limp-cbc-0.3.2.0: cbits/coin/ClpSimplex.cpp
/* $Id: ClpSimplex.cpp 1989 2013-11-11 17:27:32Z forrest $ */
// Copyright (C) 2002, International Business Machines
// Corporation and others. All Rights Reserved.
// This code is licensed under the terms of the Eclipse Public License (EPL).
//#undef NDEBUG
#include "ClpConfig.h"
#include "CoinPragma.hpp"
#include <math.h>
#if SLIM_CLP==2
#define SLIM_NOIO
#endif
#include "CoinHelperFunctions.hpp"
#include "CoinFloatEqual.hpp"
#include "ClpSimplex.hpp"
#include "ClpFactorization.hpp"
#include "ClpPackedMatrix.hpp"
#include "CoinIndexedVector.hpp"
#include "ClpDualRowDantzig.hpp"
#include "ClpDualRowSteepest.hpp"
#include "ClpPrimalColumnDantzig.hpp"
#include "ClpPrimalColumnSteepest.hpp"
#include "ClpNonLinearCost.hpp"
#include "ClpMessage.hpp"
#include "ClpEventHandler.hpp"
#include "ClpLinearObjective.hpp"
#include "ClpHelperFunctions.hpp"
#include "CoinModel.hpp"
#include "CoinLpIO.hpp"
#include <cfloat>
#if CLP_HAS_ABC
#include "CoinAbcCommon.hpp"
#endif
#include <string>
#include <stdio.h>
#include <iostream>
//#############################################################################
ClpSimplex::ClpSimplex (bool emptyMessages) :
ClpModel(emptyMessages),
bestPossibleImprovement_(0.0),
zeroTolerance_(1.0e-13),
columnPrimalSequence_(-2),
rowPrimalSequence_(-2),
bestObjectiveValue_(-COIN_DBL_MAX),
moreSpecialOptions_(2),
baseIteration_(0),
primalToleranceToGetOptimal_(-1.0),
largeValue_(1.0e15),
largestPrimalError_(0.0),
largestDualError_(0.0),
alphaAccuracy_(-1.0),
dualBound_(1.0e10),
alpha_(0.0),
theta_(0.0),
lowerIn_(0.0),
valueIn_(0.0),
upperIn_(-COIN_DBL_MAX),
dualIn_(0.0),
lowerOut_(-1),
valueOut_(-1),
upperOut_(-1),
dualOut_(-1),
dualTolerance_(1.0e-7),
primalTolerance_(1.0e-7),
sumDualInfeasibilities_(0.0),
sumPrimalInfeasibilities_(0.0),
infeasibilityCost_(1.0e10),
sumOfRelaxedDualInfeasibilities_(0.0),
sumOfRelaxedPrimalInfeasibilities_(0.0),
acceptablePivot_(1.0e-8),
lower_(NULL),
rowLowerWork_(NULL),
columnLowerWork_(NULL),
upper_(NULL),
rowUpperWork_(NULL),
columnUpperWork_(NULL),
cost_(NULL),
rowObjectiveWork_(NULL),
objectiveWork_(NULL),
sequenceIn_(-1),
directionIn_(-1),
sequenceOut_(-1),
directionOut_(-1),
pivotRow_(-1),
lastGoodIteration_(-100),
dj_(NULL),
rowReducedCost_(NULL),
reducedCostWork_(NULL),
solution_(NULL),
rowActivityWork_(NULL),
columnActivityWork_(NULL),
numberDualInfeasibilities_(0),
numberDualInfeasibilitiesWithoutFree_(0),
numberPrimalInfeasibilities_(100),
numberRefinements_(0),
pivotVariable_(NULL),
factorization_(NULL),
savedSolution_(NULL),
numberTimesOptimal_(0),
disasterArea_(NULL),
changeMade_(1),
algorithm_(0),
forceFactorization_(-1),
perturbation_(100),
nonLinearCost_(NULL),
lastBadIteration_(-999999),
lastFlaggedIteration_(-999999),
numberFake_(0),
numberChanged_(0),
progressFlag_(0),
firstFree_(-1),
numberExtraRows_(0),
maximumBasic_(0),
dontFactorizePivots_(0),
incomingInfeasibility_(1.0),
allowedInfeasibility_(10.0),
automaticScale_(0),
maximumPerturbationSize_(0),
perturbationArray_(NULL),
baseModel_(NULL)
#ifdef ABC_INHERIT
,abcSimplex_(NULL),
abcState_(0)
#endif
{
int i;
for (i = 0; i < 6; i++) {
rowArray_[i] = NULL;
columnArray_[i] = NULL;
}
for (i = 0; i < 4; i++) {
spareIntArray_[i] = 0;
spareDoubleArray_[i] = 0.0;
}
saveStatus_ = NULL;
// get an empty factorization so we can set tolerances etc
getEmptyFactorization();
// Say sparse
factorization_->sparseThreshold(1);
// say Steepest pricing
dualRowPivot_ = new ClpDualRowSteepest();
// say Steepest pricing
primalColumnPivot_ = new ClpPrimalColumnSteepest();
solveType_ = 1; // say simplex based life form
}
// Subproblem constructor
ClpSimplex::ClpSimplex ( const ClpModel * rhs,
int numberRows, const int * whichRow,
int numberColumns, const int * whichColumn,
bool dropNames, bool dropIntegers, bool fixOthers)
: ClpModel(rhs, numberRows, whichRow,
numberColumns, whichColumn, dropNames, dropIntegers),
bestPossibleImprovement_(0.0),
zeroTolerance_(1.0e-13),
columnPrimalSequence_(-2),
rowPrimalSequence_(-2),
bestObjectiveValue_(-COIN_DBL_MAX),
moreSpecialOptions_(2),
baseIteration_(0),
primalToleranceToGetOptimal_(-1.0),
largeValue_(1.0e15),
largestPrimalError_(0.0),
largestDualError_(0.0),
alphaAccuracy_(-1.0),
dualBound_(1.0e10),
alpha_(0.0),
theta_(0.0),
lowerIn_(0.0),
valueIn_(0.0),
upperIn_(-COIN_DBL_MAX),
dualIn_(0.0),
lowerOut_(-1),
valueOut_(-1),
upperOut_(-1),
dualOut_(-1),
dualTolerance_(1.0e-7),
primalTolerance_(1.0e-7),
sumDualInfeasibilities_(0.0),
sumPrimalInfeasibilities_(0.0),
infeasibilityCost_(1.0e10),
sumOfRelaxedDualInfeasibilities_(0.0),
sumOfRelaxedPrimalInfeasibilities_(0.0),
acceptablePivot_(1.0e-8),
lower_(NULL),
rowLowerWork_(NULL),
columnLowerWork_(NULL),
upper_(NULL),
rowUpperWork_(NULL),
columnUpperWork_(NULL),
cost_(NULL),
rowObjectiveWork_(NULL),
objectiveWork_(NULL),
sequenceIn_(-1),
directionIn_(-1),
sequenceOut_(-1),
directionOut_(-1),
pivotRow_(-1),
lastGoodIteration_(-100),
dj_(NULL),
rowReducedCost_(NULL),
reducedCostWork_(NULL),
solution_(NULL),
rowActivityWork_(NULL),
columnActivityWork_(NULL),
numberDualInfeasibilities_(0),
numberDualInfeasibilitiesWithoutFree_(0),
numberPrimalInfeasibilities_(100),
numberRefinements_(0),
pivotVariable_(NULL),
factorization_(NULL),
savedSolution_(NULL),
numberTimesOptimal_(0),
disasterArea_(NULL),
changeMade_(1),
algorithm_(0),
forceFactorization_(-1),
perturbation_(100),
nonLinearCost_(NULL),
lastBadIteration_(-999999),
lastFlaggedIteration_(-999999),
numberFake_(0),
numberChanged_(0),
progressFlag_(0),
firstFree_(-1),
numberExtraRows_(0),
maximumBasic_(0),
dontFactorizePivots_(0),
incomingInfeasibility_(1.0),
allowedInfeasibility_(10.0),
automaticScale_(0),
maximumPerturbationSize_(0),
perturbationArray_(NULL),
baseModel_(NULL)
#ifdef ABC_INHERIT
,abcSimplex_(NULL),
abcState_(0)
#endif
{
int i;
for (i = 0; i < 6; i++) {
rowArray_[i] = NULL;
columnArray_[i] = NULL;
}
for (i = 0; i < 4; i++) {
spareIntArray_[i] = 0;
spareDoubleArray_[i] = 0.0;
}
saveStatus_ = NULL;
// get an empty factorization so we can set tolerances etc
getEmptyFactorization();
// say Steepest pricing
dualRowPivot_ = new ClpDualRowSteepest();
// say Steepest pricing
primalColumnPivot_ = new ClpPrimalColumnSteepest();
solveType_ = 1; // say simplex based life form
if (fixOthers) {
int numberOtherColumns = rhs->numberColumns();
int numberOtherRows = rhs->numberRows();
double * solution = new double [numberOtherColumns];
CoinZeroN(solution, numberOtherColumns);
int i;
for (i = 0; i < numberColumns; i++) {
int iColumn = whichColumn[i];
if (solution[iColumn])
fixOthers = false; // duplicates
solution[iColumn] = 1.0;
}
if (fixOthers) {
const double * otherSolution = rhs->primalColumnSolution();
const double * objective = rhs->objective();
double offset = 0.0;
for (i = 0; i < numberOtherColumns; i++) {
if (solution[i]) {
solution[i] = 0.0; // in
} else {
solution[i] = otherSolution[i];
offset += objective[i] * otherSolution[i];
}
}
double * rhsModification = new double [numberOtherRows];
CoinZeroN(rhsModification, numberOtherRows);
rhs->matrix()->times(solution, rhsModification) ;
for ( i = 0; i < numberRows; i++) {
int iRow = whichRow[i];
if (rowLower_[i] > -1.0e20)
rowLower_[i] -= rhsModification[iRow];
if (rowUpper_[i] < 1.0e20)
rowUpper_[i] -= rhsModification[iRow];
}
delete [] rhsModification;
setObjectiveOffset(rhs->objectiveOffset() - offset);
// And set objective value to match
setObjectiveValue(rhs->objectiveValue());
}
delete [] solution;
}
}
// Subproblem constructor
ClpSimplex::ClpSimplex ( const ClpSimplex * rhs,
int numberRows, const int * whichRow,
int numberColumns, const int * whichColumn,
bool dropNames, bool dropIntegers, bool fixOthers)
: ClpModel(rhs, numberRows, whichRow,
numberColumns, whichColumn, dropNames, dropIntegers),
bestPossibleImprovement_(0.0),
zeroTolerance_(1.0e-13),
columnPrimalSequence_(-2),
rowPrimalSequence_(-2),
bestObjectiveValue_(-COIN_DBL_MAX),
moreSpecialOptions_(2),
baseIteration_(0),
primalToleranceToGetOptimal_(-1.0),
largeValue_(1.0e15),
largestPrimalError_(0.0),
largestDualError_(0.0),
alphaAccuracy_(-1.0),
dualBound_(1.0e10),
alpha_(0.0),
theta_(0.0),
lowerIn_(0.0),
valueIn_(0.0),
upperIn_(-COIN_DBL_MAX),
dualIn_(0.0),
lowerOut_(-1),
valueOut_(-1),
upperOut_(-1),
dualOut_(-1),
dualTolerance_(rhs->dualTolerance_),
primalTolerance_(rhs->primalTolerance_),
sumDualInfeasibilities_(0.0),
sumPrimalInfeasibilities_(0.0),
infeasibilityCost_(1.0e10),
sumOfRelaxedDualInfeasibilities_(0.0),
sumOfRelaxedPrimalInfeasibilities_(0.0),
acceptablePivot_(1.0e-8),
lower_(NULL),
rowLowerWork_(NULL),
columnLowerWork_(NULL),
upper_(NULL),
rowUpperWork_(NULL),
columnUpperWork_(NULL),
cost_(NULL),
rowObjectiveWork_(NULL),
objectiveWork_(NULL),
sequenceIn_(-1),
directionIn_(-1),
sequenceOut_(-1),
directionOut_(-1),
pivotRow_(-1),
lastGoodIteration_(-100),
dj_(NULL),
rowReducedCost_(NULL),
reducedCostWork_(NULL),
solution_(NULL),
rowActivityWork_(NULL),
columnActivityWork_(NULL),
numberDualInfeasibilities_(0),
numberDualInfeasibilitiesWithoutFree_(0),
numberPrimalInfeasibilities_(100),
numberRefinements_(0),
pivotVariable_(NULL),
factorization_(NULL),
savedSolution_(NULL),
numberTimesOptimal_(0),
disasterArea_(NULL),
changeMade_(1),
algorithm_(0),
forceFactorization_(-1),
perturbation_(100),
nonLinearCost_(NULL),
lastBadIteration_(-999999),
lastFlaggedIteration_(-999999),
numberFake_(0),
numberChanged_(0),
progressFlag_(0),
firstFree_(-1),
numberExtraRows_(0),
maximumBasic_(0),
dontFactorizePivots_(0),
incomingInfeasibility_(1.0),
allowedInfeasibility_(10.0),
automaticScale_(0),
maximumPerturbationSize_(0),
perturbationArray_(NULL),
baseModel_(NULL)
#ifdef ABC_INHERIT
,abcSimplex_(NULL),
abcState_(rhs->abcState_)
#endif
{
int i;
for (i = 0; i < 6; i++) {
rowArray_[i] = NULL;
columnArray_[i] = NULL;
}
for (i = 0; i < 4; i++) {
spareIntArray_[i] = 0;
spareDoubleArray_[i] = 0.0;
}
saveStatus_ = NULL;
factorization_ = new ClpFactorization(*rhs->factorization_, -numberRows_);
//factorization_ = new ClpFactorization(*rhs->factorization_,
// rhs->factorization_->goDenseThreshold());
ClpDualRowDantzig * pivot =
dynamic_cast< ClpDualRowDantzig*>(rhs->dualRowPivot_);
// say Steepest pricing
if (!pivot)
dualRowPivot_ = new ClpDualRowSteepest();
else
dualRowPivot_ = new ClpDualRowDantzig();
// say Steepest pricing
primalColumnPivot_ = new ClpPrimalColumnSteepest();
solveType_ = 1; // say simplex based life form
if (fixOthers) {
int numberOtherColumns = rhs->numberColumns();
int numberOtherRows = rhs->numberRows();
double * solution = new double [numberOtherColumns];
CoinZeroN(solution, numberOtherColumns);
int i;
for (i = 0; i < numberColumns; i++) {
int iColumn = whichColumn[i];
if (solution[iColumn])
fixOthers = false; // duplicates
solution[iColumn] = 1.0;
}
if (fixOthers) {
const double * otherSolution = rhs->primalColumnSolution();
const double * objective = rhs->objective();
double offset = 0.0;
for (i = 0; i < numberOtherColumns; i++) {
if (solution[i]) {
solution[i] = 0.0; // in
} else {
solution[i] = otherSolution[i];
offset += objective[i] * otherSolution[i];
}
}
double * rhsModification = new double [numberOtherRows];
CoinZeroN(rhsModification, numberOtherRows);
rhs->matrix()->times(solution, rhsModification) ;
for ( i = 0; i < numberRows; i++) {
int iRow = whichRow[i];
if (rowLower_[i] > -1.0e20)
rowLower_[i] -= rhsModification[iRow];
if (rowUpper_[i] < 1.0e20)
rowUpper_[i] -= rhsModification[iRow];
}
delete [] rhsModification;
setObjectiveOffset(rhs->objectiveOffset() - offset);
// And set objective value to match
setObjectiveValue(rhs->objectiveValue());
}
delete [] solution;
}
if (rhs->maximumPerturbationSize_) {
maximumPerturbationSize_ = 2 * numberColumns;
perturbationArray_ = new double [maximumPerturbationSize_];
for (i = 0; i < numberColumns; i++) {
int iColumn = whichColumn[i];
perturbationArray_[2*i] = rhs->perturbationArray_[2*iColumn];
perturbationArray_[2*i+1] = rhs->perturbationArray_[2*iColumn+1];
}
}
}
// Puts solution back small model
void
ClpSimplex::getbackSolution(const ClpSimplex & smallModel, const int * whichRow, const int * whichColumn)
{
setSumDualInfeasibilities(smallModel.sumDualInfeasibilities());
setNumberDualInfeasibilities(smallModel.numberDualInfeasibilities());
setSumPrimalInfeasibilities(smallModel.sumPrimalInfeasibilities());
setNumberPrimalInfeasibilities(smallModel.numberPrimalInfeasibilities());
setNumberIterations(smallModel.numberIterations());
setProblemStatus(smallModel.status());
setObjectiveValue(smallModel.objectiveValue());
const double * solution2 = smallModel.primalColumnSolution();
int i;
int numberRows2 = smallModel.numberRows();
int numberColumns2 = smallModel.numberColumns();
const double * dj2 = smallModel.dualColumnSolution();
for ( i = 0; i < numberColumns2; i++) {
int iColumn = whichColumn[i];
columnActivity_[iColumn] = solution2[i];
reducedCost_[iColumn] = dj2[i];
setStatus(iColumn, smallModel.getStatus(i));
}
const double * dual2 = smallModel.dualRowSolution();
memset(dual_, 0, numberRows_ * sizeof(double));
for (i = 0; i < numberRows2; i++) {
int iRow = whichRow[i];
setRowStatus(iRow, smallModel.getRowStatus(i));
dual_[iRow] = dual2[i];
}
CoinZeroN(rowActivity_, numberRows_);
#if 0
if (!problemStatus_) {
ClpDisjointCopyN(smallModel.objective(), smallModel.numberColumns_, smallModel.reducedCost_);
smallModel.matrix_->transposeTimes(-1.0, smallModel.dual_, smallModel.reducedCost_);
for (int i = 0; i < smallModel.numberColumns_; i++) {
if (smallModel.getColumnStatus(i) == basic)
assert (fabs(smallModel.reducedCost_[i]) < 1.0e-5);
}
ClpDisjointCopyN(objective(), numberColumns_, reducedCost_);
matrix_->transposeTimes(-1.0, dual_, reducedCost_);
for (int i = 0; i < numberColumns_; i++) {
if (getColumnStatus(i) == basic)
assert (fabs(reducedCost_[i]) < 1.0e-5);
}
}
#endif
matrix()->times(columnActivity_, rowActivity_) ;
}
//-----------------------------------------------------------------------------
ClpSimplex::~ClpSimplex ()
{
setPersistenceFlag(0);
gutsOfDelete(0);
delete nonLinearCost_;
}
//#############################################################################
void ClpSimplex::setLargeValue( double value)
{
if (value > 0.0 && value < COIN_DBL_MAX)
largeValue_ = value;
}
int
ClpSimplex::gutsOfSolution ( double * givenDuals,
const double * givenPrimals,
bool valuesPass)
{
// if values pass, save values of basic variables
double * save = NULL;
double oldValue = 0.0;
if (valuesPass) {
assert(algorithm_ > 0); // only primal at present
assert(nonLinearCost_);
int iRow;
checkPrimalSolution( rowActivityWork_, columnActivityWork_);
// get correct bounds on all variables
nonLinearCost_->checkInfeasibilities(primalTolerance_);
oldValue = nonLinearCost_->largestInfeasibility();
save = new double[numberRows_];
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
save[iRow] = solution_[iPivot];
}
}
// do work
computePrimals(rowActivityWork_, columnActivityWork_);
// If necessary - override results
if (givenPrimals) {
CoinMemcpyN(givenPrimals, numberColumns_, columnActivityWork_);
memset(rowActivityWork_, 0, numberRows_ * sizeof(double));
times(-1.0, columnActivityWork_, rowActivityWork_);
}
double objectiveModification = 0.0;
if (algorithm_ > 0 && nonLinearCost_ != NULL) {
// primal algorithm
// get correct bounds on all variables
// If 4 bit set - Force outgoing variables to exact bound (primal)
if ((specialOptions_ & 4) == 0)
nonLinearCost_->checkInfeasibilities(primalTolerance_);
else
nonLinearCost_->checkInfeasibilities(0.0);
objectiveModification += nonLinearCost_->changeInCost();
if (nonLinearCost_->numberInfeasibilities())
if (handler_->detail(CLP_SIMPLEX_NONLINEAR, messages_) < 100) {
handler_->message(CLP_SIMPLEX_NONLINEAR, messages_)
<< nonLinearCost_->changeInCost()
<< nonLinearCost_->numberInfeasibilities()
<< CoinMessageEol;
}
}
if (valuesPass) {
double badInfeasibility = nonLinearCost_->largestInfeasibility();
#ifdef CLP_DEBUG
std::cout << "Largest given infeasibility " << oldValue
<< " now " << nonLinearCost_->largestInfeasibility() << std::endl;
#endif
int numberOut = 0;
// But may be very large rhs etc
double useError = CoinMin(largestPrimalError_,
1.0e5 / maximumAbsElement(solution_, numberRows_ + numberColumns_));
if ((oldValue < incomingInfeasibility_ || badInfeasibility >
(CoinMax(10.0 * allowedInfeasibility_, 100.0 * oldValue)))
&& (badInfeasibility > CoinMax(incomingInfeasibility_, allowedInfeasibility_) ||
useError > 1.0e-3)) {
if (algorithm_>1) {
// nonlinear
//printf("Original largest infeas %g, now %g, primalError %g\n",
// oldValue,nonLinearCost_->largestInfeasibility(),
// largestPrimalError_);
//printf("going to all slack\n");
allSlackBasis(true);
CoinIotaN(pivotVariable_, numberRows_, numberColumns_);
return 1;
}
//printf("Original largest infeas %g, now %g, primalError %g\n",
// oldValue,nonLinearCost_->largestInfeasibility(),
// largestPrimalError_);
// throw out up to 1000 structurals
int maxOut = (allowedInfeasibility_==10.0) ? 1000 : 100;
int iRow;
int * sort = new int[numberRows_];
// first put back solution and store difference
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
double difference = fabs(solution_[iPivot] - save[iRow]);
solution_[iPivot] = save[iRow];
save[iRow] = difference;
}
int numberBasic = 0;
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
if (iPivot < numberColumns_) {
// column
double difference = save[iRow];
if (difference > 1.0e-4) {
sort[numberOut] = iRow;
save[numberOut++] = -difference;
if (getStatus(iPivot) == basic)
numberBasic++;
}
}
}
if (!numberBasic) {
//printf("no errors on basic - going to all slack - numberOut %d\n",numberOut);
#if 0
allSlackBasis(true);
CoinIotaN(pivotVariable_, numberRows_, numberColumns_);
#else
// allow
numberOut = 0;
#endif
}
CoinSort_2(save, save + numberOut, sort);
numberOut = CoinMin(maxOut, numberOut);
for (iRow = 0; iRow < numberOut; iRow++) {
int jRow = sort[iRow];
int iColumn = pivotVariable_[jRow];
setColumnStatus(iColumn, superBasic);
setRowStatus(jRow, basic);
pivotVariable_[jRow] = jRow + numberColumns_;
if (fabs(solution_[iColumn]) > 1.0e10) {
if (upper_[iColumn] < 0.0) {
solution_[iColumn] = upper_[iColumn];
} else if (lower_[iColumn] > 0.0) {
solution_[iColumn] = lower_[iColumn];
} else {
solution_[iColumn] = 0.0;
}
}
}
delete [] sort;
}
delete [] save;
if (numberOut)
return numberOut;
}
if ((moreSpecialOptions_ & 128) != 0 && !numberIterations_) {
//printf("trying feas pump\n");
const char * integerType = integerInformation();
assert (integerType);
assert (perturbationArray_);
CoinZeroN(cost_, numberRows_ + numberColumns_);
for (int i = 0; i < numberRows_ - numberRows_; i++) {
int iSequence = pivotVariable_[i];
if (iSequence < numberColumns_ && integerType[iSequence]) {
double lower = lower_[iSequence];
double upper = upper_[iSequence];
double value = solution_[iSequence];
if (value >= lower - primalTolerance_ &&
value <= upper + primalTolerance_) {
double sign;
if (value - lower < upper - value)
sign = 1.0;
else
sign = -1.0;
cost_[iSequence] = sign * perturbationArray_[iSequence];
}
}
}
}
#if CAN_HAVE_ZERO_OBJ>1
if ((specialOptions_&2097152)==0) {
#endif
computeDuals(givenDuals);
if ((moreSpecialOptions_ & 128) != 0 && !numberIterations_) {
const char * integerType = integerInformation();
// Need to do columns and rows to stay dual feasible
for (int iSequence = 0; iSequence < numberColumns_; iSequence++) {
if (integerType[iSequence] && getStatus(iSequence) != basic) {
double djValue = dj_[iSequence];
double change = 0.0;
if (getStatus(iSequence) == atLowerBound)
change = CoinMax(-djValue, 10.0 * perturbationArray_[iSequence]);
else if (getStatus(iSequence) == atUpperBound)
change = CoinMin(-djValue, -10.0 * perturbationArray_[iSequence]);
cost_[iSequence] = change;
dj_[iSequence] += change;
}
}
}
// now check solutions
//checkPrimalSolution( rowActivityWork_, columnActivityWork_);
//checkDualSolution();
checkBothSolutions();
objectiveValue_ += objectiveModification / (objectiveScale_ * rhsScale_);
#if CAN_HAVE_ZERO_OBJ>1
} else {
checkPrimalSolution( rowActivityWork_, columnActivityWork_);
#ifndef COIN_REUSE_RANDOM
memset(dj_,0,(numberRows_+numberColumns_)*sizeof(double));
#else
for (int iSequence=0;iSequence<numberRows_+numberColumns_;iSequence++) {
double value;
switch (getStatus(iSequence)) {
case atLowerBound:
value=1.0e-9*(1.0+CoinDrand48());
break;
case atUpperBound:
value=-1.0e-9*(1.0+CoinDrand48());
break;
default:
value=0.0;
break;
}
}
#endif
objectiveValue_=0.0;
}
#endif
if (handler_->logLevel() > 3 || (largestPrimalError_ > 1.0e-2 ||
largestDualError_ > 1.0e-2))
handler_->message(CLP_SIMPLEX_ACCURACY, messages_)
<< largestPrimalError_
<< largestDualError_
<< CoinMessageEol;
if (largestPrimalError_ > 1.0e-1 && numberRows_ > 100 && numberIterations_) {
// Change factorization tolerance
if (factorization_->zeroTolerance() > 1.0e-18)
factorization_->zeroTolerance(1.0e-18);
}
// Switch off false values pass indicator
if (!valuesPass && algorithm_ > 0)
firstFree_ = -1;
return 0;
}
void
ClpSimplex::computePrimals ( const double * rowActivities,
const double * columnActivities)
{
//work space
CoinIndexedVector * workSpace = rowArray_[0];
CoinIndexedVector * arrayVector = rowArray_[1];
arrayVector->clear();
CoinIndexedVector * previousVector = rowArray_[2];
previousVector->clear();
// accumulate non basic stuff
int iRow;
// order is this way for scaling
if (columnActivities != columnActivityWork_)
ClpDisjointCopyN(columnActivities, numberColumns_, columnActivityWork_);
if (rowActivities != rowActivityWork_)
ClpDisjointCopyN(rowActivities, numberRows_, rowActivityWork_);
double * array = arrayVector->denseVector();
int * index = arrayVector->getIndices();
int number = 0;
const double * rhsOffset = matrix_->rhsOffset(this, false, true);
if (!rhsOffset) {
// Use whole matrix every time to make it easier for ClpMatrixBase
// So zero out basic
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
assert (iPivot >= 0);
solution_[iPivot] = 0.0;
#ifdef CLP_INVESTIGATE
assert (getStatus(iPivot) == basic);
#endif
}
// Extended solution before "update"
matrix_->primalExpanded(this, 0);
times(-1.0, columnActivityWork_, array);
for (iRow = 0; iRow < numberRows_; iRow++) {
double value = array[iRow] + rowActivityWork_[iRow];
if (value) {
array[iRow] = value;
index[number++] = iRow;
} else {
array[iRow] = 0.0;
}
}
} else {
// we have an effective rhs lying around
// zero out basic (really just for slacks)
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
solution_[iPivot] = 0.0;
}
for (iRow = 0; iRow < numberRows_; iRow++) {
double value = rhsOffset[iRow] + rowActivityWork_[iRow];
if (value) {
array[iRow] = value;
index[number++] = iRow;
} else {
array[iRow] = 0.0;
}
}
}
arrayVector->setNumElements(number);
#ifdef CLP_DEBUG
if (numberIterations_ == -3840) {
int i;
for (i = 0; i < numberRows_ + numberColumns_; i++)
printf("%d status %d\n", i, status_[i]);
printf("xxxxx1\n");
for (i = 0; i < numberRows_; i++)
if (array[i])
printf("%d rhs %g\n", i, array[i]);
printf("xxxxx2\n");
for (i = 0; i < numberRows_ + numberColumns_; i++)
if (getStatus(i) != basic)
printf("%d non basic %g %g %g\n", i, lower_[i], solution_[i], upper_[i]);
printf("xxxxx3\n");
}
#endif
// Ftran adjusted RHS and iterate to improve accuracy
double lastError = COIN_DBL_MAX;
int iRefine;
CoinIndexedVector * thisVector = arrayVector;
CoinIndexedVector * lastVector = previousVector;
if (number)
factorization_->updateColumn(workSpace, thisVector);
double * work = workSpace->denseVector();
#ifdef CLP_DEBUG
if (numberIterations_ == -3840) {
int i;
for (i = 0; i < numberRows_; i++)
if (array[i])
printf("%d after rhs %g\n", i, array[i]);
printf("xxxxx4\n");
}
#endif
bool goodSolution = true;
for (iRefine = 0; iRefine < numberRefinements_ + 1; iRefine++) {
int numberIn = thisVector->getNumElements();
int * indexIn = thisVector->getIndices();
double * arrayIn = thisVector->denseVector();
// put solution in correct place
if (!rhsOffset) {
int j;
for (j = 0; j < numberIn; j++) {
iRow = indexIn[j];
int iPivot = pivotVariable_[iRow];
solution_[iPivot] = arrayIn[iRow];
//assert (fabs(solution_[iPivot])<1.0e100);
}
} else {
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
solution_[iPivot] = arrayIn[iRow];
//assert (fabs(solution_[iPivot])<1.0e100);
}
}
// Extended solution after "update"
matrix_->primalExpanded(this, 1);
// check Ax == b (for all)
// signal column generated matrix to just do basic (and gub)
unsigned int saveOptions = specialOptions();
setSpecialOptions(16);
times(-1.0, columnActivityWork_, work);
setSpecialOptions(saveOptions);
largestPrimalError_ = 0.0;
double multiplier = 131072.0;
for (iRow = 0; iRow < numberRows_; iRow++) {
double value = work[iRow] + rowActivityWork_[iRow];
work[iRow] = value * multiplier;
if (fabs(value) > largestPrimalError_) {
largestPrimalError_ = fabs(value);
}
}
if (largestPrimalError_ >= lastError) {
// restore
CoinIndexedVector * temp = thisVector;
thisVector = lastVector;
lastVector = temp;
goodSolution = false;
break;
}
if (iRefine < numberRefinements_ && largestPrimalError_ > 1.0e-10) {
// try and make better
// save this
CoinIndexedVector * temp = thisVector;
thisVector = lastVector;
lastVector = temp;
int * indexOut = thisVector->getIndices();
int number = 0;
array = thisVector->denseVector();
thisVector->clear();
for (iRow = 0; iRow < numberRows_; iRow++) {
double value = work[iRow];
if (value) {
