directed cutoff distance, normalised expected improvements, objective parallelism, integer support, sparsity, dynamism, pseudo-costs, and variable locks. In addition to the weighted sum score, there are optionally parallelism-based filtering and penalties, and density filtering. There are also additional budget constraints on the number of cuts that should be added. The input cuts array gets re-sorted such that the selected cuts come first and the remaining ones are the end.

Parameters

scip	SCIP data structure
cuts	array with cuts to perform selection algorithm
forcedcuts	array with forced cuts
cutseldata	cut selector data
root	whether we are at the root node or not
ncuts	number of cuts in cuts array
nforcedcuts	number of forced cuts
maxselectedcuts	maximal number of cuts from cuts array to select
nselectedcuts	pointer to return number of selected cuts from cuts array

Definition at line 720 of file cutsel_ensemble.c.

References filterWithDensity(), filterWithParallelism(), NULL, penaliseWithParallelism(), SCIP_CALL, SCIP_OKAY, SCIP_Real, SCIPallocBufferArray, SCIPfreeBufferArray, SCIPgetNLPCols(), SCIProwGetNNonz(), scoring(), and selectBestCut().

Referenced by SCIP_DECL_CUTSELSELECT().

SCIPselectCutsHybrid()

SCIP_RETCODE SCIPselectCutsHybrid	(	SCIP *	scip,
		SCIP_ROW **	cuts,
		SCIP_ROW **	forcedcuts,
		SCIP_RANDNUMGEN *	randnumgen,
		SCIP_Real	goodscorefac,
		SCIP_Real	badscorefac,
		SCIP_Real	goodmaxparall,
		SCIP_Real	maxparall,
		SCIP_Real	dircutoffdistweight,
		SCIP_Real	efficacyweight,
		SCIP_Real	objparalweight,
		SCIP_Real	intsupportweight,
		int	ncuts,
		int	nforcedcuts,
		int	maxselectedcuts,
		int *	nselectedcuts
	)

perform a cut selection algorithm for the given array of cuts

This is the selection method of the hybrid cut selector which uses a weighted sum of the efficacy, parallelism, directed cutoff distance, and the integral support. The input cuts array gets resorted s.t the selected cuts come first and the remaining ones are the end.

Parameters

scip	SCIP data structure
cuts	array with cuts to perform selection algorithm
forcedcuts	array with forced cuts
randnumgen	random number generator for tie-breaking, or NULL
goodscorefac	factor of best score among the given cuts to consider a cut good and filter with less strict settings of the maximum parallelism
badscorefac	factor of best score among the given cuts to consider a cut bad and discard it regardless of its parallelism to other cuts
goodmaxparall	maximum parallelism for good cuts
maxparall	maximum parallelism for non-good cuts
dircutoffdistweight	weight of directed cutoff distance in cut score calculation
efficacyweight	weight of efficacy in cut score calculation
objparalweight	weight of objective parallelism in cut score calculation
intsupportweight	weight of integral support in cut score calculation
ncuts	number of cuts in cuts array
nforcedcuts	number of forced cuts
maxselectedcuts	maximal number of cuts from cuts array to select
nselectedcuts	pointer to return number of selected cuts from cuts array

Definition at line 438 of file cutsel_hybrid.c.

References filterWithParallelism(), MAX, NULL, SCIP_CALL, SCIP_OKAY, SCIP_Real, SCIPallocBufferArray, SCIPfreeBufferArray, scoring(), and selectBestCut().

Referenced by doSeparation(), SCIP_DECL_CUTSELSELECT(), SCIP_DECL_SEPAEXECLP(), and separateRltCuts().