array[iRow] = value;
indexOut[number++] = iRow;
work[iRow] = 0.0;
}
}
thisVector->setNumElements(number);
lastError = largestPrimalError_;
factorization_->updateColumn(workSpace, thisVector);
multiplier = 1.0 / multiplier;
double * previous = lastVector->denseVector();
number = 0;
for (iRow = 0; iRow < numberRows_; iRow++) {
double value = previous[iRow] + multiplier * array[iRow];
if (value) {
array[iRow] = value;
indexOut[number++] = iRow;
} else {
array[iRow] = 0.0;
}
}
thisVector->setNumElements(number);
} else {
break;
}
}
// solution as accurate as we are going to get
ClpFillN(work, numberRows_, 0.0);
if (!goodSolution) {
array = thisVector->denseVector();
// put solution in correct place
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
solution_[iPivot] = array[iRow];
//assert (fabs(solution_[iPivot])<1.0e100);
}
}
arrayVector->clear();
previousVector->clear();
#ifdef CLP_DEBUG
if (numberIterations_ == -3840) {
exit(77);
}
#endif
}
// now dual side
void
ClpSimplex::computeDuals(double * givenDjs)
{
#ifndef SLIM_CLP
if (objective_->type() == 1 || !objective_->activated()) {
#endif
// Linear
//work space
CoinIndexedVector * workSpace = rowArray_[0];
CoinIndexedVector * arrayVector = rowArray_[1];
arrayVector->clear();
CoinIndexedVector * previousVector = rowArray_[2];
previousVector->clear();
int iRow;
#ifdef CLP_DEBUG
workSpace->checkClear();
#endif
double * array = arrayVector->denseVector();
int * index = arrayVector->getIndices();
int number = 0;
if (!givenDjs) {
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
double value = cost_[iPivot];
if (value) {
array[iRow] = value;
index[number++] = iRow;
}
}
} else {
// dual values pass - djs may not be zero
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
// make sure zero if done
if (!pivoted(iPivot))
givenDjs[iPivot] = 0.0;
double value = cost_[iPivot] - givenDjs[iPivot];
if (value) {
array[iRow] = value;
index[number++] = iRow;
}
}
}
arrayVector->setNumElements(number);
// Extended duals before "updateTranspose"
matrix_->dualExpanded(this, arrayVector, givenDjs, 0);
// Btran basic costs and get as accurate as possible
double lastError = COIN_DBL_MAX;
int iRefine;
double * work = workSpace->denseVector();
CoinIndexedVector * thisVector = arrayVector;
CoinIndexedVector * lastVector = previousVector;
factorization_->updateColumnTranspose(workSpace, thisVector);
for (iRefine = 0; iRefine < numberRefinements_ + 1; iRefine++) {
// check basic reduced costs zero
largestDualError_ = 0.0;
if (!numberExtraRows_) {
// Just basic
int * index2 = workSpace->getIndices();
// use reduced costs for slacks as work array
double * work2 = reducedCostWork_ + numberColumns_;
int numberStructurals = 0;
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
if (iPivot < numberColumns_)
index2[numberStructurals++] = iPivot;
}
matrix_->listTransposeTimes(this, array, index2, numberStructurals, work2);
numberStructurals = 0;
if (!givenDjs) {
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
double value;
if (iPivot >= numberColumns_) {
// slack
value = rowObjectiveWork_[iPivot-numberColumns_]
+ array[iPivot-numberColumns_];
} else {
// column
value = objectiveWork_[iPivot] - work2[numberStructurals++];
}
work[iRow] = value;
if (fabs(value) > largestDualError_) {
largestDualError_ = fabs(value);
}
}
} else {
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
if (iPivot >= numberColumns_) {
// slack
work[iRow] = rowObjectiveWork_[iPivot-numberColumns_]
+ array[iPivot-numberColumns_] - givenDjs[iPivot];
} else {
// column
work[iRow] = objectiveWork_[iPivot] - work2[numberStructurals++]
- givenDjs[iPivot];
}
if (fabs(work[iRow]) > largestDualError_) {
largestDualError_ = fabs(work[iRow]);
//assert (largestDualError_<1.0e-7);
//if (largestDualError_>1.0e-7)
//printf("large dual error %g\n",largestDualError_);
}
}
}
} else {
// extra rows - be more careful
#if 1
// would be faster to do just for basic but this reduces code
ClpDisjointCopyN(objectiveWork_, numberColumns_, reducedCostWork_);
transposeTimes(-1.0, array, reducedCostWork_);
#else
// Just basic
int * index2 = workSpace->getIndices();
int numberStructurals = 0;
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
if (iPivot < numberColumns_)
index2[numberStructurals++] = iPivot;
}
matrix_->listTransposeTimes(this, array, index2, numberStructurals, work);
for (iRow = 0; iRow < numberStructurals; iRow++) {
int iPivot = index2[iRow];
reducedCostWork_[iPivot] = objectiveWork_[iPivot] - work[iRow];
}
#endif
// update by duals on sets
matrix_->dualExpanded(this, NULL, NULL, 1);
if (!givenDjs) {
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
double value;
if (iPivot >= numberColumns_) {
// slack
value = rowObjectiveWork_[iPivot-numberColumns_]
+ array[iPivot-numberColumns_];
} else {
// column
value = reducedCostWork_[iPivot];
}
work[iRow] = value;
if (fabs(value) > largestDualError_) {
largestDualError_ = fabs(value);
}
}
} else {
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
if (iPivot >= numberColumns_) {
// slack
work[iRow] = rowObjectiveWork_[iPivot-numberColumns_]
+ array[iPivot-numberColumns_] - givenDjs[iPivot];
} else {
// column
work[iRow] = reducedCostWork_[iPivot] - givenDjs[iPivot];
}
if (fabs(work[iRow]) > largestDualError_) {
largestDualError_ = fabs(work[iRow]);
//assert (largestDualError_<1.0e-7);
//if (largestDualError_>1.0e-7)
//printf("large dual error %g\n",largestDualError_);
}
}
}
}
if (largestDualError_ >= lastError) {
// restore
CoinIndexedVector * temp = thisVector;
thisVector = lastVector;
lastVector = temp;
break;
}
if (iRefine < numberRefinements_ && largestDualError_ > 1.0e-10
&& !givenDjs) {
// try and make better
// save this
CoinIndexedVector * temp = thisVector;
thisVector = lastVector;
lastVector = temp;
int * indexOut = thisVector->getIndices();
int number = 0;
array = thisVector->denseVector();
thisVector->clear();
double multiplier = 131072.0;
for (iRow = 0; iRow < numberRows_; iRow++) {
double value = multiplier * work[iRow];
if (value) {
array[iRow] = value;
indexOut[number++] = iRow;
work[iRow] = 0.0;
}
work[iRow] = 0.0;
}
thisVector->setNumElements(number);
lastError = largestDualError_;
factorization_->updateColumnTranspose(workSpace, thisVector);
multiplier = 1.0 / multiplier;
double * previous = lastVector->denseVector();
number = 0;
for (iRow = 0; iRow < numberRows_; iRow++) {
double value = previous[iRow] + multiplier * array[iRow];
if (value) {
array[iRow] = value;
indexOut[number++] = iRow;
} else {
array[iRow] = 0.0;
}
}
thisVector->setNumElements(number);
} else {
break;
}
}
// now look at dual solution
array = thisVector->denseVector();
for (iRow = 0; iRow < numberRows_; iRow++) {
// slack
double value = array[iRow];
dual_[iRow] = value;
value += rowObjectiveWork_[iRow];
rowReducedCost_[iRow] = value;
}
// can use work if problem scaled (for better cache)
ClpPackedMatrix* clpMatrix =
dynamic_cast< ClpPackedMatrix*>(matrix_);
double * saveRowScale = rowScale_;
//double * saveColumnScale = columnScale_;
if (scaledMatrix_) {
rowScale_ = NULL;
clpMatrix = scaledMatrix_;
}
if (clpMatrix && (clpMatrix->flags() & 2) == 0) {
CoinIndexedVector * cVector = columnArray_[0];
int * whichColumn = cVector->getIndices();
assert (!cVector->getNumElements());
int n = 0;
for (int i = 0; i < numberColumns_; i++) {
if (getColumnStatus(i) != basic) {
whichColumn[n++] = i;
reducedCostWork_[i] = objectiveWork_[i];
} else {
reducedCostWork_[i] = 0.0;
}
}
if (numberRows_ > 4000)
clpMatrix->transposeTimesSubset(n, whichColumn, dual_, reducedCostWork_,
rowScale_, columnScale_, work);
else
clpMatrix->transposeTimesSubset(n, whichColumn, dual_, reducedCostWork_,
rowScale_, columnScale_, NULL);
} else {
ClpDisjointCopyN(objectiveWork_, numberColumns_, reducedCostWork_);
if (numberRows_ > 4000)
matrix_->transposeTimes(-1.0, dual_, reducedCostWork_,
rowScale_, columnScale_, work);
else
matrix_->transposeTimes(-1.0, dual_, reducedCostWork_,
rowScale_, columnScale_, NULL);
}
rowScale_ = saveRowScale;
//columnScale_ = saveColumnScale;
ClpFillN(work, numberRows_, 0.0);
// Extended duals and check dual infeasibility
if (!matrix_->skipDualCheck() || algorithm_ < 0 || problemStatus_ != -2)
matrix_->dualExpanded(this, NULL, NULL, 2);
// If necessary - override results
if (givenDjs) {
// restore accurate duals
CoinMemcpyN(dj_, (numberRows_ + numberColumns_), givenDjs);
}
arrayVector->clear();
previousVector->clear();
#ifndef SLIM_CLP
} else {
// Nonlinear
objective_->reducedGradient(this, dj_, false);
// get dual_ by moving from reduced costs for slacks
CoinMemcpyN(dj_ + numberColumns_, numberRows_, dual_);
}
#endif
}
/* Given an existing factorization computes and checks
primal and dual solutions. Uses input arrays for variables at
bounds. Returns feasibility states */
int ClpSimplex::getSolution ( const double * /*rowActivities*/,
const double * /*columnActivities*/)
{
if (!factorization_->status()) {
// put in standard form
createRim(7 + 8 + 16 + 32, false, -1);
if (pivotVariable_[0] < 0)
internalFactorize(0);
// do work
gutsOfSolution ( NULL, NULL);
// release extra memory
deleteRim(0);
}
return factorization_->status();
}
/* Given an existing factorization computes and checks
primal and dual solutions. Uses current problem arrays for
bounds. Returns feasibility states */
int ClpSimplex::getSolution ( )
{
double * rowActivities = new double[numberRows_];
double * columnActivities = new double[numberColumns_];
ClpDisjointCopyN ( rowActivityWork_, numberRows_ , rowActivities);
ClpDisjointCopyN ( columnActivityWork_, numberColumns_ , columnActivities);
int status = getSolution( rowActivities, columnActivities);
delete [] rowActivities;
delete [] columnActivities;
return status;
}
// Factorizes using current basis. This is for external use
// Return codes are as from ClpFactorization
int ClpSimplex::factorize ()
{
// put in standard form
createRim(7 + 8 + 16 + 32, false);
// do work
int status = internalFactorize(-1);
// release extra memory
deleteRim(0);
return status;
}
// Clean up status
void
ClpSimplex::cleanStatus()
{
int iRow, iColumn;
int numberBasic = 0;
// make row activities correct
memset(rowActivityWork_, 0, numberRows_ * sizeof(double));
times(1.0, columnActivityWork_, rowActivityWork_);
if (!status_)
createStatus();
for (iRow = 0; iRow < numberRows_; iRow++) {
if (getRowStatus(iRow) == basic)
numberBasic++;
else {
setRowStatus(iRow, superBasic);
// but put to bound if close
if (fabs(rowActivityWork_[iRow] - rowLowerWork_[iRow])
<= primalTolerance_) {
rowActivityWork_[iRow] = rowLowerWork_[iRow];
setRowStatus(iRow, atLowerBound);
} else if (fabs(rowActivityWork_[iRow] - rowUpperWork_[iRow])
<= primalTolerance_) {
rowActivityWork_[iRow] = rowUpperWork_[iRow];
setRowStatus(iRow, atUpperBound);
}
}
}
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (getColumnStatus(iColumn) == basic) {
if (numberBasic == numberRows_) {
// take out of basis
setColumnStatus(iColumn, superBasic);
// but put to bound if close
if (fabs(columnActivityWork_[iColumn] - columnLowerWork_[iColumn])
<= primalTolerance_) {
columnActivityWork_[iColumn] = columnLowerWork_[iColumn];
setColumnStatus(iColumn, atLowerBound);
} else if (fabs(columnActivityWork_[iColumn]
- columnUpperWork_[iColumn])
<= primalTolerance_) {
columnActivityWork_[iColumn] = columnUpperWork_[iColumn];
setColumnStatus(iColumn, atUpperBound);
}
} else
numberBasic++;
} else {
setColumnStatus(iColumn, superBasic);
// but put to bound if close
if (fabs(columnActivityWork_[iColumn] - columnLowerWork_[iColumn])
<= primalTolerance_) {
columnActivityWork_[iColumn] = columnLowerWork_[iColumn];
setColumnStatus(iColumn, atLowerBound);
} else if (fabs(columnActivityWork_[iColumn]
- columnUpperWork_[iColumn])
<= primalTolerance_) {
columnActivityWork_[iColumn] = columnUpperWork_[iColumn];
setColumnStatus(iColumn, atUpperBound);
}
}
}
}
/* Factorizes using current basis.
solveType - 1 iterating, 0 initial, -1 external
- 2 then iterating but can throw out of basis
If 10 added then in primal values pass
Return codes are as from ClpFactorization unless initial factorization
when total number of singularities is returned.
Special case is numberRows_+1 -> all slack basis.
*/
int ClpSimplex::internalFactorize ( int solveType)
{
int iRow, iColumn;
int totalSlacks = numberRows_;
if (!status_)
createStatus();
bool valuesPass = false;
if (solveType >= 10) {
valuesPass = true;
solveType -= 10;
}
#ifdef CLP_DEBUG
if (solveType > 0) {
int numberFreeIn = 0, numberFreeOut = 0;
double biggestDj = 0.0;
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
switch(getColumnStatus(iColumn)) {
case basic:
if (columnLower_[iColumn] < -largeValue_
&& columnUpper_[iColumn] > largeValue_)
numberFreeIn++;
break;
default:
if (columnLower_[iColumn] < -largeValue_
&& columnUpper_[iColumn] > largeValue_) {
numberFreeOut++;
biggestDj = CoinMax(fabs(dj_[iColumn]), biggestDj);
}
break;
}
}
if (numberFreeIn + numberFreeOut)
printf("%d in basis, %d out - largest dj %g\n",
numberFreeIn, numberFreeOut, biggestDj);
}
#endif
if (solveType <= 0) {
// Make sure everything is clean
for (iRow = 0; iRow < numberRows_; iRow++) {
if(getRowStatus(iRow) == isFixed) {
// double check fixed
if (rowUpperWork_[iRow] > rowLowerWork_[iRow])
setRowStatus(iRow, atLowerBound);
} else if (getRowStatus(iRow) == isFree) {
// may not be free after all
if (rowLowerWork_[iRow] > -largeValue_ || rowUpperWork_[iRow] < largeValue_)
setRowStatus(iRow, superBasic);
}
}
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
if(getColumnStatus(iColumn) == isFixed) {
// double check fixed
if (columnUpperWork_[iColumn] > columnLowerWork_[iColumn])
setColumnStatus(iColumn, atLowerBound);
} else if (getColumnStatus(iColumn) == isFree) {
// may not be free after all
if (columnLowerWork_[iColumn] > -largeValue_ || columnUpperWork_[iColumn] < largeValue_)
setColumnStatus(iColumn, superBasic);
}
}
if (!valuesPass) {
// not values pass so set to bounds
bool allSlack = true;
if (status_) {
for (iRow = 0; iRow < numberRows_; iRow++) {
if (getRowStatus(iRow) != basic) {
allSlack = false;
break;
}
}
}
if (!allSlack) {
//#define CLP_INVESTIGATE2
#ifdef CLP_INVESTIGATE3
int numberTotal = numberRows_ + numberColumns_;
double * saveSol = valuesPass ?
CoinCopyOfArray(solution_, numberTotal) : NULL;
#endif
// set values from warm start (if sensible)
int numberBasic = 0;
for (iRow = 0; iRow < numberRows_; iRow++) {
switch(getRowStatus(iRow)) {
case basic:
numberBasic++;
break;
case atUpperBound:
rowActivityWork_[iRow] = rowUpperWork_[iRow];
if (rowActivityWork_[iRow] > largeValue_) {
if (rowLowerWork_[iRow] > -largeValue_) {
rowActivityWork_[iRow] = rowLowerWork_[iRow];
setRowStatus(iRow, atLowerBound);
} else {
// say free
setRowStatus(iRow, isFree);
rowActivityWork_[iRow] = 0.0;
}
}
break;
case ClpSimplex::isFixed:
case atLowerBound:
rowActivityWork_[iRow] = rowLowerWork_[iRow];
if (rowActivityWork_[iRow] < -largeValue_) {
if (rowUpperWork_[iRow] < largeValue_) {
rowActivityWork_[iRow] = rowUpperWork_[iRow];
setRowStatus(iRow, atUpperBound);
} else {
// say free
setRowStatus(iRow, isFree);
rowActivityWork_[iRow] = 0.0;
}
}
break;
case isFree:
break;
// not really free - fall through to superbasic
case superBasic:
if (rowUpperWork_[iRow] > largeValue_) {
if (rowLowerWork_[iRow] > -largeValue_) {
rowActivityWork_[iRow] = rowLowerWork_[iRow];
setRowStatus(iRow, atLowerBound);
} else {
// say free
setRowStatus(iRow, isFree);
rowActivityWork_[iRow] = 0.0;
}
} else {
if (rowLowerWork_[iRow] > -largeValue_) {
// set to nearest
if (fabs(rowActivityWork_[iRow] - rowLowerWork_[iRow])
< fabs(rowActivityWork_[iRow] - rowLowerWork_[iRow])) {
rowActivityWork_[iRow] = rowLowerWork_[iRow];
setRowStatus(iRow, atLowerBound);
} else {
rowActivityWork_[iRow] = rowUpperWork_[iRow];
setRowStatus(iRow, atUpperBound);
}
} else {
rowActivityWork_[iRow] = rowUpperWork_[iRow];
setRowStatus(iRow, atUpperBound);
}
}
break;
}
}
totalSlacks = numberBasic;
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
switch(getColumnStatus(iColumn)) {
case basic:
if (numberBasic == maximumBasic_) {
// take out of basis
if (columnLowerWork_[iColumn] > -largeValue_) {
if (columnActivityWork_[iColumn] - columnLowerWork_[iColumn] <
columnUpperWork_[iColumn] - columnActivityWork_[iColumn]) {
columnActivityWork_[iColumn] = columnLowerWork_[iColumn];
setColumnStatus(iColumn, atLowerBound);
} else {
columnActivityWork_[iColumn] = columnUpperWork_[iColumn];
setColumnStatus(iColumn, atUpperBound);
}
} else if (columnUpperWork_[iColumn] < largeValue_) {
columnActivityWork_[iColumn] = columnUpperWork_[iColumn];
setColumnStatus(iColumn, atUpperBound);
} else {
columnActivityWork_[iColumn] = 0.0;
setColumnStatus(iColumn, isFree);
}
} else {
numberBasic++;
}
break;
case atUpperBound:
columnActivityWork_[iColumn] = columnUpperWork_[iColumn];
if (columnActivityWork_[iColumn] > largeValue_) {
if (columnLowerWork_[iColumn] < -largeValue_) {
columnActivityWork_[iColumn] = 0.0;
setColumnStatus(iColumn, isFree);
} else {
columnActivityWork_[iColumn] = columnLowerWork_[iColumn];
setColumnStatus(iColumn, atLowerBound);
}
}
break;
case isFixed:
case atLowerBound:
columnActivityWork_[iColumn] = columnLowerWork_[iColumn];
if (columnActivityWork_[iColumn] < -largeValue_) {
if (columnUpperWork_[iColumn] > largeValue_) {
columnActivityWork_[iColumn] = 0.0;
setColumnStatus(iColumn, isFree);
} else {
columnActivityWork_[iColumn] = columnUpperWork_[iColumn];
setColumnStatus(iColumn, atUpperBound);
}
}
break;
case isFree:
break;
// not really free - fall through to superbasic
case superBasic:
if (columnUpperWork_[iColumn] > largeValue_) {
if (columnLowerWork_[iColumn] > -largeValue_) {
columnActivityWork_[iColumn] = columnLowerWork_[iColumn];
setColumnStatus(iColumn, atLowerBound);
} else {
// say free
setColumnStatus(iColumn, isFree);
columnActivityWork_[iColumn] = 0.0;
}
} else {
if (columnLowerWork_[iColumn] > -largeValue_) {
// set to nearest
if (fabs(columnActivityWork_[iColumn] - columnLowerWork_[iColumn])
< fabs(columnActivityWork_[iColumn] - columnLowerWork_[iColumn])) {
columnActivityWork_[iColumn] = columnLowerWork_[iColumn];
setColumnStatus(iColumn, atLowerBound);
} else {
columnActivityWork_[iColumn] = columnUpperWork_[iColumn];
setColumnStatus(iColumn, atUpperBound);
}
} else {
columnActivityWork_[iColumn] = columnUpperWork_[iColumn];
setColumnStatus(iColumn, atUpperBound);
}
}
break;
}
}
#ifdef CLP_INVESTIGATE3
if (saveSol) {
int numberChanged = 0;
double largestChanged = 0.0;
for (int i = 0; i < numberTotal; i++) {
double difference = fabs(solution_[i] - saveSol[i]);
if (difference > 1.0e-7) {
numberChanged++;
if (difference > largestChanged)
largestChanged = difference;
}
}
if (numberChanged)
printf("%d changed, largest %g\n", numberChanged, largestChanged);
delete [] saveSol;
}
#endif
#if 0
if (numberBasic < numberRows_) {
// add some slacks in case odd warmstart
#ifdef CLP_INVESTIGATE
printf("BAD %d basic, %d rows %d slacks\n",
numberBasic, numberRows_, totalSlacks);
#endif
int iRow = numberRows_ - 1;
while (numberBasic < numberRows_) {
if (getRowStatus(iRow) != basic) {
setRowStatus(iRow, basic);
numberBasic++;
totalSlacks++;
iRow--;
} else {
break;
}
}
}
#endif
} else {
// all slack basis
int numberBasic = 0;
if (!status_) {
createStatus();
}
for (iRow = 0; iRow < numberRows_; iRow++) {
double lower = rowLowerWork_[iRow];
double upper = rowUpperWork_[iRow];
if (lower > -largeValue_ || upper < largeValue_) {
if (fabs(lower) <= fabs(upper)) {
rowActivityWork_[iRow] = lower;
} else {
rowActivityWork_[iRow] = upper;
}
} else {
rowActivityWork_[iRow] = 0.0;
}
setRowStatus(iRow, basic);
numberBasic++;
}
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
double lower = columnLowerWork_[iColumn];
double upper = columnUpperWork_[iColumn];
double big_bound = largeValue_;
if (lower > -big_bound || upper < big_bound) {
if ((getColumnStatus(iColumn) == atLowerBound &&
columnActivityWork_[iColumn] == lower) ||
(getColumnStatus(iColumn) == atUpperBound &&
columnActivityWork_[iColumn] == upper)) {
// status looks plausible
} else {
// set to sensible
if (fabs(lower) <= fabs(upper)) {
setColumnStatus(iColumn, atLowerBound);
columnActivityWork_[iColumn] = lower;
} else {
setColumnStatus(iColumn, atUpperBound);
columnActivityWork_[iColumn] = upper;
}
}
} else {
setColumnStatus(iColumn, isFree);
columnActivityWork_[iColumn] = 0.0;
}
}
}
} else {
// values pass has less coding
// make row activities correct and clean basis a bit
cleanStatus();
if (status_) {
int numberBasic = 0;
for (iRow = 0; iRow < numberRows_; iRow++) {
if (getRowStatus(iRow) == basic)
numberBasic++;
}
totalSlacks = numberBasic;
#if 0
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (getColumnStatus(iColumn) == basic)
numberBasic++;
}
#endif
} else {
// all slack basis
int numberBasic = 0;
if (!status_) {
createStatus();
}
for (iRow = 0; iRow < numberRows_; iRow++) {
setRowStatus(iRow, basic);
numberBasic++;
}
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
setColumnStatus(iColumn, superBasic);
// but put to bound if close
if (fabs(columnActivityWork_[iColumn] - columnLowerWork_[iColumn])
<= primalTolerance_) {
columnActivityWork_[iColumn] = columnLowerWork_[iColumn];
setColumnStatus(iColumn, atLowerBound);
} else if (fabs(columnActivityWork_[iColumn]
- columnUpperWork_[iColumn])
<= primalTolerance_) {
columnActivityWork_[iColumn] = columnUpperWork_[iColumn];
setColumnStatus(iColumn, atUpperBound);
}
}
}
}
numberRefinements_ = 1;
// set fixed if they are
for (iRow = 0; iRow < numberRows_; iRow++) {
if (getRowStatus(iRow) != basic ) {
if (rowLowerWork_[iRow] == rowUpperWork_[iRow]) {
rowActivityWork_[iRow] = rowLowerWork_[iRow];
setRowStatus(iRow, isFixed);
}
}
}
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (getColumnStatus(iColumn) != basic ) {
if (columnLowerWork_[iColumn] == columnUpperWork_[iColumn]) {
columnActivityWork_[iColumn] = columnLowerWork_[iColumn];
setColumnStatus(iColumn, isFixed);
}
}
}
}
//for (iRow=0;iRow<numberRows_+numberColumns_;iRow++) {
//if (fabs(solution_[iRow])>1.0e10) {
// printf("large %g at %d - status %d\n",
// solution_[iRow],iRow,status_[iRow]);
//}
//}
# if 0 //ndef _MSC_VER
// The local static var k is a problem when trying to build a DLL. Since this is
// just for debugging (likely done on *nix), just hide it from Windows
// -- lh, 101016 --
if (0) {
static int k = 0;
printf("start basis\n");
int i;
for (i = 0; i < numberRows_; i++)
printf ("xx %d %d\n", i, pivotVariable_[i]);
for (i = 0; i < numberRows_ + numberColumns_; i++)
if (getColumnStatus(i) == basic)
printf ("yy %d basic\n", i);
if (k > 20)
exit(0);
k++;
}
# endif
#if 0 //ndef NDEBUG
// Make sure everything is clean
double sumOutside=0.0;
int numberOutside=0;
//double sumOutsideLarge=0.0;
int numberOutsideLarge=0;
double sumInside=0.0;
int numberInside=0;
//double sumInsideLarge=0.0;
int numberInsideLarge=0;
int numberTotal=numberRows_+numberColumns_;
for (int iSequence = 0; iSequence < numberTotal; iSequence++) {
if(getStatus(iSequence) == isFixed) {
// double check fixed
assert (upper_[iSequence] == lower_[iSequence]);
assert (fabs(solution_[iSequence]-lower_[iSequence])<primalTolerance_);
} else if (getStatus(iSequence) == isFree) {
assert (upper_[iSequence] == COIN_DBL_MAX && lower_[iSequence]==-COIN_DBL_MAX);
} else if (getStatus(iSequence) == atLowerBound) {
assert (fabs(solution_[iSequence]-lower_[iSequence])<1000.0*primalTolerance_);
if (solution_[iSequence]<lower_[iSequence]) {
numberOutside++;
sumOutside-=solution_[iSequence]-lower_[iSequence];
if (solution_[iSequence]<lower_[iSequence]-primalTolerance_)
numberOutsideLarge++;
} else if (solution_[iSequence]>lower_[iSequence]) {
numberInside++;
sumInside+=solution_[iSequence]-lower_[iSequence];
if (solution_[iSequence]>lower_[iSequence]+primalTolerance_)
numberInsideLarge++;
}
} else if (getStatus(iSequence) == atUpperBound) {
assert (fabs(solution_[iSequence]-upper_[iSequence])<1000.0*primalTolerance_);
if (solution_[iSequence]>upper_[iSequence]) {
numberOutside++;
sumOutside+=solution_[iSequence]-upper_[iSequence];
if (solution_[iSequence]>upper_[iSequence]+primalTolerance_)
numberOutsideLarge++;
} else if (solution_[iSequence]<upper_[iSequence]) {
numberInside++;
sumInside-=solution_[iSequence]-upper_[iSequence];
if (solution_[iSequence]<upper_[iSequence]-primalTolerance_)
numberInsideLarge++;
}
} else if (getStatus(iSequence) == superBasic) {
//assert (!valuesPass);
}
}
if (numberInside+numberOutside)
printf("%d outside summing to %g (%d large), %d inside summing to %g (%d large)\n",
numberOutside,sumOutside,numberOutsideLarge,
numberInside,sumInside,numberInsideLarge);
#endif
int status = factorization_->factorize(this, solveType, valuesPass);
if (status) {
handler_->message(CLP_SIMPLEX_BADFACTOR, messages_)
<< status
<< CoinMessageEol;
return -1;
} else if (!solveType) {
// Initial basis - return number of singularities
int numberSlacks = 0;
for (iRow = 0; iRow < numberRows_; iRow++) {
if (getRowStatus(iRow) == basic)
numberSlacks++;
}
status = CoinMax(numberSlacks - totalSlacks, 0);
// special case if all slack
if (numberSlacks == numberRows_) {
status = numberRows_ + 1;
}
}
// sparse methods
//if (factorization_->sparseThreshold()) {
// get default value
factorization_->sparseThreshold(0);
if (!(moreSpecialOptions_&1024))
factorization_->goSparse();
//}
return status;
}
/*
This does basis housekeeping and does values for in/out variables.
Can also decide to re-factorize
*/
int
ClpSimplex::housekeeping(double objectiveChange)
{
// save value of incoming and outgoing
double oldIn = solution_[sequenceIn_];
double oldOut = solution_[sequenceOut_];
numberIterations_++;
changeMade_++; // something has happened
// incoming variable
if (handler_->logLevel() > 7) {
//if (handler_->detail(CLP_SIMPLEX_HOUSE1,messages_)<100) {
handler_->message(CLP_SIMPLEX_HOUSE1, messages_)
<< directionOut_
<< directionIn_ << theta_
<< dualOut_ << dualIn_ << alpha_
<< CoinMessageEol;
if (getStatus(sequenceIn_) == isFree) {
handler_->message(CLP_SIMPLEX_FREEIN, messages_)
<< sequenceIn_
<< CoinMessageEol;
}
}
#if 0
printf("h1 %d %d %g %g %g %g",
directionOut_
, directionIn_, theta_
, dualOut_, dualIn_, alpha_);
#endif
// change of incoming
char rowcol[] = {'R', 'C'};
if (pivotRow_ >= 0)
pivotVariable_[pivotRow_] = sequenceIn();
if (upper_[sequenceIn_] > 1.0e20 && lower_[sequenceIn_] < -1.0e20)
progressFlag_ |= 2; // making real progress
solution_[sequenceIn_] = valueIn_;
if (upper_[sequenceOut_] - lower_[sequenceOut_] < 1.0e-12)
progressFlag_ |= 1; // making real progress
if (sequenceIn_ != sequenceOut_) {
if (alphaAccuracy_ > 0.0) {
double value = fabs(alpha_);
if (value > 1.0)
alphaAccuracy_ *= value;
else
alphaAccuracy_ /= value;
}
//assert( getStatus(sequenceOut_)== basic);
setStatus(sequenceIn_, basic);
if (upper_[sequenceOut_] - lower_[sequenceOut_] > 0) {
// As Nonlinear costs may have moved bounds (to more feasible)
// Redo using value
if (fabs(valueOut_ - lower_[sequenceOut_]) < fabs(valueOut_ - upper_[sequenceOut_])) {
// going to lower
setStatus(sequenceOut_, atLowerBound);
oldOut = lower_[sequenceOut_];
} else {
// going to upper
setStatus(sequenceOut_, atUpperBound);
oldOut = upper_[sequenceOut_];
}
} else {
// fixed
setStatus(sequenceOut_, isFixed);
}
solution_[sequenceOut_] = valueOut_;
} else {
//if (objective_->type()<2)
//assert (fabs(theta_)>1.0e-13);
// flip from bound to bound
// As Nonlinear costs may have moved bounds (to more feasible)
// Redo using value
if (fabs(valueIn_ - lower_[sequenceIn_]) < fabs(valueIn_ - upper_[sequenceIn_])) {
// as if from upper bound
setStatus(sequenceIn_, atLowerBound);
} else {
// as if from lower bound
setStatus(sequenceIn_, atUpperBound);
}
}
// Update hidden stuff e.g. effective RHS and gub
int invertNow=matrix_->updatePivot(this, oldIn, oldOut);
objectiveValue_ += objectiveChange / (objectiveScale_ * rhsScale_);
if (handler_->logLevel() > 7) {
//if (handler_->detail(CLP_SIMPLEX_HOUSE2,messages_)<100) {
handler_->message(CLP_SIMPLEX_HOUSE2, messages_)
<< numberIterations_ << objectiveValue()
<< rowcol[isColumn(sequenceIn_)] << sequenceWithin(sequenceIn_)
<< rowcol[isColumn(sequenceOut_)] << sequenceWithin(sequenceOut_);
handler_->printing(algorithm_ < 0) << dualOut_ << theta_;
handler_->printing(algorithm_ > 0) << dualIn_ << theta_;
handler_->message() << CoinMessageEol;
}
#if 0
if (numberIterations_ > 10000)
printf(" it %d %g %c%d %c%d\n"
, numberIterations_, objectiveValue()
, rowcol[isColumn(sequenceIn_)], sequenceWithin(sequenceIn_)
, rowcol[isColumn(sequenceOut_)], sequenceWithin(sequenceOut_));
#endif
if (trustedUserPointer_ && trustedUserPointer_->typeStruct == 1) {
if (algorithm_ > 0 && integerType_ && !nonLinearCost_->numberInfeasibilities()) {
if (fabs(theta_) > 1.0e-6 || !numberIterations_) {
// For saving solutions
typedef struct {
int numberSolutions;
int maximumSolutions;
int numberColumns;
double ** solution;
int * numberUnsatisfied;
} clpSolution;
clpSolution * solution = reinterpret_cast<clpSolution *> (trustedUserPointer_->data);
if (solution->numberSolutions == solution->maximumSolutions) {
int n = solution->maximumSolutions;
int n2 = (n * 3) / 2 + 10;
solution->maximumSolutions = n2;
double ** temp = new double * [n2];
for (int i = 0; i < n; i++)
temp[i] = solution->solution[i];
delete [] solution->solution;
solution->solution = temp;
int * tempN = new int [n2];
for (int i = 0; i < n; i++)
tempN[i] = solution->numberUnsatisfied[i];
delete [] solution->numberUnsatisfied;
solution->numberUnsatisfied = tempN;
}
assert (numberColumns_ == solution->numberColumns);
double * sol = new double [numberColumns_];
solution->solution[solution->numberSolutions] = sol;
int numberFixed = 0;
int numberUnsat = 0;
int numberSat = 0;
double sumUnsat = 0.0;
double tolerance = 10.0 * primalTolerance_;
double mostAway = 0.0;
for (int i = 0; i < numberColumns_; i++) {
// Save anyway
sol[i] = columnScale_ ? solution_[i] * columnScale_[i] : solution_[i];
// rest is optional
if (upper_[i] > lower_[i]) {
double value = solution_[i];
if (value > lower_[i] + tolerance &&
value < upper_[i] - tolerance && integerType_[i]) {
// may have to modify value if scaled
if (columnScale_)
value *= columnScale_[i];
double closest = floor(value + 0.5);
// problem may be perturbed so relax test
if (fabs(value - closest) > 1.0e-4) {
numberUnsat++;
sumUnsat += fabs(value - closest);
if (mostAway < fabs(value - closest)) {
mostAway = fabs(value - closest);
}
} else {
numberSat++;
}
} else {
numberSat++;
}
} else {
numberFixed++;
}
}
solution->numberUnsatisfied[solution->numberSolutions++] = numberUnsat;
COIN_DETAIL_PRINT(printf("iteration %d, %d unsatisfied (%g,%g), %d fixed, %d satisfied\n",
numberIterations_, numberUnsat, sumUnsat, mostAway, numberFixed, numberSat));
}
}
}
if (hitMaximumIterations())
return 2;
#if 1
//if (numberIterations_>14000)
//handler_->setLogLevel(63);
//if (numberIterations_>24000)
//exit(77);
// check for small cycles
int in = sequenceIn_;
int out = sequenceOut_;
matrix_->correctSequence(this, in, out);
int cycle = progress_.cycle(in, out,
directionIn_, directionOut_);
if (cycle > 0 && objective_->type() < 2 && matrix_->type() < 15) {
//if (cycle>0) {
if (handler_->logLevel() >= 63)
printf("Cycle of %d\n", cycle);
// reset
progress_.startCheck();
double random = randomNumberGenerator_.randomDouble();
int extra = static_cast<int> (9.999 * random);
int off[] = {1, 1, 1, 1, 2, 2, 2, 3, 3, 4};
if (factorization_->pivots() > cycle) {
forceFactorization_ = CoinMax(1, cycle - off[extra]);
} else {
/* need to reject something
should be better if don't reject incoming
as it is in basis */
int iSequence;
//if (algorithm_ > 0)
// iSequence = sequenceIn_;
//else
iSequence = sequenceOut_;
char x = isColumn(iSequence) ? 'C' : 'R';
if (handler_->logLevel() >= 63)
handler_->message(CLP_SIMPLEX_FLAG, messages_)
<< x << sequenceWithin(iSequence)
<< CoinMessageEol;
setFlagged(iSequence);
//printf("flagging %d\n",iSequence);
}
return 1;
}
#endif
// only time to re-factorize if one before real time
// this is so user won't be surprised that maximumPivots has exact meaning
int numberPivots = factorization_->pivots();
int maximumPivots = factorization_->maximumPivots();
int numberDense = factorization_->numberDense();
bool dontInvert = ((specialOptions_ & 16384) != 0 && numberIterations_ * 3 >
2 * maximumIterations());
if (numberPivots == maximumPivots ||
maximumPivots < 2) {
// If dense then increase
if (maximumPivots > 100 && numberDense > 1.5 * maximumPivots) {
factorization_->maximumPivots(numberDense);
dualRowPivot_->maximumPivotsChanged();
primalColumnPivot_->maximumPivotsChanged();
// and redo arrays
for (int iRow = 0; iRow < 4; iRow++) {
int length = rowArray_[iRow]->capacity() + numberDense - maximumPivots;
rowArray_[iRow]->reserve(length);
}
}
return 1;
} else if ((factorization_->timeToRefactorize() && !dontInvert)
||invertNow) {
//printf("ret after %d pivots\n",factorization_->pivots());
return 1;
} else if (forceFactorization_ > 0 &&
factorization_->pivots() == forceFactorization_) {
// relax
forceFactorization_ = (3 + 5 * forceFactorization_) / 4;
if (forceFactorization_ > factorization_->maximumPivots())
forceFactorization_ = -1; //off
return 1;
} else if (numberIterations_ > 1000 + 10 * (numberRows_ + (numberColumns_ >> 2)) && matrix_->type()<15) {
double random = randomNumberGenerator_.randomDouble();
int maxNumber = (forceFactorization_ < 0) ? maximumPivots : CoinMin(forceFactorization_, maximumPivots);
if (factorization_->pivots() >= random * maxNumber) {
return 1;
} else if (numberIterations_ > 1000000 + 10 * (numberRows_ + (numberColumns_ >> 2)) &&
numberIterations_ < 1001000 + 10 * (numberRows_ + (numberColumns_ >> 2))) {
return 1;
} else {
// carry on iterating
return 0;
}
} else {
// carry on iterating
return 0;
}
}
// Copy constructor.
ClpSimplex::ClpSimplex(const ClpSimplex &rhs, int scalingMode) :
ClpModel(rhs, scalingMode),
bestPossibleImprovement_(0.0),
zeroTolerance_(1.0e-13),
columnPrimalSequence_(-2),
rowPrimalSequence_(-2),
bestObjectiveValue_(rhs.bestObjectiveValue_),
moreSpecialOptions_(2),
baseIteration_(0),
primalToleranceToGetOptimal_(-1.0),
largeValue_(1.0e15),
largestPrimalError_(0.0),
largestDualError_(0.0),
alphaAccuracy_(-1.0),
dualBound_(1.0e10),
alpha_(0.0),
theta_(0.0),
lowerIn_(0.0),
valueIn_(0.0),
upperIn_(-COIN_DBL_MAX),
dualIn_(0.0),
lowerOut_(-1),
valueOut_(-1),
upperOut_(-1),
dualOut_(-1),
dualTolerance_(1.0e-7),
primalTolerance_(1.0e-7),
sumDualInfeasibilities_(0.0),
sumPrimalInfeasibilities_(0.0),
infeasibilityCost_(1.0e10),
sumOfRelaxedDualInfeasibilities_(0.0),
sumOfRelaxedPrimalInfeasibilities_(0.0),
acceptablePivot_(1.0e-8),
lower_(NULL),
rowLowerWork_(NULL),
columnLowerWork_(NULL),
upper_(NULL),
rowUpperWork_(NULL),
columnUpperWork_(NULL),
cost_(NULL),
rowObjectiveWork_(NULL),
objectiveWork_(NULL),
sequenceIn_(-1),
directionIn_(-1),
sequenceOut_(-1),
directionOut_(-1),
pivotRow_(-1),
lastGoodIteration_(-100),
dj_(NULL),
rowReducedCost_(NULL),
reducedCostWork_(NULL),
solution_(NULL),
rowActivityWork_(NULL),
columnActivityWork_(NULL),
numberDualInfeasibilities_(0),
numberDualInfeasibilitiesWithoutFree_(0),
numberPrimalInfeasibilities_(100),
numberRefinements_(0),
pivotVariable_(NULL),
factorization_(NULL),
savedSolution_(NULL),
numberTimesOptimal_(0),
disasterArea_(NULL),
changeMade_(1),
algorithm_(0),
forceFactorization_(-1),
perturbation_(100),
nonLinearCost_(NULL),
lastBadIteration_(-999999),
lastFlaggedIteration_(-999999),
numberFake_(0),
numberChanged_(0),
progressFlag_(0),
firstFree_(-1),
numberExtraRows_(0),
maximumBasic_(0),
dontFactorizePivots_(0),
incomingInfeasibility_(1.0),
allowedInfeasibility_(10.0),
automaticScale_(0),
maximumPerturbationSize_(0),
perturbationArray_(NULL),
baseModel_(NULL)
#ifdef ABC_INHERIT
,abcSimplex_(NULL),
abcState_(0)
#endif
{
int i;
for (i = 0; i < 6; i++) {
rowArray_[i] = NULL;
columnArray_[i] = NULL;
}
for (i = 0; i < 4; i++) {
spareIntArray_[i] = 0;
spareDoubleArray_[i] = 0.0;
}
saveStatus_ = NULL;
factorization_ = NULL;
dualRowPivot_ = NULL;
primalColumnPivot_ = NULL;
gutsOfDelete(0);
delete nonLinearCost_;
nonLinearCost_ = NULL;
gutsOfCopy(rhs);
solveType_ = 1; // say simplex based life form
}
// Copy constructor from model
ClpSimplex::ClpSimplex(const ClpModel &rhs, int scalingMode) :
ClpModel(rhs, scalingMode),
bestPossibleImprovement_(0.0),
zeroTolerance_(1.0e-13),
columnPrimalSequence_(-2),
rowPrimalSequence_(-2),
bestObjectiveValue_(-COIN_DBL_MAX),
moreSpecialOptions_(2),
baseIteration_(0),
primalToleranceToGetOptimal_(-1.0),
largeValue_(1.0e15),
largestPrimalError_(0.0),
largestDualError_(0.0),
alphaAccuracy_(-1.0),
dualBound_(1.0e10),
alpha_(0.0),
theta_(0.0),
lowerIn_(0.0),
valueIn_(0.0),
upperIn_(-COIN_DBL_MAX),
dualIn_(0.0),
lowerOut_(-1),
valueOut_(-1),
upperOut_(-1),
dualOut_(-1),
dualTolerance_(1.0e-7),
primalTolerance_(1.0e-7),
sumDualInfeasibilities_(0.0),
sumPrimalInfeasibilities_(0.0),
infeasibilityCost_(1.0e10),
sumOfRelaxedDualInfeasibilities_(0.0),
sumOfRelaxedPrimalInfeasibilities_(0.0),
acceptablePivot_(1.0e-8),
lower_(NULL),
rowLowerWork_(NULL),
columnLowerWork_(NULL),
upper_(NULL),
rowUpperWork_(NULL),
columnUpperWork_(NULL),
cost_(NULL),
rowObjectiveWork_(NULL),
objectiveWork_(NULL),
sequenceIn_(-1),
directionIn_(-1),
sequenceOut_(-1),
directionOut_(-1),
pivotRow_(-1),
lastGoodIteration_(-100),
dj_(NULL),
rowReducedCost_(NULL),
reducedCostWork_(NULL),
solution_(NULL),
rowActivityWork_(NULL),
columnActivityWork_(NULL),
numberDualInfeasibilities_(0),
numberDualInfeasibilitiesWithoutFree_(0),
numberPrimalInfeasibilities_(100),
numberRefinements_(0),
pivotVariable_(NULL),
factorization_(NULL),
savedSolution_(NULL),
numberTimesOptimal_(0),
disasterArea_(NULL),
changeMade_(1),
algorithm_(0),
forceFactorization_(-1),
perturbation_(100),
nonLinearCost_(NULL),
lastBadIteration_(-999999),
lastFlaggedIteration_(-999999),
numberFake_(0),
numberChanged_(0),
progressFlag_(0),
firstFree_(-1),
numberExtraRows_(0),
maximumBasic_(0),
dontFactorizePivots_(0),
incomingInfeasibility_(1.0),
allowedInfeasibility_(10.0),
automaticScale_(0),
maximumPerturbationSize_(0),
perturbationArray_(NULL),
baseModel_(NULL)
#ifdef ABC_INHERIT
,abcSimplex_(NULL),
abcState_(0)
#endif
{
int i;
for (i = 0; i < 6; i++) {
rowArray_[i] = NULL;
columnArray_[i] = NULL;
}
for (i = 0; i < 4; i++) {
spareIntArray_[i] = 0;
spareDoubleArray_[i] = 0.0;
}
saveStatus_ = NULL;
// get an empty factorization so we can set tolerances etc
getEmptyFactorization();
// say Steepest pricing
dualRowPivot_ = new ClpDualRowSteepest();
// say Steepest pricing
primalColumnPivot_ = new ClpPrimalColumnSteepest();
solveType_ = 1; // say simplex based life form
}
// Assignment operator. This copies the data
ClpSimplex &
ClpSimplex::operator=(const ClpSimplex & rhs)
{
if (this != &rhs) {
gutsOfDelete(0);
delete nonLinearCost_;
nonLinearCost_ = NULL;
ClpModel::operator=(rhs);
gutsOfCopy(rhs);
}
return *this;
}
void
ClpSimplex::gutsOfCopy(const ClpSimplex & rhs)
{
assert (numberRows_ == rhs.numberRows_);
assert (numberColumns_ == rhs.numberColumns_);
numberExtraRows_ = rhs.numberExtraRows_;
maximumBasic_ = rhs.maximumBasic_;
dontFactorizePivots_ = rhs.dontFactorizePivots_;
int numberRows2 = numberRows_ + numberExtraRows_;
moreSpecialOptions_ = rhs.moreSpecialOptions_;
if ((whatsChanged_ & 1) != 0) {
int numberTotal = numberColumns_ + numberRows2;
if ((specialOptions_ & 65536) != 0 && maximumRows_ >= 0) {
assert (maximumInternalRows_ >= numberRows2);
assert (maximumInternalColumns_ >= numberColumns_);
numberTotal = 2 * (maximumInternalColumns_ + maximumInternalRows_);
}
lower_ = ClpCopyOfArray(rhs.lower_, numberTotal);
rowLowerWork_ = lower_ + numberColumns_;
columnLowerWork_ = lower_;
upper_ = ClpCopyOfArray(rhs.upper_, numberTotal);
rowUpperWork_ = upper_ + numberColumns_;
columnUpperWork_ = upper_;
cost_ = ClpCopyOfArray(rhs.cost_, numberTotal);
objectiveWork_ = cost_;
rowObjectiveWork_ = cost_ + numberColumns_;
dj_ = ClpCopyOfArray(rhs.dj_, numberTotal);
if (dj_) {
reducedCostWork_ = dj_;
rowReducedCost_ = dj_ + numberColumns_;
}
solution_ = ClpCopyOfArray(rhs.solution_, numberTotal);
if (solution_) {
columnActivityWork_ = solution_;
rowActivityWork_ = solution_ + numberColumns_;
}
if (rhs.pivotVariable_) {
pivotVariable_ = new int[numberRows2];
CoinMemcpyN ( rhs.pivotVariable_, numberRows2 , pivotVariable_);
} else {
pivotVariable_ = NULL;
}
savedSolution_ = ClpCopyOfArray(rhs.savedSolution_, numberTotal);
int i;
for (i = 0; i < 6; i++) {
rowArray_[i] = NULL;
if (rhs.rowArray_[i])
rowArray_[i] = new CoinIndexedVector(*rhs.rowArray_[i]);
columnArray_[i] = NULL;
if (rhs.columnArray_[i])
columnArray_[i] = new CoinIndexedVector(*rhs.columnArray_[i]);
}
if (rhs.saveStatus_) {
saveStatus_ = ClpCopyOfArray( rhs.saveStatus_, numberTotal);
}
} else {
lower_ = NULL;
rowLowerWork_ = NULL;
columnLowerWork_ = NULL;
upper_ = NULL;
rowUpperWork_ = NULL;
columnUpperWork_ = NULL;
cost_ = NULL;
objectiveWork_ = NULL;
rowObjectiveWork_ = NULL;
dj_ = NULL;
reducedCostWork_ = NULL;
rowReducedCost_ = NULL;
solution_ = NULL;
columnActivityWork_ = NULL;
rowActivityWork_ = NULL;
pivotVariable_ = NULL;
savedSolution_ = NULL;
int i;
for (i = 0; i < 6; i++) {
rowArray_[i] = NULL;
columnArray_[i] = NULL;
}
saveStatus_ = NULL;
}
if (rhs.factorization_) {
setFactorization(*rhs.factorization_);
} else {
delete factorization_;
factorization_ = NULL;
}
bestPossibleImprovement_ = rhs.bestPossibleImprovement_;
columnPrimalSequence_ = rhs.columnPrimalSequence_;
zeroTolerance_ = rhs.zeroTolerance_;
rowPrimalSequence_ = rhs.rowPrimalSequence_;
bestObjectiveValue_ = rhs.bestObjectiveValue_;
baseIteration_ = rhs.baseIteration_;
primalToleranceToGetOptimal_ = rhs.primalToleranceToGetOptimal_;
largeValue_ = rhs.largeValue_;
largestPrimalError_ = rhs.largestPrimalError_;
largestDualError_ = rhs.largestDualError_;
alphaAccuracy_ = rhs.alphaAccuracy_;
dualBound_ = rhs.dualBound_;
alpha_ = rhs.alpha_;
theta_ = rhs.theta_;
lowerIn_ = rhs.lowerIn_;
valueIn_ = rhs.valueIn_;
upperIn_ = rhs.upperIn_;
dualIn_ = rhs.dualIn_;
sequenceIn_ = rhs.sequenceIn_;
directionIn_ = rhs.directionIn_;
lowerOut_ = rhs.lowerOut_;
valueOut_ = rhs.valueOut_;
upperOut_ = rhs.upperOut_;
dualOut_ = rhs.dualOut_;
sequenceOut_ = rhs.sequenceOut_;
directionOut_ = rhs.directionOut_;
pivotRow_ = rhs.pivotRow_;
lastGoodIteration_ = rhs.lastGoodIteration_;
numberRefinements_ = rhs.numberRefinements_;
dualTolerance_ = rhs.dualTolerance_;
primalTolerance_ = rhs.primalTolerance_;
sumDualInfeasibilities_ = rhs.sumDualInfeasibilities_;
numberDualInfeasibilities_ = rhs.numberDualInfeasibilities_;
numberDualInfeasibilitiesWithoutFree_ =
rhs.numberDualInfeasibilitiesWithoutFree_;
sumPrimalInfeasibilities_ = rhs.sumPrimalInfeasibilities_;
numberPrimalInfeasibilities_ = rhs.numberPrimalInfeasibilities_;
dualRowPivot_ = rhs.dualRowPivot_->clone(true);
dualRowPivot_->setModel(this);
primalColumnPivot_ = rhs.primalColumnPivot_->clone(true);
primalColumnPivot_->setModel(this);
numberTimesOptimal_ = rhs.numberTimesOptimal_;
disasterArea_ = NULL;
changeMade_ = rhs.changeMade_;
algorithm_ = rhs.algorithm_;
forceFactorization_ = rhs.forceFactorization_;
perturbation_ = rhs.perturbation_;
infeasibilityCost_ = rhs.infeasibilityCost_;
lastBadIteration_ = rhs.lastBadIteration_;
lastFlaggedIteration_ = rhs.lastFlaggedIteration_;
numberFake_ = rhs.numberFake_;
numberChanged_ = rhs.numberChanged_;
progressFlag_ = rhs.progressFlag_;
firstFree_ = rhs.firstFree_;
incomingInfeasibility_ = rhs.incomingInfeasibility_;
allowedInfeasibility_ = rhs.allowedInfeasibility_;
automaticScale_ = rhs.automaticScale_;
#ifdef ABC_INHERIT
abcSimplex_ = NULL;
abcState_ = rhs.abcState_;
#endif
maximumPerturbationSize_ = rhs.maximumPerturbationSize_;
if (maximumPerturbationSize_ && maximumPerturbationSize_ >= 2 * numberColumns_) {
perturbationArray_ = CoinCopyOfArray(rhs.perturbationArray_,
maximumPerturbationSize_);
} else {
maximumPerturbationSize_ = 0;
perturbationArray_ = NULL;
}
if (rhs.baseModel_) {
baseModel_ = new ClpSimplex(*rhs.baseModel_);
} else {
baseModel_ = NULL;
}
progress_ = rhs.progress_;
for (int i = 0; i < 4; i++) {
spareIntArray_[i] = rhs.spareIntArray_[i];
spareDoubleArray_[i] = rhs.spareDoubleArray_[i];
}
sumOfRelaxedDualInfeasibilities_ = rhs.sumOfRelaxedDualInfeasibilities_;
sumOfRelaxedPrimalInfeasibilities_ = rhs.sumOfRelaxedPrimalInfeasibilities_;
acceptablePivot_ = rhs.acceptablePivot_;
if (rhs.nonLinearCost_ != NULL)
nonLinearCost_ = new ClpNonLinearCost(*rhs.nonLinearCost_);
else
nonLinearCost_ = NULL;
solveType_ = rhs.solveType_;
}
// type == 0 do everything, most + pivot data, 2 factorization data as well
void
ClpSimplex::gutsOfDelete(int type)
{
if (!type || (specialOptions_ & 65536) == 0) {
maximumInternalColumns_ = -1;
maximumInternalRows_ = -1;
delete [] lower_;
lower_ = NULL;
rowLowerWork_ = NULL;
columnLowerWork_ = NULL;
delete [] upper_;
upper_ = NULL;
rowUpperWork_ = NULL;
columnUpperWork_ = NULL;
delete [] cost_;
cost_ = NULL;
objectiveWork_ = NULL;
rowObjectiveWork_ = NULL;
delete [] dj_;
dj_ = NULL;
reducedCostWork_ = NULL;
rowReducedCost_ = NULL;
delete [] solution_;
solution_ = NULL;
rowActivityWork_ = NULL;
columnActivityWork_ = NULL;
delete [] savedSolution_;
savedSolution_ = NULL;
}
if ((specialOptions_ & 2) == 0) {
delete nonLinearCost_;
nonLinearCost_ = NULL;
}
int i;
if ((specialOptions_ & 65536) == 0) {
for (i = 0; i < 6; i++) {
delete rowArray_[i];
rowArray_[i] = NULL;
delete columnArray_[i];
columnArray_[i] = NULL;
}
}
delete [] saveStatus_;
saveStatus_ = NULL;
if (type != 1) {
delete rowCopy_;
rowCopy_ = NULL;
}
if (!type) {
// delete everything
setEmptyFactorization();
delete [] pivotVariable_;
pivotVariable_ = NULL;
delete dualRowPivot_;
dualRowPivot_ = NULL;
delete primalColumnPivot_;
primalColumnPivot_ = NULL;
delete baseModel_;
baseModel_ = NULL;
delete [] perturbationArray_;
perturbationArray_ = NULL;
maximumPerturbationSize_ = 0;
} else {
// delete any size information in methods
if (type > 1) {
//assert (factorization_);
if (factorization_)
factorization_->clearArrays();
delete [] pivotVariable_;
pivotVariable_ = NULL;
}
dualRowPivot_->clearArrays();
primalColumnPivot_->clearArrays();
}
}
// This sets largest infeasibility and most infeasible
void
ClpSimplex::checkPrimalSolution(const double * rowActivities,
const double * columnActivities)
{
double * solution;
int iRow, iColumn;
objectiveValue_ = 0.0;
// now look at primal solution
solution = rowActivityWork_;
sumPrimalInfeasibilities_ = 0.0;
numberPrimalInfeasibilities_ = 0;
double primalTolerance = primalTolerance_;
double relaxedTolerance = primalTolerance_;
// we can't really trust infeasibilities if there is primal error
double error = CoinMin(1.0e-2, largestPrimalError_);
// allow tolerance at least slightly bigger than standard
relaxedTolerance = relaxedTolerance + error;
sumOfRelaxedPrimalInfeasibilities_ = 0.0;
for (iRow = 0; iRow < numberRows_; iRow++) {
//assert (fabs(solution[iRow])<1.0e15||getRowStatus(iRow) == basic);
double infeasibility = 0.0;
objectiveValue_ += solution[iRow] * rowObjectiveWork_[iRow];
if (solution[iRow] > rowUpperWork_[iRow]) {
infeasibility = solution[iRow] - rowUpperWork_[iRow];
} else if (solution[iRow] < rowLowerWork_[iRow]) {
infeasibility = rowLowerWork_[iRow] - solution[iRow];
}
if (infeasibility > primalTolerance) {
sumPrimalInfeasibilities_ += infeasibility - primalTolerance_;
if (infeasibility > relaxedTolerance)
sumOfRelaxedPrimalInfeasibilities_ += infeasibility - relaxedTolerance;
numberPrimalInfeasibilities_ ++;
}
infeasibility = fabs(rowActivities[iRow] - solution[iRow]);
}
// Check any infeasibilities from dynamic rows
matrix_->primalExpanded(this, 2);
solution = columnActivityWork_;
if (!matrix_->rhsOffset(this)) {
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
//assert (fabs(solution[iColumn])<1.0e15||getColumnStatus(iColumn) == basic);
double infeasibility = 0.0;
objectiveValue_ += objectiveWork_[iColumn] * solution[iColumn];
if (solution[iColumn] > columnUpperWork_[iColumn]) {
infeasibility = solution[iColumn] - columnUpperWork_[iColumn];
} else if (solution[iColumn] < columnLowerWork_[iColumn]) {
infeasibility = columnLowerWork_[iColumn] - solution[iColumn];
}
if (infeasibility > primalTolerance) {
sumPrimalInfeasibilities_ += infeasibility - primalTolerance_;
if (infeasibility > relaxedTolerance)
sumOfRelaxedPrimalInfeasibilities_ += infeasibility - relaxedTolerance;
numberPrimalInfeasibilities_ ++;
}
infeasibility = fabs(columnActivities[iColumn] - solution[iColumn]);
}
} else {
// as we are using effective rhs we only check basics
// But we do need to get objective
objectiveValue_ += innerProduct(objectiveWork_, numberColumns_, solution);
for (int j = 0; j < numberRows_; j++) {
int iColumn = pivotVariable_[j];
//assert (fabs(solution[iColumn])<1.0e15||getColumnStatus(iColumn) == basic);
double infeasibility = 0.0;
if (solution[iColumn] > columnUpperWork_[iColumn]) {
infeasibility = solution[iColumn] - columnUpperWork_[iColumn];
} else if (solution[iColumn] < columnLowerWork_[iColumn]) {
infeasibility = columnLowerWork_[iColumn] - solution[iColumn];
}
if (infeasibility > primalTolerance) {
sumPrimalInfeasibilities_ += infeasibility - primalTolerance_;
if (infeasibility > relaxedTolerance)
sumOfRelaxedPrimalInfeasibilities_ += infeasibility - relaxedTolerance;
numberPrimalInfeasibilities_ ++;
}
infeasibility = fabs(columnActivities[iColumn] - solution[iColumn]);
}
}
objectiveValue_ += objective_->nonlinearOffset();
objectiveValue_ /= (objectiveScale_ * rhsScale_);
}
void
ClpSimplex::checkDualSolution()
{
int iRow, iColumn;
sumDualInfeasibilities_ = 0.0;
numberDualInfeasibilities_ = 0;
numberDualInfeasibilitiesWithoutFree_ = 0;
if (matrix_->skipDualCheck() && algorithm_ > 0 && problemStatus_ == -2) {
// pretend we found dual infeasibilities
sumOfRelaxedDualInfeasibilities_ = 1.0;
sumDualInfeasibilities_ = 1.0;
numberDualInfeasibilities_ = 1;
return;
}
int firstFreePrimal = -1;
int firstFreeDual = -1;
int numberSuperBasicWithDj = 0;
bestPossibleImprovement_ = 0.0;
// we can't really trust infeasibilities if there is dual error
double error = CoinMin(1.0e-2, largestDualError_);
// allow tolerance at least slightly bigger than standard
double relaxedTolerance = dualTolerance_ + error;
// allow bigger tolerance for possible improvement
double possTolerance = 5.0 * relaxedTolerance;
sumOfRelaxedDualInfeasibilities_ = 0.0;
// Check any djs from dynamic rows
matrix_->dualExpanded(this, NULL, NULL, 3);
numberDualInfeasibilitiesWithoutFree_ = numberDualInfeasibilities_;
objectiveValue_ = 0.0;
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
objectiveValue_ += objectiveWork_[iColumn] * columnActivityWork_[iColumn];
if (getColumnStatus(iColumn) != basic && !flagged(iColumn)) {
// not basic
double distanceUp = columnUpperWork_[iColumn] -
columnActivityWork_[iColumn];
double distanceDown = columnActivityWork_[iColumn] -
columnLowerWork_[iColumn];
if (distanceUp > primalTolerance_) {
double value = reducedCostWork_[iColumn];
// Check if "free"
if (distanceDown > primalTolerance_) {
if (fabs(value) > 1.0e2 * relaxedTolerance) {
numberSuperBasicWithDj++;
if (firstFreeDual < 0)
firstFreeDual = iColumn;
}
if (firstFreePrimal < 0)
firstFreePrimal = iColumn;
}
// should not be negative
if (value < 0.0) {
value = - value;
if (value > dualTolerance_) {
if (getColumnStatus(iColumn) != isFree) {
numberDualInfeasibilitiesWithoutFree_ ++;
sumDualInfeasibilities_ += value - dualTolerance_;
if (value > possTolerance)
bestPossibleImprovement_ += CoinMin(distanceUp, 1.0e10) * value;
if (value > relaxedTolerance)
sumOfRelaxedDualInfeasibilities_ += value - relaxedTolerance;
numberDualInfeasibilities_ ++;
} else {
// free so relax a lot
value *= 0.01;
if (value > dualTolerance_) {
sumDualInfeasibilities_ += value - dualTolerance_;
if (value > possTolerance)
bestPossibleImprovement_ = 1.0e100;
if (value > relaxedTolerance)
sumOfRelaxedDualInfeasibilities_ += value - relaxedTolerance;
numberDualInfeasibilities_ ++;
}
}
}
}
}
if (distanceDown > primalTolerance_) {
double value = reducedCostWork_[iColumn];
// should not be positive
if (value > 0.0) {
if (value > dualTolerance_) {
sumDualInfeasibilities_ += value - dualTolerance_;
if (value > possTolerance)
bestPossibleImprovement_ += value * CoinMin(distanceDown, 1.0e10);
if (value > relaxedTolerance)
sumOfRelaxedDualInfeasibilities_ += value - relaxedTolerance;
numberDualInfeasibilities_ ++;
if (getColumnStatus(iColumn) != isFree)
numberDualInfeasibilitiesWithoutFree_ ++;
// maybe we can make feasible by increasing tolerance
}
}
}
}
}
for (iRow = 0; iRow < numberRows_; iRow++) {
objectiveValue_ += rowActivityWork_[iRow] * rowObjectiveWork_[iRow];
if (getRowStatus(iRow) != basic && !flagged(iRow + numberColumns_)) {
// not basic
double distanceUp = rowUpperWork_[iRow] - rowActivityWork_[iRow];
double distanceDown = rowActivityWork_[iRow] - rowLowerWork_[iRow];
if (distanceUp > primalTolerance_) {
double value = rowReducedCost_[iRow];
// Check if "free"
if (distanceDown > primalTolerance_) {
if (fabs(value) > 1.0e2 * relaxedTolerance) {
numberSuperBasicWithDj++;
if (firstFreeDual < 0)
firstFreeDual = iRow + numberColumns_;
}
if (firstFreePrimal < 0)
firstFreePrimal = iRow + numberColumns_;
}
// should not be negative
if (value < 0.0) {
value = - value;
if (value > dualTolerance_) {
sumDualInfeasibilities_ += value - dualTolerance_;
if (value > possTolerance)
bestPossibleImprovement_ += value * CoinMin(distanceUp, 1.0e10);
if (value > relaxedTolerance)
sumOfRelaxedDualInfeasibilities_ += value - relaxedTolerance;
numberDualInfeasibilities_ ++;
if (getRowStatus(iRow) != isFree)
numberDualInfeasibilitiesWithoutFree_ ++;
}
}
}
if (distanceDown > primalTolerance_) {
double value = rowReducedCost_[iRow];
// should not be positive
if (value > 0.0) {
if (value > dualTolerance_) {
sumDualInfeasibilities_ += value - dualTolerance_;
if (value > possTolerance)
bestPossibleImprovement_ += value * CoinMin(distanceDown, 1.0e10);
if (value > relaxedTolerance)
sumOfRelaxedDualInfeasibilities_ += value - relaxedTolerance;
numberDualInfeasibilities_ ++;
if (getRowStatus(iRow) != isFree)
numberDualInfeasibilitiesWithoutFree_ ++;
// maybe we can make feasible by increasing tolerance
}
}
}
}
}
if (algorithm_ < 0 && firstFreeDual >= 0) {
// dual
firstFree_ = firstFreeDual;
} else if (numberSuperBasicWithDj ||
(progress_.lastIterationNumber(0) <= 0)) {
firstFree_ = firstFreePrimal;
}
objectiveValue_ += objective_->nonlinearOffset();
objectiveValue_ /= (objectiveScale_ * rhsScale_);
}
/* This sets sum and number of infeasibilities (Dual and Primal) */
void
ClpSimplex::checkBothSolutions()
{
if ((matrix_->skipDualCheck() && algorithm_ > 0 && problemStatus_ == -2) ||
matrix_->rhsOffset(this)) {
// Say may be free or superbasic
moreSpecialOptions_ &= ~8;
// old way
checkPrimalSolution(rowActivityWork_, columnActivityWork_);
checkDualSolution();
return;
}
int iSequence;
assert (dualTolerance_ > 0.0 && dualTolerance_ < 1.0e10);
assert (primalTolerance_ > 0.0 && primalTolerance_ < 1.0e10);
objectiveValue_ = 0.0;
sumPrimalInfeasibilities_ = 0.0;
numberPrimalInfeasibilities_ = 0;
double primalTolerance = primalTolerance_;
double relaxedToleranceP = primalTolerance_;
// we can't really trust infeasibilities if there is primal error
double error = CoinMin(1.0e-2, largestPrimalError_);
// allow tolerance at least slightly bigger than standard
relaxedToleranceP = relaxedToleranceP + error;
sumOfRelaxedPrimalInfeasibilities_ = 0.0;
sumDualInfeasibilities_ = 0.0;
numberDualInfeasibilities_ = 0;
double dualTolerance = dualTolerance_;
double relaxedToleranceD = dualTolerance;
// we can't really trust infeasibilities if there is dual error
error = CoinMin(1.0e-2, largestDualError_);
// allow tolerance at least slightly bigger than standard
relaxedToleranceD = relaxedToleranceD + error;
// allow bigger tolerance for possible improvement
double possTolerance = 5.0 * relaxedToleranceD;
sumOfRelaxedDualInfeasibilities_ = 0.0;
bestPossibleImprovement_ = 0.0;
// Check any infeasibilities from dynamic rows
matrix_->primalExpanded(this, 2);
// Check any djs from dynamic rows
matrix_->dualExpanded(this, NULL, NULL, 3);
int numberDualInfeasibilitiesFree = 0;
int firstFreePrimal = -1;
int firstFreeDual = -1;
int numberSuperBasicWithDj = 0;
int numberTotal = numberRows_ + numberColumns_;
// Say no free or superbasic
moreSpecialOptions_ |= 8;
//#define PRINT_INFEAS
#ifdef PRINT_INFEAS
int seqInf[10];
#endif
for (iSequence = 0; iSequence < numberTotal; iSequence++) {
double value = solution_[iSequence];
#ifdef COIN_DEBUG
if (fabs(value) > 1.0e20)
printf("%d values %g %g %g - status %d\n", iSequence, lower_[iSequence],
solution_[iSequence], upper_[iSequence], status_[iSequence]);
#endif
objectiveValue_ += value * cost_[iSequence];
double distanceUp = upper_[iSequence] - value;
double distanceDown = value - lower_[iSequence];
if (distanceUp < -primalTolerance) {
double infeasibility = -distanceUp;
sumPrimalInfeasibilities_ += infeasibility - primalTolerance_;
if (infeasibility > relaxedToleranceP)
sumOfRelaxedPrimalInfeasibilities_ += infeasibility - relaxedToleranceP;
#ifdef PRINT_INFEAS
if (numberPrimalInfeasibilities_<10) {
seqInf[numberPrimalInfeasibilities_]=iSequence;
}
#endif
numberPrimalInfeasibilities_ ++;
} else if (distanceDown < -primalTolerance) {
double infeasibility = -distanceDown;
sumPrimalInfeasibilities_ += infeasibility - primalTolerance_;
if (infeasibility > relaxedToleranceP)
sumOfRelaxedPrimalInfeasibilities_ += infeasibility - relaxedToleranceP;
#ifdef PRINT_INFEAS
if (numberPrimalInfeasibilities_<10) {
seqInf[numberPrimalInfeasibilities_]=iSequence;
}
#endif
numberPrimalInfeasibilities_ ++;
} else {
// feasible (so could be free)
if (getStatus(iSequence) != basic && !flagged(iSequence)) {
// not basic
double djValue = dj_[iSequence];
if (distanceDown < primalTolerance) {
if (distanceUp > primalTolerance && djValue < -dualTolerance) {
sumDualInfeasibilities_ -= djValue + dualTolerance;
if (djValue < -possTolerance)
bestPossibleImprovement_ -= distanceUp * djValue;
if (djValue < -relaxedToleranceD)
sumOfRelaxedDualInfeasibilities_ -= djValue + relaxedToleranceD;
numberDualInfeasibilities_ ++;
}
} else if (distanceUp < primalTolerance) {
if (djValue > dualTolerance) {
sumDualInfeasibilities_ += djValue - dualTolerance;
if (djValue > possTolerance)
bestPossibleImprovement_ += distanceDown * djValue;
if (djValue > relaxedToleranceD)
sumOfRelaxedDualInfeasibilities_ += djValue - relaxedToleranceD;
numberDualInfeasibilities_ ++;
}
} else {
// may be free
// Say free or superbasic
moreSpecialOptions_ &= ~8;
djValue *= 0.01;
if (fabs(djValue) > dualTolerance) {
if (getStatus(iSequence) == isFree)
numberDualInfeasibilitiesFree++;
sumDualInfeasibilities_ += fabs(djValue) - dualTolerance;
bestPossibleImprovement_ = 1.0e100;
numberDualInfeasibilities_ ++;
if (fabs(djValue) > relaxedToleranceD) {
sumOfRelaxedDualInfeasibilities_ += value - relaxedToleranceD;
numberSuperBasicWithDj++;
if (firstFreeDual < 0)
firstFreeDual = iSequence;
}
}
if (firstFreePrimal < 0)
firstFreePrimal = iSequence;
}
}
}
}
objectiveValue_ += objective_->nonlinearOffset();
objectiveValue_ /= (objectiveScale_ * rhsScale_);
numberDualInfeasibilitiesWithoutFree_ = numberDualInfeasibilities_ -
numberDualInfeasibilitiesFree;
#ifdef PRINT_INFEAS
if (numberPrimalInfeasibilities_<=10) {
printf("---------------start-----------\n");
if (!rowScale_) {
for (int i=0;i<numberPrimalInfeasibilities_;i++) {
int iSeq = seqInf[i];
double infeas;
if (solution_[iSeq]<lower_[iSeq])
infeas = lower_[iSeq]-solution_[iSeq];
else
infeas = solution_[iSeq]-upper_[iSeq];
if (iSeq<numberColumns_) {
printf("INF C%d %.10g <= %.10g <= %.10g - infeas %g\n",
iSeq,lower_[iSeq],solution_[iSeq],upper_[iSeq],infeas);
} else {
printf("INF R%d %.10g <= %.10g <= %.10g - infeas %g\n",
iSeq-numberColumns_,lower_[iSeq],solution_[iSeq],upper_[iSeq],infeas);
}
}
} else {
for (int i=0;i<numberPrimalInfeasibilities_;i++) {
int iSeq = seqInf[i];
double infeas;
if (solution_[iSeq]<lower_[iSeq])
infeas = lower_[iSeq]-solution_[iSeq];
else
infeas = solution_[iSeq]-upper_[iSeq];
double unscaled = infeas;
if (iSeq<numberColumns_) {
unscaled *= columnScale_[iSeq];
printf("INF C%d %.10g <= %.10g <= %.10g - infeas %g - unscaled %g\n",
iSeq,lower_[iSeq],solution_[iSeq],upper_[iSeq],infeas,unscaled);
} else {
unscaled /= rowScale_[iSeq-numberColumns_];
printf("INF R%d %.10g <= %.10g <= %.10g - infeas %g - unscaled %g\n",
iSeq-numberColumns_,lower_[iSeq],solution_[iSeq],upper_[iSeq],infeas,unscaled);
}
}
}
}
#endif
if (algorithm_ < 0 && firstFreeDual >= 0) {
// dual
firstFree_ = firstFreeDual;
} else if (numberSuperBasicWithDj ||
(progress_.lastIterationNumber(0) <= 0)) {
firstFree_ = firstFreePrimal;
}
}
/* Adds multiple of a column into an array */
void
ClpSimplex::add(double * array,
int sequence, double multiplier) const
{
if (sequence >= numberColumns_ && sequence < numberColumns_ + numberRows_) {
//slack
array [sequence-numberColumns_] -= multiplier;
} else {
// column
matrix_->add(this, array, sequence, multiplier);
}
}
/*
Unpacks one column of the matrix into indexed array
*/
void
ClpSimplex::unpack(CoinIndexedVector * rowArray) const
{
rowArray->clear();
if (sequenceIn_ >= numberColumns_ && sequenceIn_ < numberColumns_ + numberRows_) {
//slack
rowArray->insert(sequenceIn_ - numberColumns_, -1.0);
} else {
// column
matrix_->unpack(this, rowArray, sequenceIn_);
}
}
void
ClpSimplex::unpack(CoinIndexedVector * rowArray, int sequence) const
{
rowArray->clear();
if (sequence >= numberColumns_ && sequence < numberColumns_ + numberRows_) {
//slack
rowArray->insert(sequence - numberColumns_, -1.0);
} else {
// column
matrix_->unpack(this, rowArray, sequence);
}
}
/*
Unpacks one column of the matrix into indexed array
*/
void
ClpSimplex::unpackPacked(CoinIndexedVector * rowArray)
{
rowArray->clear();
if (sequenceIn_ >= numberColumns_ && sequenceIn_ < numberColumns_ + numberRows_) {
//slack
int * index = rowArray->getIndices();
double * array = rowArray->denseVector();
array[0] = -1.0;
index[0] = sequenceIn_ - numberColumns_;
rowArray->setNumElements(1);
rowArray->setPackedMode(true);
} else {
// column
matrix_->unpackPacked(this, rowArray, sequenceIn_);
}
}
void
ClpSimplex::unpackPacked(CoinIndexedVector * rowArray, int sequence)
{
rowArray->clear();
if (sequence >= numberColumns_ && sequence < numberColumns_ + numberRows_) {
//slack
int * index = rowArray->getIndices();
double * array = rowArray->denseVector();
array[0] = -1.0;
index[0] = sequence - numberColumns_;
rowArray->setNumElements(1);
rowArray->setPackedMode(true);
} else {
// column
matrix_->unpackPacked(this, rowArray, sequence);
}
}
//static int x_gaps[4]={0,0,0,0};
//static int scale_times[]={0,0,0,0};
bool
ClpSimplex::createRim(int what, bool makeRowCopy, int startFinishOptions)
{
bool goodMatrix = true;
int saveLevel = handler_->logLevel();
spareIntArray_[0] = 0;
if (!matrix_->canGetRowCopy())
makeRowCopy = false; // switch off row copy if can't produce
// Arrays will be there and correct size unless what is 63
bool newArrays = (what == 63);
// We may be restarting with same size
bool keepPivots = false;
if (startFinishOptions == -1) {
startFinishOptions = 0;
keepPivots = true;
}
bool oldMatrix = ((startFinishOptions & 4) != 0 && (whatsChanged_ & 1) != 0);
if (what == 63) {
pivotRow_ = -1;
if (!status_)
createStatus();
if (oldMatrix)
newArrays = false;
if (problemStatus_ == 10) {
handler_->setLogLevel(0); // switch off messages
if (rowArray_[0]) {
// stuff is still there
oldMatrix = true;
newArrays = false;
keepPivots = true;
for (int iRow = 0; iRow < 4; iRow++) {
rowArray_[iRow]->clear();
}
for (int iColumn = 0; iColumn < 2; iColumn++) {
columnArray_[iColumn]->clear();
}
}
} else if (factorization_) {
// match up factorization messages
if (handler_->logLevel() < 3)
factorization_->messageLevel(0);
else
factorization_->messageLevel(CoinMax(3, factorization_->messageLevel()));
/* Faster to keep pivots rather than re-scan matrix. Matrix may have changed
i.e. oldMatrix false but okay as long as same number rows and status array exists
*/
if ((startFinishOptions & 2) != 0 && factorization_->numberRows() == numberRows_ && status_)
keepPivots = true;
}
numberExtraRows_ = matrix_->generalExpanded(this, 2, maximumBasic_);
if (numberExtraRows_ && newArrays) {
// make sure status array large enough
assert (status_);
int numberOld = numberRows_ + numberColumns_;
int numberNew = numberRows_ + numberColumns_ + numberExtraRows_;
unsigned char * newStatus = new unsigned char [numberNew];
memset(newStatus + numberOld, 0, numberExtraRows_);
CoinMemcpyN(status_, numberOld, newStatus);
delete [] status_;
status_ = newStatus;
}
}
int numberRows2 = numberRows_ + numberExtraRows_;
int numberTotal = numberRows2 + numberColumns_;
if ((specialOptions_ & 65536) != 0) {
assert (!numberExtraRows_);
if (!cost_ || numberRows2 > maximumInternalRows_ ||
numberColumns_ > maximumInternalColumns_) {
newArrays = true;
keepPivots = false;
COIN_DETAIL_PRINT(printf("createrim a %d rows, %d maximum rows %d maxinternal\n",
numberRows_, maximumRows_, maximumInternalRows_));
int oldMaximumRows = maximumInternalRows_;
int oldMaximumColumns = maximumInternalColumns_;
if (cost_) {
if (numberRows2 > maximumInternalRows_)
maximumInternalRows_ = numberRows2;
if (numberColumns_ > maximumInternalColumns_)
maximumInternalColumns_ = numberColumns_;
} else {
maximumInternalRows_ = numberRows2;
maximumInternalColumns_ = numberColumns_;
}
//maximumRows_=CoinMax(maximumInternalRows_,maximumRows_);
//maximumColumns_=CoinMax(maximumInternalColumns_,maximumColumns_);
assert(maximumInternalRows_ == maximumRows_);
assert(maximumInternalColumns_ == maximumColumns_);
COIN_DETAIL_PRINT(printf("createrim b %d rows, %d maximum rows, %d maxinternal\n",
numberRows_, maximumRows_, maximumInternalRows_));
int numberTotal2 = (maximumInternalRows_ + maximumInternalColumns_) * 2;
delete [] cost_;
cost_ = new double[numberTotal2];
delete [] lower_;
delete [] upper_;
lower_ = new double[numberTotal2];
upper_ = new double[numberTotal2];
delete [] dj_;
dj_ = new double[numberTotal2];
delete [] solution_;
solution_ = new double[numberTotal2];
// ***** should be non NULL but seems to be too much
//printf("resize %d savedRowScale %x\n",maximumRows_,savedRowScale_);
if (savedRowScale_) {
assert (oldMaximumRows > 0);
double * temp;
temp = new double [4*maximumRows_];
CoinFillN(temp, 4 * maximumRows_, 1.0);
CoinMemcpyN(savedRowScale_, numberRows_, temp);
CoinMemcpyN(savedRowScale_ + oldMaximumRows, numberRows_, temp + maximumRows_);
CoinMemcpyN(savedRowScale_ + 2 * oldMaximumRows, numberRows_, temp + 2 * maximumRows_);
CoinMemcpyN(savedRowScale_ + 3 * oldMaximumRows, numberRows_, temp + 3 * maximumRows_);
delete [] savedRowScale_;
savedRowScale_ = temp;
temp = new double [4*maximumColumns_];
CoinFillN(temp, 4 * maximumColumns_, 1.0);
CoinMemcpyN(savedColumnScale_, numberColumns_, temp);
CoinMemcpyN(savedColumnScale_ + oldMaximumColumns, numberColumns_, temp + maximumColumns_);
CoinMemcpyN(savedColumnScale_ + 2 * oldMaximumColumns, numberColumns_, temp + 2 * maximumColumns_);
CoinMemcpyN(savedColumnScale_ + 3 * oldMaximumColumns, numberColumns_, temp + 3 * maximumColumns_);
delete [] savedColumnScale_;
savedColumnScale_ = temp;
}
}
}
int i;
bool doSanityCheck = true;
if (what == 63) {
// We may want to switch stuff off for speed
if ((specialOptions_ & 256) != 0)
makeRowCopy = false; // no row copy
if ((specialOptions_ & 128) != 0)
doSanityCheck = false; // no sanity check
//check matrix
if (!matrix_)
matrix_ = new ClpPackedMatrix();
int checkType = (doSanityCheck) ? 15 : 14;
if (oldMatrix)
checkType = 14;
bool inCbcOrOther = (specialOptions_ & 0x03000000) != 0;
if (inCbcOrOther)
checkType -= 4; // don't check for duplicates
if (!matrix_->allElementsInRange(this, smallElement_, 1.0e20, checkType)) {
problemStatus_ = 4;
secondaryStatus_ = 8;
//goodMatrix= false;
return false;
}
bool rowCopyIsScaled;
if (makeRowCopy) {
if(!oldMatrix || !rowCopy_) {
delete rowCopy_;
// may return NULL if can't give row copy
rowCopy_ = matrix_->reverseOrderedCopy();
rowCopyIsScaled = false;
} else {
rowCopyIsScaled = true;
}
}
#if 0
if (what == 63) {
int k = rowScale_ ? 1 : 0;
if (oldMatrix)
k += 2;
scale_times[k]++;
if ((scale_times[0] + scale_times[1] + scale_times[2] + scale_times[3]) % 1000 == 0)
printf("scale counts %d %d %d %d\n",
scale_times[0], scale_times[1], scale_times[2], scale_times[3]);
}
#endif
// do scaling if needed
if (!oldMatrix && scalingFlag_ < 0) {
if (scalingFlag_ < 0 && rowScale_) {
//if (handler_->logLevel()>0)
printf("How did we get scalingFlag_ %d and non NULL rowScale_? - switching off scaling\n",
scalingFlag_);
scalingFlag_ = 0;
}
delete [] rowScale_;
delete [] columnScale_;
rowScale_ = NULL;
columnScale_ = NULL;
}
inverseRowScale_ = NULL;
inverseColumnScale_ = NULL;
if (scalingFlag_ > 0 &&(specialOptions_ & 65536) != 0&&
rowScale_&&rowScale_==savedRowScale_)
rowScale_=NULL;
if (scalingFlag_ > 0 && !rowScale_) {
if ((specialOptions_ & 65536) != 0) {
assert (!rowScale_);
rowScale_ = savedRowScale_;
columnScale_ = savedColumnScale_;
// put back original
if (savedRowScale_) {
inverseRowScale_ = savedRowScale_ + maximumInternalRows_;
inverseColumnScale_ = savedColumnScale_ + maximumInternalColumns_;
CoinMemcpyN(savedRowScale_ + 2 * maximumInternalRows_,
numberRows2, savedRowScale_);
CoinMemcpyN(savedRowScale_ + 3 * maximumInternalRows_,
numberRows2, inverseRowScale_);
CoinMemcpyN(savedColumnScale_ + 2 * maximumColumns_,
numberColumns_, savedColumnScale_);
CoinMemcpyN(savedColumnScale_ + 3 * maximumColumns_,
numberColumns_, inverseColumnScale_);
}
}
if (matrix_->scale(this))
scalingFlag_ = -scalingFlag_; // not scaled after all
if (rowScale_ && automaticScale_) {
if (!savedRowScale_) {
inverseRowScale_ = rowScale_ + numberRows2;
inverseColumnScale_ = columnScale_ + numberColumns_;
}
// try automatic scaling
double smallestObj = 1.0e100;
double largestObj = 0.0;
double largestRhs = 0.0;
const double * obj = objective();
for (i = 0; i < numberColumns_; i++) {
double value = fabs(obj[i]);
value *= columnScale_[i];
if (value && columnLower_[i] != columnUpper_[i]) {
smallestObj = CoinMin(smallestObj, value);
largestObj = CoinMax(largestObj, value);
}
if (columnLower_[i] > 0.0 || columnUpper_[i] < 0.0) {
double scale = 1.0 * inverseColumnScale_[i];
//printf("%d %g %g %g %g\n",i,scale,lower_[i],upper_[i],largestRhs);
if (columnLower_[i] > 0)
largestRhs = CoinMax(largestRhs, columnLower_[i] * scale);
if (columnUpper_[i] < 0.0)
largestRhs = CoinMax(largestRhs, -columnUpper_[i] * scale);
}
}
for (i = 0; i < numberRows_; i++) {
if (rowLower_[i] > 0.0 || rowUpper_[i] < 0.0) {
double scale = rowScale_[i];
//printf("%d %g %g %g %g\n",i,scale,lower_[i],upper_[i],largestRhs);
if (rowLower_[i] > 0)
largestRhs = CoinMax(largestRhs, rowLower_[i] * scale);
if (rowUpper_[i] < 0.0)
largestRhs = CoinMax(largestRhs, -rowUpper_[i] * scale);
}
}
COIN_DETAIL_PRINT(printf("small obj %g, large %g - rhs %g\n", smallestObj, largestObj, largestRhs));
bool scalingDone = false;
// look at element range
double smallestNegative;
double largestNegative;
double smallestPositive;
double largestPositive;
matrix_->rangeOfElements(smallestNegative, largestNegative,
smallestPositive, largestPositive);
smallestPositive = CoinMin(fabs(smallestNegative), smallestPositive);
largestPositive = CoinMax(fabs(largestNegative), largestPositive);
if (largestObj) {
double ratio = largestObj / smallestObj;
double scale = 1.0;
if (ratio < 1.0e8) {
// reasonable
if (smallestObj < 1.0e-4) {
// may as well scale up
scalingDone = true;
scale = 1.0e-3 / smallestObj;
} else if (largestObj < 1.0e6 || (algorithm_ > 0 && largestObj < 1.0e-4 * infeasibilityCost_)) {
//done=true;
} else {
scalingDone = true;
if (algorithm_ < 0) {
scale = 1.0e6 / largestObj;
} else {
scale = CoinMax(1.0e6, 1.0e-4 * infeasibilityCost_) / largestObj;
}
}
} else if (ratio < 1.0e12) {
// not so good
if (smallestObj < 1.0e-7) {
// may as well scale up
scalingDone = true;
scale = 1.0e-6 / smallestObj;
} else if (largestObj < 1.0e7 || (algorithm_ > 0 && largestObj < 1.0e-3 * infeasibilityCost_)) {
//done=true;
} else {
scalingDone = true;
if (algorithm_ < 0) {
scale = 1.0e7 / largestObj;
} else {
scale = CoinMax(1.0e7, 1.0e-3 * infeasibilityCost_) / largestObj;
}
}
} else {
// Really nasty problem
if (smallestObj < 1.0e-8) {
// may as well scale up
scalingDone = true;
scale = 1.0e-7 / smallestObj;
largestObj *= scale;
}
if (largestObj < 1.0e7 || (algorithm_ > 0 && largestObj < 1.0e-3 * infeasibilityCost_)) {
//done=true;
} else {
scalingDone = true;
if (algorithm_ < 0) {
scale = 1.0e7 / largestObj;
} else {
scale = CoinMax(1.0e7, 1.0e-3 * infeasibilityCost_) / largestObj;
}
}
}
objectiveScale_ = scale;
}
if (largestRhs > 1.0e12) {
scalingDone = true;
rhsScale_ = 1.0e9 / largestRhs;
} else if (largestPositive > 1.0e-14 * smallestPositive && largestRhs > 1.0e6) {
scalingDone = true;
rhsScale_ = 1.0e6 / largestRhs;
} else {
rhsScale_ = 1.0;
}
if (scalingDone) {
handler_->message(CLP_RIM_SCALE, messages_)
<< objectiveScale_ << rhsScale_
<< CoinMessageEol;
}
}
} else if (makeRowCopy && scalingFlag_ > 0 && !rowCopyIsScaled) {
matrix_->scaleRowCopy(this);
}
if (rowScale_ && !savedRowScale_) {
inverseRowScale_ = rowScale_ + numberRows2;
inverseColumnScale_ = columnScale_ + numberColumns_;
}
// See if we can try for faster row copy
if (makeRowCopy && !oldMatrix) {
ClpPackedMatrix* clpMatrix =
dynamic_cast< ClpPackedMatrix*>(matrix_);
if (clpMatrix && numberThreads_)
clpMatrix->specialRowCopy(this, rowCopy_);
if (clpMatrix)
clpMatrix->specialColumnCopy(this);
}
}
if (what == 63) {
#if 0
{
x_gaps[0]++;
ClpPackedMatrix* clpMatrix =
dynamic_cast< ClpPackedMatrix*>(matrix_);
if (clpMatrix) {
if (!clpMatrix->getPackedMatrix()->hasGaps())
x_gaps[1]++;
if ((clpMatrix->flags() & 2) == 0)
x_gaps[3]++;
} else {
x_gaps[2]++;
}
if ((x_gaps[0] % 1000) == 0)
printf("create %d times, no gaps %d times - not clp %d times - flagged %d\n",
x_gaps[0], x_gaps[1], x_gaps[2], x_gaps[3]);
}
#endif
if (newArrays && (specialOptions_ & 65536) == 0) {
delete [] cost_;
cost_ = new double[2*numberTotal];
delete [] lower_;
delete [] upper_;
lower_ = new double[numberTotal];
upper_ = new double[numberTotal];
delete [] dj_;
dj_ = new double[numberTotal];
delete [] solution_;
solution_ = new double[numberTotal];
}
reducedCostWork_ = dj_;
rowReducedCost_ = dj_ + numberColumns_;
columnActivityWork_ = solution_;
rowActivityWork_ = solution_ + numberColumns_;
objectiveWork_ = cost_;
rowObjectiveWork_ = cost_ + numberColumns_;
rowLowerWork_ = lower_ + numberColumns_;
columnLowerWork_ = lower_;
rowUpperWork_ = upper_ + numberColumns_;
columnUpperWork_ = upper_;
}
if ((what & 4) != 0) {
double direction = optimizationDirection_ * objectiveScale_;
const double * obj = objective();
const double * rowScale = rowScale_;
const double * columnScale = columnScale_;
// and also scale by scale factors
if (rowScale) {
if (rowObjective_) {
for (i = 0; i < numberRows_; i++)
rowObjectiveWork_[i] = rowObjective_[i] * direction / rowScale[i];
} else {
memset(rowObjectiveWork_, 0, numberRows_ * sizeof(double));
}
// If scaled then do all columns later in one loop
if (what != 63) {
for (i = 0; i < numberColumns_; i++) {
CoinAssert(fabs(obj[i]) < 1.0e25);
objectiveWork_[i] = obj[i] * direction * columnScale[i];
}
}
} else {
if (rowObjective_) {
for (i = 0; i < numberRows_; i++)
rowObjectiveWork_[i] = rowObjective_[i] * direction;
} else {
memset(rowObjectiveWork_, 0, numberRows_ * sizeof(double));
}
for (i = 0; i < numberColumns_; i++) {
CoinAssert(fabs(obj[i]) < 1.0e25);
objectiveWork_[i] = obj[i] * direction;
}
}
}
if ((what & 1) != 0) {
const double * rowScale = rowScale_;
// clean up any mismatches on infinity
// and fix any variables with tiny gaps
double primalTolerance = dblParam_[ClpPrimalTolerance];
if(rowScale) {
// If scaled then do all columns later in one loop
if (what != 63) {
const double * inverseScale = inverseColumnScale_;
for (i = 0; i < numberColumns_; i++) {
double multiplier = rhsScale_ * inverseScale[i];
double lowerValue = columnLower_[i];
double upperValue = columnUpper_[i];
if (lowerValue > -1.0e20) {
columnLowerWork_[i] = lowerValue * multiplier;
if (upperValue >= 1.0e20) {
columnUpperWork_[i] = COIN_DBL_MAX;
} else {
columnUpperWork_[i] = upperValue * multiplier;
if (fabs(columnUpperWork_[i] - columnLowerWork_[i]) <= primalTolerance) {
if (columnLowerWork_[i] >= 0.0) {
columnUpperWork_[i] = columnLowerWork_[i];
} else if (columnUpperWork_[i] <= 0.0) {
columnLowerWork_[i] = columnUpperWork_[i];
} else {
columnUpperWork_[i] = 0.0;
columnLowerWork_[i] = 0.0;
}
}
}
} else if (upperValue < 1.0e20) {
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = upperValue * multiplier;
} else {
// free
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = COIN_DBL_MAX;
}
}
}
for (i = 0; i < numberRows_; i++) {
double multiplier = rhsScale_ * rowScale[i];
double lowerValue = rowLower_[i];
double upperValue = rowUpper_[i];
if (lowerValue > -1.0e20) {
rowLowerWork_[i] = lowerValue * multiplier;
if (upperValue >= 1.0e20) {
rowUpperWork_[i] = COIN_DBL_MAX;
} else {
rowUpperWork_[i] = upperValue * multiplier;
if (fabs(rowUpperWork_[i] - rowLowerWork_[i]) <= primalTolerance) {
if (rowLowerWork_[i] >= 0.0) {
rowUpperWork_[i] = rowLowerWork_[i];
} else if (rowUpperWork_[i] <= 0.0) {
rowLowerWork_[i] = rowUpperWork_[i];
} else {
rowUpperWork_[i] = 0.0;
rowLowerWork_[i] = 0.0;
}
}
}
} else if (upperValue < 1.0e20) {
rowLowerWork_[i] = -COIN_DBL_MAX;
rowUpperWork_[i] = upperValue * multiplier;
} else {
// free
rowLowerWork_[i] = -COIN_DBL_MAX;
rowUpperWork_[i] = COIN_DBL_MAX;
}
}
} else if (rhsScale_ != 1.0) {
for (i = 0; i < numberColumns_; i++) {
double lowerValue = columnLower_[i];
double upperValue = columnUpper_[i];
if (lowerValue > -1.0e20) {
columnLowerWork_[i] = lowerValue * rhsScale_;
if (upperValue >= 1.0e20) {
columnUpperWork_[i] = COIN_DBL_MAX;
} else {
columnUpperWork_[i] = upperValue * rhsScale_;
if (fabs(columnUpperWork_[i] - columnLowerWork_[i]) <= primalTolerance) {
if (columnLowerWork_[i] >= 0.0) {
columnUpperWork_[i] = columnLowerWork_[i];
} else if (columnUpperWork_[i] <= 0.0) {
columnLowerWork_[i] = columnUpperWork_[i];
} else {
columnUpperWork_[i] = 0.0;
columnLowerWork_[i] = 0.0;
}
}
}
} else if (upperValue < 1.0e20) {
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = upperValue * rhsScale_;
} else {
// free
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = COIN_DBL_MAX;
}
}
for (i = 0; i < numberRows_; i++) {
double lowerValue = rowLower_[i];
double upperValue = rowUpper_[i];
if (lowerValue > -1.0e20) {
rowLowerWork_[i] = lowerValue * rhsScale_;
if (upperValue >= 1.0e20) {
rowUpperWork_[i] = COIN_DBL_MAX;
} else {
rowUpperWork_[i] = upperValue * rhsScale_;
if (fabs(rowUpperWork_[i] - rowLowerWork_[i]) <= primalTolerance) {
if (rowLowerWork_[i] >= 0.0) {
rowUpperWork_[i] = rowLowerWork_[i];
} else if (rowUpperWork_[i] <= 0.0) {
rowLowerWork_[i] = rowUpperWork_[i];
} else {
rowUpperWork_[i] = 0.0;
rowLowerWork_[i] = 0.0;
}
}
}
} else if (upperValue < 1.0e20) {
rowLowerWork_[i] = -COIN_DBL_MAX;
rowUpperWork_[i] = upperValue * rhsScale_;
} else {
// free
rowLowerWork_[i] = -COIN_DBL_MAX;
rowUpperWork_[i] = COIN_DBL_MAX;
}
}
} else {
for (i = 0; i < numberColumns_; i++) {
double lowerValue = columnLower_[i];
double upperValue = columnUpper_[i];
if (lowerValue > -1.0e20) {
columnLowerWork_[i] = lowerValue;
if (upperValue >= 1.0e20) {
columnUpperWork_[i] = COIN_DBL_MAX;
} else {
columnUpperWork_[i] = upperValue;
if (fabs(columnUpperWork_[i] - columnLowerWork_[i]) <= primalTolerance) {
if (columnLowerWork_[i] >= 0.0) {
columnUpperWork_[i] = columnLowerWork_[i];
} else if (columnUpperWork_[i] <= 0.0) {
columnLowerWork_[i] = columnUpperWork_[i];
} else {
columnUpperWork_[i] = 0.0;
columnLowerWork_[i] = 0.0;
}
}
}
} else if (upperValue < 1.0e20) {
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = upperValue;
} else {
// free
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = COIN_DBL_MAX;
}
}
for (i = 0; i < numberRows_; i++) {
double lowerValue = rowLower_[i];
double upperValue = rowUpper_[i];
if (lowerValue > -1.0e20) {
rowLowerWork_[i] = lowerValue;
if (upperValue >= 1.0e20) {
rowUpperWork_[i] = COIN_DBL_MAX;
} else {
rowUpperWork_[i] = upperValue;
if (fabs(rowUpperWork_[i] - rowLowerWork_[i]) <= primalTolerance) {
if (rowLowerWork_[i] >= 0.0) {
rowUpperWork_[i] = rowLowerWork_[i];
} else if (rowUpperWork_[i] <= 0.0) {
rowLowerWork_[i] = rowUpperWork_[i];
} else {
rowUpperWork_[i] = 0.0;
rowLowerWork_[i] = 0.0;
}
}
}
} else if (upperValue < 1.0e20) {
rowLowerWork_[i] = -COIN_DBL_MAX;
rowUpperWork_[i] = upperValue;
} else {
// free
rowLowerWork_[i] = -COIN_DBL_MAX;
rowUpperWork_[i] = COIN_DBL_MAX;
}
}
}
}
if (what == 63) {
// move information to work arrays
double direction = optimizationDirection_;
// direction is actually scale out not scale in
if (direction)
direction = 1.0 / direction;
if (direction != 1.0) {
// reverse all dual signs
for (i = 0; i < numberColumns_; i++)
reducedCost_[i] *= direction;
for (i = 0; i < numberRows_; i++)
dual_[i] *= direction;
}
for (i = 0; i < numberRows_ + numberColumns_; i++) {
setFakeBound(i, noFake);
}
if (rowScale_) {
const double * obj = objective();
double direction = optimizationDirection_ * objectiveScale_;
// clean up any mismatches on infinity
// and fix any variables with tiny gaps
double primalTolerance = dblParam_[ClpPrimalTolerance];
// on entry
const double * inverseScale = inverseColumnScale_;
for (i = 0; i < numberColumns_; i++) {
CoinAssert(fabs(obj[i]) < 1.0e25);
double scaleFactor = columnScale_[i];
double multiplier = rhsScale_ * inverseScale[i];
scaleFactor *= direction;
objectiveWork_[i] = obj[i] * scaleFactor;
reducedCostWork_[i] = reducedCost_[i] * scaleFactor;
double lowerValue = columnLower_[i];
double upperValue = columnUpper_[i];
if (lowerValue > -1.0e20) {
columnLowerWork_[i] = lowerValue * multiplier;
if (upperValue >= 1.0e20) {
columnUpperWork_[i] = COIN_DBL_MAX;
} else {
columnUpperWork_[i] = upperValue * multiplier;
if (fabs(columnUpperWork_[i] - columnLowerWork_[i]) <= primalTolerance) {
if (columnLowerWork_[i] >= 0.0) {
columnUpperWork_[i] = columnLowerWork_[i];
} else if (columnUpperWork_[i] <= 0.0) {
columnLowerWork_[i] = columnUpperWork_[i];
} else {
columnUpperWork_[i] = 0.0;
columnLowerWork_[i] = 0.0;
}
}
}
} else if (upperValue < 1.0e20) {
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = upperValue * multiplier;
} else {
// free
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = COIN_DBL_MAX;
}
double value = columnActivity_[i] * multiplier;
if (fabs(value) > 1.0e20) {
//printf("bad value of %g for column %d\n",value,i);
setColumnStatus(i, superBasic);
if (columnUpperWork_[i] < 0.0) {
value = columnUpperWork_[i];
} else if (columnLowerWork_[i] > 0.0) {
value = columnLowerWork_[i];
} else {
value = 0.0;
}
}
columnActivityWork_[i] = value;
}
inverseScale = inverseRowScale_;
for (i = 0; i < numberRows_; i++) {
dual_[i] *= inverseScale[i];
dual_[i] *= objectiveScale_;
rowReducedCost_[i] = dual_[i];
double multiplier = rhsScale_ * rowScale_[i];
double value = rowActivity_[i] * multiplier;
if (fabs(value) > 1.0e20) {
//printf("bad value of %g for row %d\n",value,i);
setRowStatus(i, superBasic);
if (rowUpperWork_[i] < 0.0) {
value = rowUpperWork_[i];
} else if (rowLowerWork_[i] > 0.0) {
value = rowLowerWork_[i];
} else {
value = 0.0;
}
}
rowActivityWork_[i] = value;
}
} else if (objectiveScale_ != 1.0 || rhsScale_ != 1.0) {
// on entry
for (i = 0; i < numberColumns_; i++) {
double value = columnActivity_[i];
value *= rhsScale_;
if (fabs(value) > 1.0e20) {
//printf("bad value of %g for column %d\n",value,i);
setColumnStatus(i, superBasic);
if (columnUpperWork_[i] < 0.0) {
value = columnUpperWork_[i];
} else if (columnLowerWork_[i] > 0.0) {
value = columnLowerWork_[i];
} else {
value = 0.0;
}
}
columnActivityWork_[i] = value;
reducedCostWork_[i] = reducedCost_[i] * objectiveScale_;
}
for (i = 0; i < numberRows_; i++) {
double value = rowActivity_[i];
value *= rhsScale_;
if (fabs(value) > 1.0e20) {
//printf("bad value of %g for row %d\n",value,i);
setRowStatus(i, superBasic);
if (rowUpperWork_[i] < 0.0) {
value = rowUpperWork_[i];
} else if (rowLowerWork_[i] > 0.0) {
value = rowLowerWork_[i];
} else {
value = 0.0;
}
}
rowActivityWork_[i] = value;
dual_[i] *= objectiveScale_;
rowReducedCost_[i] = dual_[i];
}
} else {
// on entry
for (i = 0; i < numberColumns_; i++) {
double value = columnActivity_[i];
if (fabs(value) > 1.0e20) {
//printf("bad value of %g for column %d\n",value,i);
setColumnStatus(i, superBasic);
if (columnUpperWork_[i] < 0.0) {
value = columnUpperWork_[i];
} else if (columnLowerWork_[i] > 0.0) {
value = columnLowerWork_[i];
} else {
value = 0.0;
}
}
columnActivityWork_[i] = value;
reducedCostWork_[i] = reducedCost_[i];
}
for (i = 0; i < numberRows_; i++) {
double value = rowActivity_[i];
if (fabs(value) > 1.0e20) {
//printf("bad value of %g for row %d\n",value,i);
setRowStatus(i, superBasic);
if (rowUpperWork_[i] < 0.0) {
value = rowUpperWork_[i];
} else if (rowLowerWork_[i] > 0.0) {
value = rowLowerWork_[i];
} else {
value = 0.0;
}
}
rowActivityWork_[i] = value;
rowReducedCost_[i] = dual_[i];
}
}
}
if (what == 63 && doSanityCheck) {
// check rim of problem okay
if (!sanityCheck())
goodMatrix = false;
}
// we need to treat matrix as if each element by rowScaleIn and columnScaleout??
// maybe we need to move scales to SimplexModel for factorization?
if ((what == 63 && !pivotVariable_) || (newArrays && !keepPivots)) {
delete [] pivotVariable_;
pivotVariable_ = new int[numberRows2];
for (int i = 0; i < numberRows2; i++)
pivotVariable_[i] = -1;
} else if (what == 63 && !keepPivots) {
// just reset
for (int i = 0; i < numberRows2; i++)
pivotVariable_[i] = -1;
} else if (what == 63) {
// check pivots
for (int i = 0; i < numberRows2; i++) {
int iSequence = pivotVariable_[i];
if (iSequence<0||(iSequence < numberRows_ + numberColumns_ &&
getStatus(iSequence) != basic)) {
keepPivots = false;
break;
}
}
if (!keepPivots) {
// reset
for (int i = 0; i < numberRows2; i++)
pivotVariable_[i] = -1;
} else {
// clean
for (int i = 0; i < numberColumns_ + numberRows_; i++) {
Status status = getStatus(i);
if (status != basic) {
if (upper_[i] == lower_[i]) {
setStatus(i, isFixed);
solution_[i] = lower_[i];
} else if (status == atLowerBound) {
if (lower_[i] > -1.0e20) {
solution_[i] = lower_[i];
} else {
//printf("seq %d at lower of %g\n",i,lower_[i]);
if (upper_[i] < 1.0e20) {
solution_[i] = upper_[i];
setStatus(i, atUpperBound);
} else {
setStatus(i, isFree);
}
}
} else if (status == atUpperBound) {
if (upper_[i] < 1.0e20) {
solution_[i] = upper_[i];
} else {
//printf("seq %d at upper of %g\n",i,upper_[i]);
if (lower_[i] > -1.0e20) {
solution_[i] = lower_[i];
setStatus(i, atLowerBound);
} else {
setStatus(i, isFree);
}
}
} else if (status == isFixed && upper_[i] > lower_[i]) {
// was fixed - not now
if (solution_[i] <= lower_[i]) {
setStatus(i, atLowerBound);
} else if (solution_[i] >= upper_[i]) {
setStatus(i, atUpperBound);
} else {
setStatus(i, superBasic);
}
}
}
}
}
}
if (what == 63) {
if (newArrays) {
// get some arrays
int iRow, iColumn;
// these are "indexed" arrays so we always know where nonzeros are
/**********************************************************
rowArray_[3] is long enough for rows+columns
rowArray_[1] is long enough for max(rows,columns)
*********************************************************/
for (iRow = 0; iRow < 4; iRow++) {
int length = numberRows2 + factorization_->maximumPivots();
if (iRow == 3 || objective_->type() > 1)
length += numberColumns_;
else if (iRow == 1)
length = CoinMax(length, numberColumns_);
if ((specialOptions_ & 65536) == 0 || !rowArray_[iRow]) {
delete rowArray_[iRow];
rowArray_[iRow] = new CoinIndexedVector();
}
rowArray_[iRow]->reserve(length);
}
for (iColumn = 0; iColumn < 2; iColumn++) {
if ((specialOptions_ & 65536) == 0 || !columnArray_[iColumn]) {
delete columnArray_[iColumn];
columnArray_[iColumn] = new CoinIndexedVector();
}
columnArray_[iColumn]->reserve(numberColumns_+numberRows2);
}
} else {
int iRow, iColumn;
for (iRow = 0; iRow < 4; iRow++) {
int length = numberRows2 + factorization_->maximumPivots();
if (iRow == 3 || objective_->type() > 1)
length += numberColumns_;
if(rowArray_[iRow]->capacity() >= length) {
rowArray_[iRow]->clear();
} else {
// model size or maxinv changed
rowArray_[iRow]->reserve(length);
}
#ifndef NDEBUG
rowArray_[iRow]->checkClear();
#endif
}
for (iColumn = 0; iColumn < 2; iColumn++) {
int length = numberColumns_;
if (iColumn)
length = CoinMax(numberRows2, numberColumns_);
if(columnArray_[iColumn]->capacity() >= length) {
columnArray_[iColumn]->clear();
} else {
// model size or maxinv changed
columnArray_[iColumn]->reserve(length);
}
#ifndef NDEBUG
columnArray_[iColumn]->checkClear();
#endif
}
}
}
if (problemStatus_ == 10) {
problemStatus_ = -1;
handler_->setLogLevel(saveLevel); // switch back messages
}
if ((what & 5) != 0)
matrix_->generalExpanded(this, 9, what); // update costs and bounds if necessary
if (goodMatrix && (specialOptions_ & 65536) != 0) {
int save = maximumColumns_ + maximumRows_;
CoinMemcpyN(cost_, numberTotal, cost_ + save);
CoinMemcpyN(lower_, numberTotal, lower_ + save);
CoinMemcpyN(upper_, numberTotal, upper_ + save);
CoinMemcpyN(dj_, numberTotal, dj_ + save);
CoinMemcpyN(solution_, numberTotal, solution_ + save);
if (rowScale_ && !savedRowScale_) {
double * temp;
temp = new double [4*maximumRows_];
CoinFillN(temp, 4 * maximumRows_, 1.0);
CoinMemcpyN(rowScale_, numberRows2, temp);
CoinMemcpyN(rowScale_ + numberRows2, numberRows2, temp + maximumRows_);
CoinMemcpyN(rowScale_, numberRows2, temp + 2 * maximumRows_);
CoinMemcpyN(rowScale_ + numberRows2, numberRows2, temp + 3 * maximumRows_);
delete [] rowScale_;
savedRowScale_ = temp;
rowScale_ = savedRowScale_;
inverseRowScale_ = savedRowScale_ + maximumInternalRows_;
temp = new double [4*maximumColumns_];
CoinFillN(temp, 4 * maximumColumns_, 1.0);
CoinMemcpyN(columnScale_, numberColumns_, temp);
CoinMemcpyN(columnScale_ + numberColumns_, numberColumns_, temp + maximumColumns_);
CoinMemcpyN(columnScale_, numberColumns_, temp + 2 * maximumColumns_);
CoinMemcpyN(columnScale_ + numberColumns_, numberColumns_, temp + 3 * maximumColumns_);
delete [] columnScale_;
savedColumnScale_ = temp;
columnScale_ = savedColumnScale_;
inverseColumnScale_ = savedColumnScale_ + maximumInternalColumns_;
}
}
#ifdef CLP_USER_DRIVEN
eventHandler_->event(ClpEventHandler::endOfCreateRim);
#endif
return goodMatrix;
}
// Does rows and columns
void
ClpSimplex::createRim1(bool initial)
{
int i;
int numberRows2 = numberRows_ + numberExtraRows_;
int numberTotal = numberRows2 + numberColumns_;
if ((specialOptions_ & 65536) != 0 && true) {
assert (!initial);
int save = maximumColumns_ + maximumRows_;
CoinMemcpyN(lower_ + save, numberTotal, lower_);
CoinMemcpyN(upper_ + save, numberTotal, upper_);
return;
}
const double * rowScale = rowScale_;
// clean up any mismatches on infinity
// and fix any variables with tiny gaps
double primalTolerance = dblParam_[ClpPrimalTolerance];
if(rowScale) {
// If scaled then do all columns later in one loop
if (!initial) {
const double * inverseScale = inverseColumnScale_;
for (i = 0; i < numberColumns_; i++) {
double multiplier = rhsScale_ * inverseScale[i];
double lowerValue = columnLower_[i];
double upperValue = columnUpper_[i];
if (lowerValue > -1.0e20) {
columnLowerWork_[i] = lowerValue * multiplier;
if (upperValue >= 1.0e20) {
columnUpperWork_[i] = COIN_DBL_MAX;
} else {
columnUpperWork_[i] = upperValue * multiplier;
if (fabs(columnUpperWork_[i] - columnLowerWork_[i]) <= primalTolerance) {
if (columnLowerWork_[i] >= 0.0) {
columnUpperWork_[i] = columnLowerWork_[i];
} else if (columnUpperWork_[i] <= 0.0) {
columnLowerWork_[i] = columnUpperWork_[i];
} else {
columnUpperWork_[i] = 0.0;
columnLowerWork_[i] = 0.0;
}
}
}
} else if (upperValue < 1.0e20) {
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = upperValue * multiplier;
} else {
// free
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = COIN_DBL_MAX;
}
}
}
for (i = 0; i < numberRows_; i++) {
double multiplier = rhsScale_ * rowScale[i];
double lowerValue = rowLower_[i];
double upperValue = rowUpper_[i];
if (lowerValue > -1.0e20) {
rowLowerWork_[i] = lowerValue * multiplier;
if (upperValue >= 1.0e20) {
rowUpperWork_[i] = COIN_DBL_MAX;
} else {
rowUpperWork_[i] = upperValue * multiplier;
if (fabs(rowUpperWork_[i] - rowLowerWork_[i]) <= primalTolerance) {
if (rowLowerWork_[i] >= 0.0) {
rowUpperWork_[i] = rowLowerWork_[i];
} else if (rowUpperWork_[i] <= 0.0) {
rowLowerWork_[i] = rowUpperWork_[i];
} else {
rowUpperWork_[i] = 0.0;
rowLowerWork_[i] = 0.0;
}
}
}
} else if (upperValue < 1.0e20) {
rowLowerWork_[i] = -COIN_DBL_MAX;
rowUpperWork_[i] = upperValue * multiplier;
} else {
// free
rowLowerWork_[i] = -COIN_DBL_MAX;
rowUpperWork_[i] = COIN_DBL_MAX;
}
}
} else if (rhsScale_ != 1.0) {
for (i = 0; i < numberColumns_; i++) {
double lowerValue = columnLower_[i];
double upperValue = columnUpper_[i];
if (lowerValue > -1.0e20) {
columnLowerWork_[i] = lowerValue * rhsScale_;
if (upperValue >= 1.0e20) {
columnUpperWork_[i] = COIN_DBL_MAX;
} else {
columnUpperWork_[i] = upperValue * rhsScale_;
if (fabs(columnUpperWork_[i] - columnLowerWork_[i]) <= primalTolerance) {
if (columnLowerWork_[i] >= 0.0) {
columnUpperWork_[i] = columnLowerWork_[i];
} else if (columnUpperWork_[i] <= 0.0) {
columnLowerWork_[i] = columnUpperWork_[i];
} else {
columnUpperWork_[i] = 0.0;
columnLowerWork_[i] = 0.0;
}
}
}
} else if (upperValue < 1.0e20) {
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = upperValue * rhsScale_;
} else {
// free
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = COIN_DBL_MAX;
}
}
for (i = 0; i < numberRows_; i++) {
double lowerValue = rowLower_[i];
double upperValue = rowUpper_[i];
if (lowerValue > -1.0e20) {
rowLowerWork_[i] = lowerValue * rhsScale_;
if (upperValue >= 1.0e20) {
rowUpperWork_[i] = COIN_DBL_MAX;
} else {
rowUpperWork_[i] = upperValue * rhsScale_;
if (fabs(rowUpperWork_[i] - rowLowerWork_[i]) <= primalTolerance) {
if (rowLowerWork_[i] >= 0.0) {
rowUpperWork_[i] = rowLowerWork_[i];
} else if (rowUpperWork_[i] <= 0.0) {
rowLowerWork_[i] = rowUpperWork_[i];
} else {
rowUpperWork_[i] = 0.0;
rowLowerWork_[i] = 0.0;
}
}
}
} else if (upperValue < 1.0e20) {
rowLowerWork_[i] = -COIN_DBL_MAX;
rowUpperWork_[i] = upperValue * rhsScale_;
} else {
// free
rowLowerWork_[i] = -COIN_DBL_MAX;
rowUpperWork_[i] = COIN_DBL_MAX;
}
}
} else {
for (i = 0; i < numberColumns_; i++) {
double lowerValue = columnLower_[i];
double upperValue = columnUpper_[i];
if (lowerValue > -1.0e20) {
columnLowerWork_[i] = lowerValue;
if (upperValue >= 1.0e20) {
columnUpperWork_[i] = COIN_DBL_MAX;
} else {
columnUpperWork_[i] = upperValue;
if (fabs(columnUpperWork_[i] - columnLowerWork_[i]) <= primalTolerance) {
if (columnLowerWork_[i] >= 0.0) {
columnUpperWork_[i] = columnLowerWork_[i];
} else if (columnUpperWork_[i] <= 0.0) {
columnLowerWork_[i] = columnUpperWork_[i];
} else {
columnUpperWork_[i] = 0.0;
columnLowerWork_[i] = 0.0;
}
}
}
} else if (upperValue < 1.0e20) {
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = upperValue;
} else {
// free
columnLowerWork_[i] = -COIN_DBL_MAX;
columnUpperWork_[i] = COIN_DBL_MAX;
}
}
for (i = 0; i < numberRows_; i++) {
double lowerValue = rowLower_[i];
double upperValue = rowUpper_[i];
if (lowerValue > -1.0e20) {
rowLowerWork_[i] = lowerValue;
if (upperValue >= 1.0e20) {
rowUpperWork_[i] = COIN_DBL_MAX;
} else {
rowUpperWork_[i] = upperValue;
if (fabs(rowUpperWork_[i] - rowLowerWork_[i]) <= primalTolerance) {
if (rowLowerWork_[i] >= 0.0) {
rowUpperWork_[i] = rowLowerWork_[i];
} else if (rowUpperWork_[i] <= 0.0) {
rowLowerWork_[i] = rowUpperWork_[i];
} else {
rowUpperWork_[i] = 0.0;
rowLowerWork_[i] = 0.0;
}
}
}
} else if (upperValue < 1.0e20) {
rowLowerWork_[i] = -COIN_DBL_MAX;
rowUpperWork_[i] = upperValue;
} else {
// free
rowLowerWork_[i] = -COIN_DBL_MAX;
rowUpperWork_[i] = COIN_DBL_MAX;
}
}
}
#ifndef NDEBUG
if ((specialOptions_ & 65536) != 0 && false) {
assert (!initial);
int save = maximumColumns_ + maximumRows_;
for (int i = 0; i < numberTotal; i++) {
assert (fabs(lower_[i] - lower_[i+save]) < 1.0e-5);
assert (fabs(upper_[i] - upper_[i+save]) < 1.0e-5);
}
}
#endif
}
// Does objective
void
ClpSimplex::createRim4(bool initial)
{
int i;
int numberRows2 = numberRows_ + numberExtraRows_;
int numberTotal = numberRows2 + numberColumns_;
if ((specialOptions_ & 65536) != 0 && true) {
assert (!initial);
int save = maximumColumns_ + maximumRows_;
CoinMemcpyN(cost_ + save, numberTotal, cost_);
return;
}
double direction = optimizationDirection_ * objectiveScale_;
const double * obj = objective();
const double * rowScale = rowScale_;
const double * columnScale = columnScale_;
// and also scale by scale factors
if (rowScale) {
if (rowObjective_) {
for (i = 0; i < numberRows_; i++)
rowObjectiveWork_[i] = rowObjective_[i] * direction / rowScale[i];
} else {
memset(rowObjectiveWork_, 0, numberRows_ * sizeof(double));
}
// If scaled then do all columns later in one loop
if (!initial) {
for (i = 0; i < numberColumns_; i++) {
CoinAssert(fabs(obj[i]) < 1.0e25);
objectiveWork_[i] = obj[i] * direction * columnScale[i];
}
}
} else {
if (rowObjective_) {
for (i = 0; i < numberRows_; i++)
rowObjectiveWork_[i] = rowObjective_[i] * direction;
} else {
memset(rowObjectiveWork_, 0, numberRows_ * sizeof(double));
}
for (i = 0; i < numberColumns_; i++) {
CoinAssert(fabs(obj[i]) < 1.0e25);
objectiveWork_[i] = obj[i] * direction;
}
}
}
// Does rows and columns and objective
void
ClpSimplex::createRim5(bool initial)
{
createRim4(initial);
createRim1(initial);
}
void
ClpSimplex::deleteRim(int getRidOfFactorizationData)
{
// Just possible empty problem
int numberRows = numberRows_;
int numberColumns = numberColumns_;
if (!numberRows || !numberColumns) {
numberRows = 0;
if (objective_->type() < 2)
numberColumns = 0;
}
int i;
if (problemStatus_ != 1 && problemStatus_ != 2) {
delete [] ray_;
ray_ = NULL;
}
// set upperOut_ to furthest away from bound so can use in dual for dualBound_
upperOut_ = 1.0;
#if 0
{
int nBad = 0;
for (i = 0; i < numberColumns; i++) {
if (lower_[i] == upper_[i] && getColumnStatus(i) == basic)
nBad++;
}
if (nBad)
printf("yy %d basic fixed\n", nBad);
}
#endif
// ray may be null if in branch and bound
if (rowScale_) {
// Collect infeasibilities
int numberPrimalScaled = 0;
int numberPrimalUnscaled = 0;
int numberDualScaled = 0;
int numberDualUnscaled = 0;
double scaleC = 1.0 / objectiveScale_;
double scaleR = 1.0 / rhsScale_;
const double * inverseScale = inverseColumnScale_;
for (i = 0; i < numberColumns; i++) {
double scaleFactor = columnScale_[i];
double valueScaled = columnActivityWork_[i];
double lowerScaled = columnLowerWork_[i];
double upperScaled = columnUpperWork_[i];
if (lowerScaled > -1.0e20 || upperScaled < 1.0e20) {
if (valueScaled < lowerScaled - primalTolerance_ ||
valueScaled > upperScaled + primalTolerance_)
numberPrimalScaled++;
else
upperOut_ = CoinMax(upperOut_, CoinMin(valueScaled - lowerScaled, upperScaled - valueScaled));
}
columnActivity_[i] = valueScaled * scaleFactor * scaleR;
double value = columnActivity_[i];
if (value < columnLower_[i] - primalTolerance_)
numberPrimalUnscaled++;
else if (value > columnUpper_[i] + primalTolerance_)
numberPrimalUnscaled++;
double valueScaledDual = reducedCostWork_[i];
if (valueScaled > columnLowerWork_[i] + primalTolerance_ && valueScaledDual > dualTolerance_)
numberDualScaled++;
if (valueScaled < columnUpperWork_[i] - primalTolerance_ && valueScaledDual < -dualTolerance_)
numberDualScaled++;
reducedCost_[i] = (valueScaledDual * scaleC) * inverseScale[i];
double valueDual = reducedCost_[i];
if (value > columnLower_[i] + primalTolerance_ && valueDual > dualTolerance_)
numberDualUnscaled++;
if (value < columnUpper_[i] - primalTolerance_ && valueDual < -dualTolerance_)
numberDualUnscaled++;
}
inverseScale = inverseRowScale_;
for (i = 0; i < numberRows; i++) {
double scaleFactor = rowScale_[i];
double valueScaled = rowActivityWork_[i];
double lowerScaled = rowLowerWork_[i];
double upperScaled = rowUpperWork_[i];
if (lowerScaled > -1.0e20 || upperScaled < 1.0e20) {
if (valueScaled < lowerScaled - primalTolerance_ ||
valueScaled > upperScaled + primalTolerance_)
numberPrimalScaled++;
else
upperOut_ = CoinMax(upperOut_, CoinMin(valueScaled - lowerScaled, upperScaled - valueScaled));
}
rowActivity_[i] = (valueScaled * scaleR) * inverseScale[i];
double value = rowActivity_[i];
if (value < rowLower_[i] - primalTolerance_)
numberPrimalUnscaled++;
else if (value > rowUpper_[i] + primalTolerance_)
numberPrimalUnscaled++;
double valueScaledDual = dual_[i] + rowObjectiveWork_[i];;
if (valueScaled > rowLowerWork_[i] + primalTolerance_ && valueScaledDual > dualTolerance_)
numberDualScaled++;
if (valueScaled < rowUpperWork_[i] - primalTolerance_ && valueScaledDual < -dualTolerance_)
numberDualScaled++;
dual_[i] *= scaleFactor * scaleC;
double valueDual = dual_[i];
if (rowObjective_)
valueDual += rowObjective_[i];
if (value > rowLower_[i] + primalTolerance_ && valueDual > dualTolerance_)
numberDualUnscaled++;
if (value < rowUpper_[i] - primalTolerance_ && valueDual < -dualTolerance_)
numberDualUnscaled++;
}
if (!problemStatus_ && !secondaryStatus_) {
// See if we need to set secondary status
if (numberPrimalUnscaled) {
if (numberDualUnscaled)
secondaryStatus_ = 4;
else
secondaryStatus_ = 2;
} else {
if (numberDualUnscaled)
secondaryStatus_ = 3;
}
}
if (problemStatus_ == 2 && ray_) {
for (i = 0; i < numberColumns; i++) {
ray_[i] *= columnScale_[i];
}
} else if (problemStatus_ == 1 && ray_) {
for (i = 0; i < numberRows; i++) {
ray_[i] *= rowScale_[i];
}
}
} else if (rhsScale_ != 1.0 || objectiveScale_ != 1.0) {
// Collect infeasibilities
int numberPrimalScaled = 0;
int numberPrimalUnscaled = 0;
int numberDualScaled = 0;
int numberDualUnscaled = 0;
double scaleC = 1.0 / objectiveScale_;
double scaleR = 1.0 / rhsScale_;
for (i = 0; i < numberColumns; i++) {
double valueScaled = columnActivityWork_[i];
double lowerScaled = columnLowerWork_[i];
double upperScaled = columnUpperWork_[i];
if (lowerScaled > -1.0e20 || upperScaled < 1.0e20) {
if (valueScaled < lowerScaled - primalTolerance_ ||
valueScaled > upperScaled + primalTolerance_)
numberPrimalScaled++;
else
upperOut_ = CoinMax(upperOut_, CoinMin(valueScaled - lowerScaled, upperScaled - valueScaled));
}
columnActivity_[i] = valueScaled * scaleR;
double value = columnActivity_[i];
if (value < columnLower_[i] - primalTolerance_)
numberPrimalUnscaled++;
else if (value > columnUpper_[i] + primalTolerance_)
numberPrimalUnscaled++;
double valueScaledDual = reducedCostWork_[i];
if (valueScaled > columnLowerWork_[i] + primalTolerance_ && valueScaledDual > dualTolerance_)
numberDualScaled++;
if (valueScaled < columnUpperWork_[i] - primalTolerance_ && valueScaledDual < -dualTolerance_)
numberDualScaled++;
reducedCost_[i] = valueScaledDual * scaleC;
double valueDual = reducedCost_[i];
if (value > columnLower_[i] + primalTolerance_ && valueDual > dualTolerance_)
numberDualUnscaled++;
if (value < columnUpper_[i] - primalTolerance_ && valueDual < -dualTolerance_)
numberDualUnscaled++;
}
for (i = 0; i < numberRows; i++) {
double valueScaled = rowActivityWork_[i];
double lowerScaled = rowLowerWork_[i];
double upperScaled = rowUpperWork_[i];
if (lowerScaled > -1.0e20 || upperScaled < 1.0e20) {
if (valueScaled < lowerScaled - primalTolerance_ ||
valueScaled > upperScaled + primalTolerance_)
numberPrimalScaled++;
else
upperOut_ = CoinMax(upperOut_, CoinMin(valueScaled - lowerScaled, upperScaled - valueScaled));
}
rowActivity_[i] = valueScaled * scaleR;
double value = rowActivity_[i];
if (value < rowLower_[i] - primalTolerance_)
numberPrimalUnscaled++;
else if (value > rowUpper_[i] + primalTolerance_)
numberPrimalUnscaled++;
double valueScaledDual = dual_[i] + rowObjectiveWork_[i];;
if (valueScaled > rowLowerWork_[i] + primalTolerance_ && valueScaledDual > dualTolerance_)
numberDualScaled++;
if (valueScaled < rowUpperWork_[i] - primalTolerance_ && valueScaledDual < -dualTolerance_)
numberDualScaled++;
dual_[i] *= scaleC;
double valueDual = dual_[i];
if (rowObjective_)
valueDual += rowObjective_[i];
if (value > rowLower_[i] + primalTolerance_ && valueDual > dualTolerance_)
numberDualUnscaled++;
if (value < rowUpper_[i] - primalTolerance_ && valueDual < -dualTolerance_)
numberDualUnscaled++;
}
if (!problemStatus_ && !secondaryStatus_) {
// See if we need to set secondary status
if (numberPrimalUnscaled) {
if (numberDualUnscaled)
secondaryStatus_ = 4;
else
secondaryStatus_ = 2;
} else {
if (numberDualUnscaled)
secondaryStatus_ = 3;
}
}
} else {
if (columnActivityWork_) {
for (i = 0; i < numberColumns; i++) {
double value = columnActivityWork_[i];
double lower = columnLowerWork_[i];
double upper = columnUpperWork_[i];
if (lower > -1.0e20 || upper < 1.0e20) {
if (value > lower && value < upper)
upperOut_ = CoinMax(upperOut_, CoinMin(value - lower, upper - value));
}
columnActivity_[i] = columnActivityWork_[i];
reducedCost_[i] = reducedCostWork_[i];
}
for (i = 0; i < numberRows; i++) {
double value = rowActivityWork_[i];
double lower = rowLowerWork_[i];
double upper = rowUpperWork_[i];
if (lower > -1.0e20 || upper < 1.0e20) {
if (value > lower && value < upper)
upperOut_ = CoinMax(upperOut_, CoinMin(value - lower, upper - value));
}
rowActivity_[i] = rowActivityWork_[i];
}
}
}
// switch off scalefactor if auto
if (automaticScale_) {
rhsScale_ = 1.0;
objectiveScale_ = 1.0;
}
if (optimizationDirection_ != 1.0) {
// and modify all dual signs
for (i = 0; i < numberColumns; i++)
reducedCost_[i] *= optimizationDirection_;
for (i = 0; i < numberRows; i++)
dual_[i] *= optimizationDirection_;
}
// scaling may have been turned off
scalingFlag_ = abs(scalingFlag_);
if(getRidOfFactorizationData > 0) {
gutsOfDelete(getRidOfFactorizationData + 1);
} else {
// at least get rid of nonLinearCost_
delete nonLinearCost_;
nonLinearCost_ = NULL;
}
if (!rowObjective_ && problemStatus_ == 0 && objective_->type() == 1 &&
numberRows && numberColumns) {
// Redo objective value
double objectiveValue = 0.0;
const double * cost = objective();
for (int i = 0; i < numberColumns; i++) {
double value = columnActivity_[i];
objectiveValue += value * cost[i];
}
//if (fabs(objectiveValue_ -objectiveValue*optimizationDirection())>1.0e-5)
//printf("old obj %g new %g\n",objectiveValue_, objectiveValue*optimizationDirection());
objectiveValue_ = objectiveValue * optimizationDirection();
}
// get rid of data
matrix_->generalExpanded(this, 13, scalingFlag_);
}
void
ClpSimplex::setDualBound(double value)
{
if (value > 0.0)
dualBound_ = value;
}
void
ClpSimplex::setInfeasibilityCost(double value)
{
if (value > 0.0)
infeasibilityCost_ = value;
}
void ClpSimplex::setNumberRefinements( int value)
{
if (value >= 0 && value < 10)
numberRefinements_ = value;
}
// Sets row pivot choice algorithm in dual
void
ClpSimplex::setDualRowPivotAlgorithm(ClpDualRowPivot & choice)
{
delete dualRowPivot_;
dualRowPivot_ = choice.clone(true);
dualRowPivot_->setModel(this);
}
// Sets row pivot choice algorithm in dual
void
ClpSimplex::setPrimalColumnPivotAlgorithm(ClpPrimalColumnPivot & choice)
{
delete primalColumnPivot_;
primalColumnPivot_ = choice.clone(true);
primalColumnPivot_->setModel(this);
}
void
ClpSimplex::setFactorization( ClpFactorization & factorization)
{
if (factorization_)
factorization_->setFactorization(factorization);
else
factorization_ = new ClpFactorization(factorization,
numberRows_);
}
// Swaps factorization
ClpFactorization *
ClpSimplex::swapFactorization( ClpFactorization * factorization)
{
ClpFactorization * swap = factorization_;
factorization_ = factorization;
return swap;
}
// Copies in factorization to existing one
void
ClpSimplex::copyFactorization( ClpFactorization & factorization)
{
*factorization_ = factorization;
}
/* Perturbation:
-50 to +50 - perturb by this power of ten (-6 sounds good)
100 - auto perturb if takes too long (1.0e-6 largest nonzero)
101 - we are perturbed
102 - don't try perturbing again
default is 100
*/
void
ClpSimplex::setPerturbation(int value)
{
if(value <= 100 && value >= -1000) {
perturbation_ = value;
}
}
// Sparsity on or off
bool
ClpSimplex::sparseFactorization() const
{
return factorization_->sparseThreshold() != 0;
}
void
ClpSimplex::setSparseFactorization(bool value)
{
if (value) {
if (!factorization_->sparseThreshold())
factorization_->goSparse();
} else {
factorization_->sparseThreshold(0);
}
}
void checkCorrect(ClpSimplex * /*model*/, int iRow,
const double * element, const int * rowStart, const int * rowLength,
const int * column,
const double * columnLower_, const double * columnUpper_,
int /*infiniteUpperC*/,
int /*infiniteLowerC*/,
double &maximumUpC,
double &maximumDownC)
{
int infiniteUpper = 0;
int infiniteLower = 0;
double maximumUp = 0.0;
double maximumDown = 0.0;
CoinBigIndex rStart = rowStart[iRow];
CoinBigIndex rEnd = rowStart[iRow] + rowLength[iRow];
CoinBigIndex j;
double large = 1.0e15;
int iColumn;
// Compute possible lower and upper ranges
for (j = rStart; j < rEnd; ++j) {
double value = element[j];
iColumn = column[j];
if (value > 0.0) {
if (columnUpper_[iColumn] >= large) {
++infiniteUpper;
} else {
maximumUp += columnUpper_[iColumn] * value;
}
if (columnLower_[iColumn] <= -large) {
++infiniteLower;
} else {
maximumDown += columnLower_[iColumn] * value;
}
} else if (value < 0.0) {
if (columnUpper_[iColumn] >= large) {
++infiniteLower;
} else {
maximumDown += columnUpper_[iColumn] * value;
}
if (columnLower_[iColumn] <= -large) {
++infiniteUpper;
} else {
maximumUp += columnLower_[iColumn] * value;
}
}
}
//assert (infiniteLowerC==infiniteLower);
//assert (infiniteUpperC==infiniteUpper);
if (fabs(maximumUp - maximumUpC) > 1.0e-12 * CoinMax(fabs(maximumUp), fabs(maximumUpC)))
COIN_DETAIL_PRINT(printf("row %d comp up %g, true up %g\n", iRow,
maximumUpC, maximumUp));
if (fabs(maximumDown - maximumDownC) > 1.0e-12 * CoinMax(fabs(maximumDown), fabs(maximumDownC)))
COIN_DETAIL_PRINT(printf("row %d comp down %g, true down %g\n", iRow,
maximumDownC, maximumDown));
maximumUpC = maximumUp;
maximumDownC = maximumDown;
}
/* Tightens primal bounds to make dual faster. Unless
fixed, bounds are slightly looser than they could be.
This is to make dual go faster and is probably not needed
with a presolve. Returns non-zero if problem infeasible
Fudge for branch and bound - put bounds on columns of factor *
largest value (at continuous) - should improve stability
in branch and bound on infeasible branches (0.0 is off)
*/
int
ClpSimplex::tightenPrimalBounds(double factor, int doTight, bool tightIntegers)
{
// Get a row copy in standard format
CoinPackedMatrix copy;
copy.setExtraGap(0.0);
copy.setExtraMajor(0.0);
copy.reverseOrderedCopyOf(*matrix());
// Matrix may have been created so get rid of it
matrix_->releasePackedMatrix();
// get matrix data pointers
const int * column = copy.getIndices();
const CoinBigIndex * rowStart = copy.getVectorStarts();
const int * rowLength = copy.getVectorLengths();
const double * element = copy.getElements();
int numberChanged = 1, iPass = 0;
double large = largeValue(); // treat bounds > this as infinite
#ifndef NDEBUG
double large2 = 1.0e10 * large;
#endif
int numberInfeasible = 0;
int totalTightened = 0;
double tolerance = primalTolerance();
// Save column bounds
double * saveLower = new double [numberColumns_];
CoinMemcpyN(columnLower_, numberColumns_, saveLower);
double * saveUpper = new double [numberColumns_];
CoinMemcpyN(columnUpper_, numberColumns_, saveUpper);
int iRow, iColumn;
// If wanted compute a reasonable dualBound_
if (factor == COIN_DBL_MAX) {
factor = 0.0;
if (dualBound_ == 1.0e10) {
// get largest scaled away from bound
double largest = 1.0e-12;
double largestScaled = 1.0e-12;
int iRow;
for (iRow = 0; iRow < numberRows_; iRow++) {
double value = rowActivity_[iRow];
double above = value - rowLower_[iRow];
double below = rowUpper_[iRow] - value;
if (above < 1.0e12) {
largest = CoinMax(largest, above);
}
if (below < 1.0e12) {
largest = CoinMax(largest, below);
}
if (rowScale_) {
double multiplier = rowScale_[iRow];
above *= multiplier;
below *= multiplier;
}
if (above < 1.0e12) {
largestScaled = CoinMax(largestScaled, above);
}
if (below < 1.0e12) {
largestScaled = CoinMax(largestScaled, below);
}
}
int iColumn;
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
double value = columnActivity_[iColumn];
double above = value - columnLower_[iColumn];
double below = columnUpper_[iColumn] - value;
if (above < 1.0e12) {
largest = CoinMax(largest, above);
}
if (below < 1.0e12) {
largest = CoinMax(largest, below);
}
if (columnScale_) {
double multiplier = 1.0 / columnScale_[iColumn];
above *= multiplier;
below *= multiplier;
}
if (above < 1.0e12) {
largestScaled = CoinMax(largestScaled, above);
}
if (below < 1.0e12) {
largestScaled = CoinMax(largestScaled, below);
}
}
std::cout << "Largest (scaled) away from bound " << largestScaled
<< " unscaled " << largest << std::endl;
dualBound_ = CoinMax(1.0001e7, CoinMin(100.0 * largest, 1.00001e10));
}
}
// If wanted - tighten column bounds using solution
if (factor) {
double largest = 0.0;
if (factor > 0.0) {
assert (factor > 1.0);
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (columnUpper_[iColumn] - columnLower_[iColumn] > tolerance) {
largest = CoinMax(largest, fabs(columnActivity_[iColumn]));
}
}
largest *= factor;
} else {
// absolute
largest = - factor;
}
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (columnUpper_[iColumn] - columnLower_[iColumn] > tolerance) {
columnUpper_[iColumn] = CoinMin(columnUpper_[iColumn], largest);
columnLower_[iColumn] = CoinMax(columnLower_[iColumn], -largest);
}
}
}
#define MAXPASS 10
// Loop round seeing if we can tighten bounds
// Would be faster to have a stack of possible rows
// and we put altered rows back on stack
int numberCheck = -1;
while(numberChanged > numberCheck) {
numberChanged = 0; // Bounds tightened this pass
if (iPass == MAXPASS) break;
iPass++;
for (iRow = 0; iRow < numberRows_; iRow++) {
if (rowLower_[iRow] > -large || rowUpper_[iRow] < large) {
// possible row
int infiniteUpper = 0;
int infiniteLower = 0;
double maximumUp = 0.0;
double maximumDown = 0.0;
double newBound;
CoinBigIndex rStart = rowStart[iRow];
CoinBigIndex rEnd = rowStart[iRow] + rowLength[iRow];
CoinBigIndex j;
// Compute possible lower and upper ranges
for (j = rStart; j < rEnd; ++j) {
double value = element[j];
iColumn = column[j];
if (value > 0.0) {
if (columnUpper_[iColumn] >= large) {
++infiniteUpper;
} else {
maximumUp += columnUpper_[iColumn] * value;
}
if (columnLower_[iColumn] <= -large) {
++infiniteLower;
} else {
maximumDown += columnLower_[iColumn] * value;
}
} else if (value < 0.0) {
if (columnUpper_[iColumn] >= large) {
++infiniteLower;
} else {
maximumDown += columnUpper_[iColumn] * value;
}
if (columnLower_[iColumn] <= -large) {
++infiniteUpper;
} else {
maximumUp += columnLower_[iColumn] * value;
}
}
}
// Build in a margin of error
maximumUp += 1.0e-8 * fabs(maximumUp);
maximumDown -= 1.0e-8 * fabs(maximumDown);
double maxUp = maximumUp + infiniteUpper * 1.0e31;
double maxDown = maximumDown - infiniteLower * 1.0e31;
if (maxUp <= rowUpper_[iRow] + tolerance &&
maxDown >= rowLower_[iRow] - tolerance) {
// Row is redundant - make totally free
// NO - can't do this for postsolve
// rowLower_[iRow]=-COIN_DBL_MAX;
// rowUpper_[iRow]=COIN_DBL_MAX;
//printf("Redundant row in presolveX %d\n",iRow);
} else {
if (maxUp < rowLower_[iRow] - 100.0 * tolerance ||
maxDown > rowUpper_[iRow] + 100.0 * tolerance) {
// problem is infeasible - exit at once
numberInfeasible++;
break;
}
double lower = rowLower_[iRow];
double upper = rowUpper_[iRow];
for (j = rStart; j < rEnd; ++j) {
double value = element[j];
iColumn = column[j];
double nowLower = columnLower_[iColumn];
double nowUpper = columnUpper_[iColumn];
if (value > 0.0) {
// positive value
if (lower > -large) {
if (!infiniteUpper) {
assert(nowUpper < large2);
newBound = nowUpper +
(lower - maximumUp) / value;
// relax if original was large
if (fabs(maximumUp) > 1.0e8)
newBound -= 1.0e-12 * fabs(maximumUp);
} else if (infiniteUpper == 1 && nowUpper > large) {
newBound = (lower - maximumUp) / value;
// relax if original was large
if (fabs(maximumUp) > 1.0e8)
newBound -= 1.0e-12 * fabs(maximumUp);
} else {
newBound = -COIN_DBL_MAX;
}
if (newBound > nowLower + 1.0e-12 && newBound > -large) {
// Tighten the lower bound
numberChanged++;
// check infeasible (relaxed)
if (nowUpper < newBound) {
if (nowUpper - newBound <
-100.0 * tolerance)
numberInfeasible++;
else
newBound = nowUpper;
}
columnLower_[iColumn] = newBound;
// adjust
double now;
if (nowLower < -large) {
now = 0.0;
infiniteLower--;
} else {
now = nowLower;
}
maximumDown += (newBound - now) * value;
nowLower = newBound;
#ifdef DEBUG
checkCorrect(this, iRow,
element, rowStart, rowLength,
column,
columnLower_, columnUpper_,
infiniteUpper,
infiniteLower,
maximumUp,
maximumDown);
#endif
}
}
if (upper < large) {
if (!infiniteLower) {
assert(nowLower > - large2);
newBound = nowLower +
(upper - maximumDown) / value;
// relax if original was large
if (fabs(maximumDown) > 1.0e8)
newBound += 1.0e-12 * fabs(maximumDown);
} else if (infiniteLower == 1 && nowLower < -large) {
newBound = (upper - maximumDown) / value;
// relax if original was large
if (fabs(maximumDown) > 1.0e8)
newBound += 1.0e-12 * fabs(maximumDown);
} else {
newBound = COIN_DBL_MAX;
}
if (newBound < nowUpper - 1.0e-12 && newBound < large) {
// Tighten the upper bound
numberChanged++;
// check infeasible (relaxed)
if (nowLower > newBound) {
if (newBound - nowLower <
-100.0 * tolerance)
numberInfeasible++;
else
newBound = nowLower;
}
columnUpper_[iColumn] = newBound;
// adjust
double now;
if (nowUpper > large) {
now = 0.0;
infiniteUpper--;
} else {
now = nowUpper;
}
maximumUp += (newBound - now) * value;
nowUpper = newBound;
#ifdef DEBUG
checkCorrect(this, iRow,
element, rowStart, rowLength,
column,
columnLower_, columnUpper_,
infiniteUpper,
infiniteLower,
maximumUp,
maximumDown);
#endif
}
}
} else {
// negative value
if (lower > -large) {
if (!infiniteUpper) {
assert(nowLower < large2);
newBound = nowLower +
(lower - maximumUp) / value;
// relax if original was large
if (fabs(maximumUp) > 1.0e8)
newBound += 1.0e-12 * fabs(maximumUp);
} else if (infiniteUpper == 1 && nowLower < -large) {
newBound = (lower - maximumUp) / value;
// relax if original was large
if (fabs(maximumUp) > 1.0e8)
newBound += 1.0e-12 * fabs(maximumUp);
} else {
newBound = COIN_DBL_MAX;
}
if (newBound < nowUpper - 1.0e-12 && newBound < large) {
// Tighten the upper bound
numberChanged++;
// check infeasible (relaxed)
if (nowLower > newBound) {
if (newBound - nowLower <
-100.0 * tolerance)
numberInfeasible++;
else
newBound = nowLower;
}
columnUpper_[iColumn] = newBound;
// adjust
double now;
if (nowUpper > large) {
now = 0.0;
infiniteLower--;
} else {
now = nowUpper;
}
maximumDown += (newBound - now) * value;
nowUpper = newBound;
#ifdef DEBUG
checkCorrect(this, iRow,
element, rowStart, rowLength,
column,
columnLower_, columnUpper_,
infiniteUpper,
infiniteLower,
maximumUp,
maximumDown);
#endif
}
}
if (upper < large) {
if (!infiniteLower) {
assert(nowUpper < large2);
newBound = nowUpper +
(upper - maximumDown) / value;
// relax if original was large
if (fabs(maximumDown) > 1.0e8)
newBound -= 1.0e-12 * fabs(maximumDown);
} else if (infiniteLower == 1 && nowUpper > large) {
newBound = (upper - maximumDown) / value;
// relax if original was large
if (fabs(maximumDown) > 1.0e8)
newBound -= 1.0e-12 * fabs(maximumDown);
} else {
newBound = -COIN_DBL_MAX;
}
if (newBound > nowLower + 1.0e-12 && newBound > -large) {
// Tighten the lower bound
numberChanged++;
// check infeasible (relaxed)
if (nowUpper < newBound) {
if (nowUpper - newBound <
-100.0 * tolerance)
numberInfeasible++;
else
newBound = nowUpper;
}
columnLower_[iColumn] = newBound;
// adjust
double now;
if (nowLower < -large) {
now = 0.0;
infiniteUpper--;
} else {
now = nowLower;
}
maximumUp += (newBound - now) * value;
nowLower = newBound;
#ifdef DEBUG
checkCorrect(this, iRow,
element, rowStart, rowLength,
column,
columnLower_, columnUpper_,
infiniteUpper,
infiniteLower,
maximumUp,
maximumDown);
#endif
}
}
}
}
}
}
}
totalTightened += numberChanged;
if (iPass == 1)
numberCheck = numberChanged >> 4;
if (numberInfeasible) break;
}
if (!numberInfeasible) {
handler_->message(CLP_SIMPLEX_BOUNDTIGHTEN, messages_)
<< totalTightened
<< CoinMessageEol;
// Set bounds slightly loose
double useTolerance = 1.0e-3;
if (doTight > 0) {
if (doTight > 10) {
useTolerance = 0.0;
} else {
while (doTight) {
useTolerance *= 0.1;
doTight--;
}
}
}
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (saveUpper[iColumn] > saveLower[iColumn] + useTolerance) {
// Make large bounds stay infinite
if (saveUpper[iColumn] > 1.0e30 && columnUpper_[iColumn] > 1.0e10) {
columnUpper_[iColumn] = COIN_DBL_MAX;
}
if (saveLower[iColumn] < -1.0e30 && columnLower_[iColumn] < -1.0e10) {
columnLower_[iColumn] = -COIN_DBL_MAX;
}
#ifdef KEEP_GOING_IF_FIXED
double multiplier = 5.0e-3 * floor(100.0 * randomNumberGenerator_.randomDouble()) + 1.0;
multiplier *= 100.0;
#else
double multiplier = 100.0;
#endif
if (columnUpper_[iColumn] - columnLower_[iColumn] < useTolerance + 1.0e-8) {
// relax enough so will have correct dj
#if 1
columnLower_[iColumn] = CoinMax(saveLower[iColumn],
columnLower_[iColumn] - multiplier * useTolerance);
columnUpper_[iColumn] = CoinMin(saveUpper[iColumn],
columnUpper_[iColumn] + multiplier * useTolerance);
#else
if (fabs(columnUpper_[iColumn]) < fabs(columnLower_[iColumn])) {
if (columnUpper_[iColumn] - multiplier * useTolerance > saveLower[iColumn]) {
columnLower_[iColumn] = columnUpper_[iColumn] - multiplier * useTolerance;
} else {
columnLower_[iColumn] = saveLower[iColumn];
columnUpper_[iColumn] = CoinMin(saveUpper[iColumn],
saveLower[iColumn] + multiplier * useTolerance);
}
} else {
if (columnLower_[iColumn] + multiplier * useTolerance < saveUpper[iColumn]) {
columnUpper_[iColumn] = columnLower_[iColumn] + multiplier * useTolerance;
} else {
columnUpper_[iColumn] = saveUpper[iColumn];
columnLower_[iColumn] = CoinMax(saveLower[iColumn],
saveUpper[iColumn] - multiplier * useTolerance);
}
}
#endif
} else {
if (columnUpper_[iColumn] < saveUpper[iColumn]) {
// relax a bit
columnUpper_[iColumn] = CoinMin(columnUpper_[iColumn] + multiplier * useTolerance,
saveUpper[iColumn]);
}
if (columnLower_[iColumn] > saveLower[iColumn]) {
// relax a bit
columnLower_[iColumn] = CoinMax(columnLower_[iColumn] - multiplier * useTolerance,
saveLower[iColumn]);
}
}
}
}
if (tightIntegers && integerType_) {
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (integerType_[iColumn]) {
double value;
value = floor(columnLower_[iColumn] + 0.5);
if (fabs(value - columnLower_[iColumn]) > primalTolerance_)
value = ceil(columnLower_[iColumn]);
columnLower_[iColumn] = value;
value = floor(columnUpper_[iColumn] + 0.5);
if (fabs(value - columnUpper_[iColumn]) > primalTolerance_)
value = floor(columnUpper_[iColumn]);
columnUpper_[iColumn] = value;
if (columnLower_[iColumn] > columnUpper_[iColumn])
numberInfeasible++;
}
}
if (numberInfeasible) {
handler_->message(CLP_SIMPLEX_INFEASIBILITIES, messages_)
<< numberInfeasible
<< CoinMessageEol;
// restore column bounds
CoinMemcpyN(saveLower, numberColumns_, columnLower_);
CoinMemcpyN(saveUpper, numberColumns_, columnUpper_);
}
}
} else {
handler_->message(CLP_SIMPLEX_INFEASIBILITIES, messages_)
<< numberInfeasible
<< CoinMessageEol;
// restore column bounds
CoinMemcpyN(saveLower, numberColumns_, columnLower_);
CoinMemcpyN(saveUpper, numberColumns_, columnUpper_);
}
delete [] saveLower;
delete [] saveUpper;
return (numberInfeasible);
}
//#define SAVE_AND_RESTORE
// dual
#include "ClpSimplexDual.hpp"
#include "ClpSimplexPrimal.hpp"
#ifndef SAVE_AND_RESTORE
int ClpSimplex::dual (int ifValuesPass , int startFinishOptions)
#else
int ClpSimplex::dual (int ifValuesPass , int startFinishOptions)
{
// May be empty problem
if (numberRows_ && numberColumns_) {
// Save on file for debug
int returnCode;
returnCode = saveModel("debug.sav");
if (returnCode) {
printf("** Unable to save model to debug.sav\n");
abort();
}
ClpSimplex temp;
returnCode = temp.restoreModel("debug.sav");
if (returnCode) {
printf("** Unable to restore model from debug.sav\n");
abort();
}
temp.setLogLevel(handler_->logLevel());
// Now do dual
returnCode = temp.dualDebug(ifValuesPass, startFinishOptions);
// Move status and solution back
int numberTotal = numberRows_ + numberColumns_;
CoinMemcpyN(temp.statusArray(), numberTotal, status_);
CoinMemcpyN(temp.primalColumnSolution(), numberColumns_, columnActivity_);
CoinMemcpyN(temp.primalRowSolution(), numberRows_, rowActivity_);
CoinMemcpyN(temp.dualColumnSolution(), numberColumns_, reducedCost_);
CoinMemcpyN(temp.dualRowSolution(), numberRows_, dual_);
problemStatus_ = temp.problemStatus_;
setObjectiveValue(temp.objectiveValue());
setSumDualInfeasibilities(temp.sumDualInfeasibilities());
setNumberDualInfeasibilities(temp.numberDualInfeasibilities());
setSumPrimalInfeasibilities(temp.sumPrimalInfeasibilities());
setNumberPrimalInfeasibilities(temp.numberPrimalInfeasibilities());
setNumberIterations(temp.numberIterations());
onStopped(); // set secondary status if stopped
return returnCode;
} else {
// empty
return dualDebug(ifValuesPass, startFinishOptions);
}
}
int ClpSimplex::dualDebug (int ifValuesPass , int startFinishOptions)
#endif
{
//double savedPivotTolerance = factorization_->pivotTolerance();
int saveQuadraticActivated = 0;
if (objective_) {
saveQuadraticActivated = objective_->activated();
objective_->setActivated(0);
} else {
// create dummy stuff
assert (!numberColumns_);
if (!numberRows_)
problemStatus_ = 0; // say optimal
return 0;
}
ClpObjective * saveObjective = objective_;
CoinAssert (ifValuesPass >= 0 && ifValuesPass < 3);
/* Note use of "down casting". The only class the user sees is ClpSimplex.
Classes ClpSimplexDual, ClpSimplexPrimal, (ClpSimplexNonlinear)
and ClpSimplexOther all exist and inherit from ClpSimplex but have no
additional data and have no destructor or (non-default) constructor.
This is to stop classes becoming too unwieldy and so I (JJF) can use e.g. "perturb"
in primal and dual.
As far as I can see this is perfectly safe.
*/
#ifdef COIN_DEVELOP
//#define EXPENSIVE
#endif
#ifdef EXPENSIVE
static int dualCount = 0;
static int dualCheckCount = -1;
dualCount++;
if (dualCount == dualCheckCount) {
printf("Bad dual coming up\n");
}
ClpSimplex saveModel = *this;
#endif
int returnCode = static_cast<ClpSimplexDual *> (this)->dual(ifValuesPass, startFinishOptions);
#ifdef EXPENSIVE
if (problemStatus_ == 1) {
saveModel.allSlackBasis(true);
static_cast<ClpSimplexDual *> (&saveModel)->dual(0, 0);
if (saveModel.problemStatus_ == 0) {
if (saveModel.objectiveValue() < dblParam_[0] - 1.0e-8 * (1.0 + fabs(dblParam_[0]))) {
if (objectiveValue() < dblParam_[0] - 1.0e-6 * (1.0 + fabs(dblParam_[0]))) {
printf("BAD dual - objs %g ,savemodel %g cutoff %g at count %d\n",
objectiveValue(), saveModel.objectiveValue(), dblParam_[0], dualCount);
saveModel = *this;
saveModel.setLogLevel(63);
static_cast<ClpSimplexDual *> (&saveModel)->dual(0, 0);
// flatten solution and try again
int iRow, iColumn;
for (iRow = 0; iRow < numberRows_; iRow++) {
if (getRowStatus(iRow) != basic) {
setRowStatus(iRow, superBasic);
// but put to bound if close
if (fabs(rowActivity_[iRow] - rowLower_[iRow])
<= primalTolerance_) {
rowActivity_[iRow] = rowLower_[iRow];
setRowStatus(iRow, atLowerBound);
} else if (fabs(rowActivity_[iRow] - rowUpper_[iRow])
<= primalTolerance_) {
rowActivity_[iRow] = rowUpper_[iRow];
setRowStatus(iRow, atUpperBound);
}
}
}
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (getColumnStatus(iColumn) != basic) {
setColumnStatus(iColumn, superBasic);
// but put to bound if close
if (fabs(columnActivity_[iColumn] - columnLower_[iColumn])
<= primalTolerance_) {
columnActivity_[iColumn] = columnLower_[iColumn];
setColumnStatus(iColumn, atLowerBound);
} else if (fabs(columnActivity_[iColumn]
- columnUpper_[iColumn])
<= primalTolerance_) {
columnActivity_[iColumn] = columnUpper_[iColumn];
setColumnStatus(iColumn, atUpperBound);
}
}
}
static_cast<ClpSimplexPrimal *> (&saveModel)->primal(0, 0);
} else {
printf("bad? dual - objs %g ,savemodel %g cutoff %g at count %d\n",
objectiveValue(), saveModel.objectiveValue(), dblParam_[0], dualCount);
}
if (dualCount > dualCheckCount && dualCheckCount >= 0)
abort();
}
}
}
#endif
//int lastAlgorithm = -1;
if ((specialOptions_ & 2048) != 0 && problemStatus_ == 10 && !numberPrimalInfeasibilities_
&& sumDualInfeasibilities_ < 1000.0 * dualTolerance_ && perturbation_ >= 100)
problemStatus_ = 0; // ignore
if (problemStatus_==1&&((specialOptions_&(1024 | 4096)) == 0 || (specialOptions_ & 32) != 0)
&&numberFake_) {
problemStatus_ = 10; // clean up in primal as fake bounds
}
if (problemStatus_ == 10) {
//printf("Cleaning up with primal\n");
#ifdef COIN_DEVELOP
int saveNumberIterations = numberIterations_;
#endif
//lastAlgorithm=1;
int savePerturbation = perturbation_;
int saveLog = handler_->logLevel();
//handler_->setLogLevel(63);
perturbation_ = 100;
bool denseFactorization = initialDenseFactorization();
// It will be safe to allow dense
setInitialDenseFactorization(true);
// Allow for catastrophe
int saveMax = intParam_[ClpMaxNumIteration];
if (numberIterations_) {
// normal
if (intParam_[ClpMaxNumIteration] > 100000 + numberIterations_)
intParam_[ClpMaxNumIteration]
= numberIterations_ + 1000 + 2 * numberRows_ + numberColumns_;
} else {
// Not normal allow more
baseIteration_ += 2 * (numberRows_ + numberColumns_);
}
// check which algorithms allowed
int dummy;
ClpPackedMatrix * ordinary =
dynamic_cast< ClpPackedMatrix*>(matrix_);
if (problemStatus_ == 10 && saveObjective == objective_ &&
ordinary)
startFinishOptions |= 2;
baseIteration_ = numberIterations_;
// Say second call
moreSpecialOptions_ |= 256;
if ((matrix_->generalExpanded(this, 4, dummy) & 1) != 0)
returnCode = static_cast<ClpSimplexPrimal *> (this)->primal(1, startFinishOptions);
else
returnCode = static_cast<ClpSimplexDual *> (this)->dual(0, startFinishOptions);
// Say not second call
moreSpecialOptions_ &= ~256;
baseIteration_ = 0;
if (saveObjective != objective_) {
// We changed objective to see if infeasible
delete objective_;
objective_ = saveObjective;
if (!problemStatus_) {
// carry on
returnCode = static_cast<ClpSimplexPrimal *> (this)->primal(1, startFinishOptions);
}
}
if (problemStatus_ == 3 && numberIterations_ < saveMax) {
#ifdef COIN_DEVELOP
if (handler_->logLevel() > 0)
printf("looks like trouble - too many iterations in clean up - trying again\n");
#endif
// flatten solution and try again
int iRow, iColumn;
for (iRow = 0; iRow < numberRows_; iRow++) {
if (getRowStatus(iRow) != basic) {
setRowStatus(iRow, superBasic);
// but put to bound if close
if (fabs(rowActivity_[iRow] - rowLower_[iRow])
<= primalTolerance_) {
rowActivity_[iRow] = rowLower_[iRow];
setRowStatus(iRow, atLowerBound);
} else if (fabs(rowActivity_[iRow] - rowUpper_[iRow])
<= primalTolerance_) {
rowActivity_[iRow] = rowUpper_[iRow];
setRowStatus(iRow, atUpperBound);
}
}
}
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (getColumnStatus(iColumn) != basic) {
setColumnStatus(iColumn, superBasic);
// but put to bound if close
if (fabs(columnActivity_[iColumn] - columnLower_[iColumn])
<= primalTolerance_) {
columnActivity_[iColumn] = columnLower_[iColumn];
setColumnStatus(iColumn, atLowerBound);
} else if (fabs(columnActivity_[iColumn]
- columnUpper_[iColumn])
<= primalTolerance_) {
columnActivity_[iColumn] = columnUpper_[iColumn];
setColumnStatus(iColumn, atUpperBound);
}
}
}
problemStatus_ = -1;
intParam_[ClpMaxNumIteration] = CoinMin(numberIterations_ + 1000 +
2 * numberRows_ + numberColumns_, saveMax);
perturbation_ = savePerturbation;
baseIteration_ = numberIterations_;
// Say second call
moreSpecialOptions_ |= 256;
returnCode = static_cast<ClpSimplexPrimal *> (this)->primal(0, startFinishOptions);
// Say not second call
moreSpecialOptions_ &= ~256;
baseIteration_ = 0;
computeObjectiveValue();
// can't rely on djs either
memset(reducedCost_, 0, numberColumns_ * sizeof(double));
#ifdef COIN_DEVELOP
if (problemStatus_ == 3 && numberIterations_ < saveMax &&
handler_->logLevel() > 0)
printf("looks like real trouble - too many iterations in second clean up - giving up\n");
#endif
}
intParam_[ClpMaxNumIteration] = saveMax;
setInitialDenseFactorization(denseFactorization);
perturbation_ = savePerturbation;
if (problemStatus_ == 10) {
if (!numberPrimalInfeasibilities_)
problemStatus_ = 0;
else
problemStatus_ = 4;
}
handler_->setLogLevel(saveLog);
#ifdef COIN_DEVELOP
if (numberIterations_ > 200)
printf("after primal status %d - %d iterations (save %d)\n",
problemStatus_, numberIterations_, saveNumberIterations);
#endif
}
objective_->setActivated(saveQuadraticActivated);
//factorization_->pivotTolerance(savedPivotTolerance);
onStopped(); // set secondary status if stopped
//if (problemStatus_==1&&lastAlgorithm==1)
//returnCode=10; // so will do primal after postsolve
if (!problemStatus_) {
//assert (!numberPrimalInfeasibilities_);
//if (returnCode!=10)
//assert (!numberDualInfeasibilities_);
}
return returnCode;
}
// primal
int ClpSimplex::primal (int ifValuesPass , int startFinishOptions)
{
//double savedPivotTolerance = factorization_->pivotTolerance();
#ifndef SLIM_CLP
// See if nonlinear
if (objective_->type() > 1 && objective_->activated())
return reducedGradient();
#endif
CoinAssert ((ifValuesPass >= 0 && ifValuesPass < 3) ||
(ifValuesPass >= 12 && ifValuesPass < 100) ||
(ifValuesPass >= 112 && ifValuesPass < 200));
if (ifValuesPass >= 12) {
int numberProblems = (ifValuesPass - 10) % 100;
ifValuesPass = (ifValuesPass < 100) ? 1 : 2;
// Go parallel to do solve
// Only if all slack basis
int i;
for ( i = 0; i < numberColumns_; i++) {
if (getColumnStatus(i) == basic)
break;
}
if (i == numberColumns_) {
// try if vaguely feasible
CoinZeroN(rowActivity_, numberRows_);
const int * row = matrix_->getIndices();
const CoinBigIndex * columnStart = matrix_->getVectorStarts();
const int * columnLength = matrix_->getVectorLengths();
const double * element = matrix_->getElements();
for (int iColumn = 0; iColumn < numberColumns_; iColumn++) {
CoinBigIndex j;
double value = columnActivity_[iColumn];
if (value) {
CoinBigIndex start = columnStart[iColumn];
CoinBigIndex end = start + columnLength[iColumn];
for (j = start; j < end; j++) {
int iRow = row[j];
rowActivity_[iRow] += value * element[j];
}
}
}
checkSolutionInternal();
if (sumPrimalInfeasibilities_ * sqrt(static_cast<double>(numberRows_)) < 1.0) {
// Could do better if can decompose
// correction to get feasible
double scaleFactor = 1.0 / numberProblems;
double * correction = new double [numberRows_];
for (int iRow = 0; iRow < numberRows_; iRow++) {
double value = rowActivity_[iRow];
if (value > rowUpper_[iRow])
value = rowUpper_[iRow] - value;
else if (value < rowLower_[iRow])
value = rowLower_[iRow] - value;
else
value = 0.0;
correction[iRow] = value * scaleFactor;
}
int numberColumns = (numberColumns_ + numberProblems - 1) / numberProblems;
int * whichRows = new int [numberRows_];
for (int i = 0; i < numberRows_; i++)
whichRows[i] = i;
int * whichColumns = new int [numberColumns_];
ClpSimplex ** model = new ClpSimplex * [numberProblems];
int startColumn = 0;
double * saveLower = CoinCopyOfArray(rowLower_, numberRows_);
double * saveUpper = CoinCopyOfArray(rowUpper_, numberRows_);
for (int i = 0; i < numberProblems; i++) {
int endColumn = CoinMin(startColumn + numberColumns, numberColumns_);
CoinZeroN(rowActivity_, numberRows_);
for (int iColumn = startColumn; iColumn < endColumn; iColumn++) {
whichColumns[iColumn-startColumn] = iColumn;
CoinBigIndex j;
double value = columnActivity_[iColumn];
if (value) {
CoinBigIndex start = columnStart[iColumn];
CoinBigIndex end = start + columnLength[iColumn];
for (j = start; j < end; j++) {
int iRow = row[j];
rowActivity_[iRow] += value * element[j];
}
}
}
// adjust rhs
for (int iRow = 0; iRow < numberRows_; iRow++) {
double value = rowActivity_[iRow] + correction[iRow];
if (saveUpper[iRow] < 1.0e30)
rowUpper_[iRow] = value;
if (saveLower[iRow] > -1.0e30)
rowLower_[iRow] = value;
}
model[i] = new ClpSimplex(this, numberRows_, whichRows,
endColumn - startColumn, whichColumns);
//#define FEB_TRY
#ifdef FEB_TRY
model[i]->setPerturbation(perturbation_);
#endif
startColumn = endColumn;
}
memcpy(rowLower_, saveLower, numberRows_ * sizeof(double));
memcpy(rowUpper_, saveUpper, numberRows_ * sizeof(double));
delete [] saveLower;
delete [] saveUpper;
delete [] correction;
// solve (in parallel)
for (int i = 0; i < numberProblems; i++) {
model[i]->primal(1/*ifValuesPass*/);
}
startColumn = 0;
int numberBasic = 0;
// use whichRows as counter
for (int iRow = 0; iRow < numberRows_; iRow++) {
int startValue = 0;
if (rowUpper_[iRow] > rowLower_[iRow])
startValue++;
if (rowUpper_[iRow] > 1.0e30)
startValue++;
if (rowLower_[iRow] < -1.0e30)
startValue++;
whichRows[iRow] = 1000 * startValue;
}
for (int i = 0; i < numberProblems; i++) {
int endColumn = CoinMin(startColumn + numberColumns, numberColumns_);
ClpSimplex * simplex = model[i];
const double * solution = simplex->columnActivity_;
for (int iColumn = startColumn; iColumn < endColumn; iColumn++) {
columnActivity_[iColumn] = solution[iColumn-startColumn];
Status status = simplex->getColumnStatus(iColumn - startColumn);
setColumnStatus(iColumn, status);
if (status == basic)
numberBasic++;
}
for (int iRow = 0; iRow < numberRows_; iRow++) {
if (simplex->getRowStatus(iRow) == basic)
whichRows[iRow]++;
}
delete model[i];
startColumn = endColumn;
}
delete [] model;
for (int iRow = 0; iRow < numberRows_; iRow++)
setRowStatus(iRow, superBasic);
CoinZeroN(rowActivity_, numberRows_);
for (int iColumn = 0; iColumn < numberColumns_; iColumn++) {
CoinBigIndex j;
double value = columnActivity_[iColumn];
if (value) {
CoinBigIndex start = columnStart[iColumn];
CoinBigIndex end = start + columnLength[iColumn];
for (j = start; j < end; j++) {
int iRow = row[j];
rowActivity_[iRow] += value * element[j];
}
}
}
checkSolutionInternal();
if (numberBasic < numberRows_) {
int * order = new int [numberRows_];
for (int iRow = 0; iRow < numberRows_; iRow++) {
setRowStatus(iRow, superBasic);
int nTimes = whichRows[iRow] % 1000;
if (nTimes)
nTimes += whichRows[iRow] / 500;
whichRows[iRow] = -nTimes;
order[iRow] = iRow;
}
CoinSort_2(whichRows, whichRows + numberRows_, order);
int nPut = numberRows_ - numberBasic;
for (int i = 0; i < nPut; i++) {
int iRow = order[i];
setRowStatus(iRow, basic);
}
delete [] order;
} else if (numberBasic > numberRows_) {
double * away = new double [numberBasic];
numberBasic = 0;
for (int iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (getColumnStatus(iColumn) == basic) {
double value = columnActivity_[iColumn];
value = CoinMin(value - columnLower_[iColumn],
columnUpper_[iColumn] - value);
away[numberBasic] = value;
whichColumns[numberBasic++] = iColumn;
}
}
CoinSort_2(away, away + numberBasic, whichColumns);
int nPut = numberBasic - numberRows_;
for (int i = 0; i < nPut; i++) {
int iColumn = whichColumns[i];
double value = columnActivity_[iColumn];
if (value - columnLower_[iColumn] <
columnUpper_[iColumn] - value)
setColumnStatus(iColumn, atLowerBound);
else
setColumnStatus(iColumn, atUpperBound);
}
delete [] away;
}
delete [] whichColumns;
delete [] whichRows;
}
}
}
//firstFree_=-1;
/* Note use of "down casting". The only class the user sees is ClpSimplex.
Classes ClpSimplexDual, ClpSimplexPrimal, (ClpSimplexNonlinear)
and ClpSimplexOther all exist and inherit from ClpSimplex but have no
additional data and have no destructor or (non-default) constructor.
This is to stop classes becoming too unwieldy and so I (JJF) can use e.g. "perturb"
in primal and dual.
As far as I can see this is perfectly safe.
*/
int returnCode = static_cast<ClpSimplexPrimal *> (this)->primal(ifValuesPass, startFinishOptions);
//int lastAlgorithm=1;
if (problemStatus_ == 10) {
//lastAlgorithm=-1;
//printf("Cleaning up with dual\n");
int savePerturbation = perturbation_;
perturbation_ = 100;
bool denseFactorization = initialDenseFactorization();
// It will be safe to allow dense
setInitialDenseFactorization(true);
// check which algorithms allowed
int dummy;
baseIteration_ = numberIterations_;
// Say second call
moreSpecialOptions_ |= 256;
if ((matrix_->generalExpanded(this, 4, dummy) & 2) != 0 && (specialOptions_ & 8192) == 0) {
double saveBound = dualBound_;
// upperOut_ has largest away from bound
dualBound_ = CoinMin(CoinMax(2.0 * upperOut_, 1.0e8), dualBound_);
returnCode = static_cast<ClpSimplexDual *> (this)->dual(0, startFinishOptions);
dualBound_ = saveBound;
} else {
returnCode = static_cast<ClpSimplexPrimal *> (this)->primal(0, startFinishOptions);
}
// Say not second call
moreSpecialOptions_ &= ~256;
baseIteration_ = 0;
setInitialDenseFactorization(denseFactorization);
perturbation_ = savePerturbation;
if (problemStatus_ == 10) {
if (!numberPrimalInfeasibilities_) {
problemStatus_ = 0;
numberDualInfeasibilities_ = 0;
} else {
problemStatus_ = 4;
}
}
}
//factorization_->pivotTolerance(savedPivotTolerance);
onStopped(); // set secondary status if stopped
//if (problemStatus_==1&&lastAlgorithm==1)
//returnCode=10; // so will do primal after postsolve
return returnCode;
}
#ifndef SLIM_CLP
#include "ClpQuadraticObjective.hpp"
/* Dual ranging.
This computes increase/decrease in cost for each given variable and corresponding
sequence numbers which would change basis. Sequence numbers are 0..numberColumns
and numberColumns.. for artificials/slacks.
For non-basic variables the sequence number will be that of the non-basic variables.
Up to user to provide correct length arrays.
Returns non-zero if infeasible unbounded etc
*/
#include "ClpSimplexOther.hpp"
int ClpSimplex::dualRanging(int numberCheck, const int * which,
double * costIncrease, int * sequenceIncrease,
double * costDecrease, int * sequenceDecrease,
double * valueIncrease, double * valueDecrease)
{
int savePerturbation = perturbation_;
perturbation_ = 100;
/*int returnCode =*/ static_cast<ClpSimplexPrimal *> (this)->primal(0, 1);
if (problemStatus_ == 10) {
//printf("Cleaning up with dual\n");
bool denseFactorization = initialDenseFactorization();
// It will be safe to allow dense
setInitialDenseFactorization(true);
// check which algorithms allowed
int dummy;
if ((matrix_->generalExpanded(this, 4, dummy) & 2) != 0) {
// upperOut_ has largest away from bound
double saveBound = dualBound_;
if (upperOut_ > 0.0)
dualBound_ = 2.0 * upperOut_;
/*returnCode =*/ static_cast<ClpSimplexDual *> (this)->dual(0, 1);
dualBound_ = saveBound;
} else {
/*returnCode =*/ static_cast<ClpSimplexPrimal *> (this)->primal(0, 1);
}
setInitialDenseFactorization(denseFactorization);
if (problemStatus_ == 10)
problemStatus_ = 0;
}
perturbation_ = savePerturbation;
if (problemStatus_ || secondaryStatus_ == 6) {
finish(); // get rid of arrays
return 1; // odd status
}
static_cast<ClpSimplexOther *> (this)->dualRanging(numberCheck, which,
costIncrease, sequenceIncrease,
costDecrease, sequenceDecrease,
valueIncrease, valueDecrease);
finish(); // get rid of arrays
return 0;
}
/* Primal ranging.
This computes increase/decrease in value for each given variable and corresponding
sequence numbers which would change basis. Sequence numbers are 0..numberColumns
and numberColumns.. for artificials/slacks.
For basic variables the sequence number will be that of the basic variables.
Up to user to providen correct length arrays.
Returns non-zero if infeasible unbounded etc
*/
int ClpSimplex::primalRanging(int numberCheck, const int * which,
double * valueIncrease, int * sequenceIncrease,
double * valueDecrease, int * sequenceDecrease)
{
int savePerturbation = perturbation_;
perturbation_ = 100;
/*int returnCode =*/ static_cast<ClpSimplexPrimal *> (this)->primal(0, 1);
if (problemStatus_ == 10) {
//printf("Cleaning up with dual\n");
bool denseFactorization = initialDenseFactorization();
// It will be safe to allow dense
setInitialDenseFactorization(true);
// check which algorithms allowed
int dummy;
if ((matrix_->generalExpanded(this, 4, dummy) & 2) != 0) {
// upperOut_ has largest away from bound
double saveBound = dualBound_;
if (upperOut_ > 0.0)
dualBound_ = 2.0 * upperOut_;
/*returnCode =*/ static_cast<ClpSimplexDual *> (this)->dual(0, 1);
dualBound_ = saveBound;
} else {
/*returnCode =*/ static_cast<ClpSimplexPrimal *> (this)->primal(0, 1);
}
setInitialDenseFactorization(denseFactorization);
if (problemStatus_ == 10)
problemStatus_ = 0;
}
perturbation_ = savePerturbation;
if (problemStatus_ || secondaryStatus_ == 6) {
finish(); // get rid of arrays
return 1; // odd status
}
static_cast<ClpSimplexOther *> (this)->primalRanging(numberCheck, which,
valueIncrease, sequenceIncrease,
valueDecrease, sequenceDecrease);
finish(); // get rid of arrays
return 0;
}
/* Write the basis in MPS format to the specified file.
If writeValues true writes values of structurals
(and adds VALUES to end of NAME card)
Row and column names may be null.
formatType is
<ul>
<li> 0 - normal
<li> 1 - extra accuracy
<li> 2 - IEEE hex (later)
</ul>
Returns non-zero on I/O error
*/
int
ClpSimplex::writeBasis(const char *filename,
bool writeValues,
int formatType) const
{
return static_cast<const ClpSimplexOther *> (this)->writeBasis(filename, writeValues,
formatType);
}
// Read a basis from the given filename
int
ClpSimplex::readBasis(const char *filename)
{
return static_cast<ClpSimplexOther *> (this)->readBasis(filename);
}
#include "ClpSimplexNonlinear.hpp"
/* Solves nonlinear problem using SLP - may be used as crash
for other algorithms when number of iterations small
*/
int
ClpSimplex::nonlinearSLP(int numberPasses, double deltaTolerance)
{
return static_cast<ClpSimplexNonlinear *> (this)->primalSLP(numberPasses, deltaTolerance);
}
/* Solves problem with nonlinear constraints using SLP - may be used as crash
for other algorithms when number of iterations small.
Also exits if all problematical variables are changing
less than deltaTolerance
*/
int
ClpSimplex::nonlinearSLP(int numberConstraints, ClpConstraint ** constraints,
int numberPasses, double deltaTolerance)
{
return static_cast<ClpSimplexNonlinear *> (this)->primalSLP(numberConstraints, constraints, numberPasses, deltaTolerance);
}
// Solves non-linear using reduced gradient
int ClpSimplex::reducedGradient(int phase)
{
if (objective_->type() < 2 || !objective_->activated()) {
// no quadratic part
return primal(0);
}
// get feasible
if ((this->status() < 0 || numberPrimalInfeasibilities()) && phase == 0) {
objective_->setActivated(0);
double saveDirection = optimizationDirection();
setOptimizationDirection(0.0);
primal(1);
setOptimizationDirection(saveDirection);
objective_->setActivated(1);
// still infeasible
if (numberPrimalInfeasibilities())
return 0;
}
// Now enter method
int returnCode = static_cast<ClpSimplexNonlinear *> (this)->primal();
return returnCode;
}
#include "ClpPredictorCorrector.hpp"
#include "ClpCholeskyBase.hpp"
#include "ClpPresolve.hpp"
/* Solves using barrier (assumes you have good cholesky factor code).
Does crossover to simplex if asked*/
int
ClpSimplex::barrier(bool crossover)
{
ClpSimplex * model2 = this;
int savePerturbation = perturbation_;
ClpInterior barrier;
barrier.borrowModel(*model2);
// See if quadratic objective
ClpQuadraticObjective * quadraticObj = NULL;
if (objective_->type() == 2)
quadraticObj = (static_cast< ClpQuadraticObjective*>(objective_));
// If Quadratic we need KKT
bool doKKT = (quadraticObj != NULL);
// Preference is WSSMP, UFL, MUMPS, TAUCS then base
#ifdef WSSMP_BARRIER
if (!doKKT) {
ClpCholeskyWssmp * cholesky = new ClpCholeskyWssmp(CoinMax(100, model2->numberRows() / 10));
barrier.setCholesky(cholesky);
} else {
//ClpCholeskyWssmp * cholesky = new ClpCholeskyWssmp();
ClpCholeskyWssmpKKT * cholesky = new ClpCholeskyWssmpKKT(CoinMax(100, model2->numberRows() / 10));
barrier.setCholesky(cholesky);
}
#elif defined(COIN_HAS_AMD) || defined(COIN_HAS_CHOLMOD)
if (!doKKT) {
ClpCholeskyUfl * cholesky = new ClpCholeskyUfl();
barrier.setCholesky(cholesky);
} else {
ClpCholeskyBase * cholesky = new ClpCholeskyBase();
// not yetClpCholeskyUfl * cholesky = new ClpCholeskyUfl();
cholesky->setKKT(true);
barrier.setCholesky(cholesky);
}
#elif TAUCS_BARRIER
assert (!doKKT);
ClpCholeskyTaucs * cholesky = new ClpCholeskyTaucs();
barrier.setCholesky(cholesky);
#elifdef COIN_HAS_MUMPS
if (!doKKT) {
ClpCholeskyMumps * cholesky = new ClpCholeskyMumps();
barrier.setCholesky(cholesky);
} else {
printf("***** Unable to do Mumps with KKT\n");
ClpCholeskyBase * cholesky = new ClpCholeskyBase();
cholesky->setKKT(true);
barrier.setCholesky(cholesky);
}
#else
if (!doKKT) {
ClpCholeskyBase * cholesky = new ClpCholeskyBase();
barrier.setCholesky(cholesky);
} else {
ClpCholeskyBase * cholesky = new ClpCholeskyBase();
cholesky->setKKT(true);
barrier.setCholesky(cholesky);
}
#endif
barrier.setDiagonalPerturbation(1.0e-14);
int numberRows = model2->numberRows();
int numberColumns = model2->numberColumns();
int saveMaxIts = model2->maximumIterations();
if (saveMaxIts < 1000) {
barrier.setMaximumBarrierIterations(saveMaxIts);
model2->setMaximumIterations(1000000);
}
barrier.primalDual();
int barrierStatus = barrier.status();
double gap = static_cast<double>(barrier.complementarityGap());
// get which variables are fixed
double * saveLower = NULL;
double * saveUpper = NULL;
ClpPresolve pinfo2;
ClpSimplex * saveModel2 = NULL;
int numberFixed = barrier.numberFixed();
if (numberFixed * 20 > barrier.numberRows() && numberFixed > 5000 && crossover && 0) {
// may as well do presolve
int numberRows = barrier.numberRows();
int numberColumns = barrier.numberColumns();
int numberTotal = numberRows + numberColumns;
saveLower = new double [numberTotal];
saveUpper = new double [numberTotal];
CoinMemcpyN(barrier.columnLower(), numberColumns, saveLower);
CoinMemcpyN(barrier.rowLower(), numberRows, saveLower + numberColumns);
CoinMemcpyN(barrier.columnUpper(), numberColumns, saveUpper);
CoinMemcpyN(barrier.rowUpper(), numberRows, saveUpper + numberColumns);
barrier.fixFixed();
saveModel2 = model2;
}
barrier.returnModel(*model2);
double * rowPrimal = new double [numberRows];
double * columnPrimal = new double [numberColumns];
double * rowDual = new double [numberRows];
double * columnDual = new double [numberColumns];
// move solutions other way
CoinMemcpyN(model2->primalRowSolution(),
numberRows, rowPrimal);
CoinMemcpyN(model2->dualRowSolution(),
numberRows, rowDual);
CoinMemcpyN(model2->primalColumnSolution(),
numberColumns, columnPrimal);
CoinMemcpyN(model2->dualColumnSolution(),
numberColumns, columnDual);
if (saveModel2) {
// do presolve
model2 = pinfo2.presolvedModel(*model2, 1.0e-8,
false, 5, true);
}
if (barrierStatus < 4 && crossover) {
// make sure no status left
model2->createStatus();
// solve
model2->setPerturbation(100);
// throw some into basis
{
int numberRows = model2->numberRows();
int numberColumns = model2->numberColumns();
double * dsort = new double[numberColumns];
int * sort = new int[numberColumns];
int n = 0;
const double * columnLower = model2->columnLower();
const double * columnUpper = model2->columnUpper();
const double * primalSolution = model2->primalColumnSolution();
double tolerance = 10.0 * primalTolerance_;
int i;
for ( i = 0; i < numberRows; i++)
model2->setRowStatus(i, superBasic);
for ( i = 0; i < numberColumns; i++) {
double distance = CoinMin(columnUpper[i] - primalSolution[i],
primalSolution[i] - columnLower[i]);
if (distance > tolerance) {
dsort[n] = -distance;
sort[n++] = i;
model2->setStatus(i, superBasic);
} else if (distance > primalTolerance_) {
model2->setStatus(i, superBasic);
} else if (primalSolution[i] <= columnLower[i] + primalTolerance_) {
model2->setStatus(i, atLowerBound);
} else {
model2->setStatus(i, atUpperBound);
}
}
CoinSort_2(dsort, dsort + n, sort);
n = CoinMin(numberRows, n);
for ( i = 0; i < n; i++) {
int iColumn = sort[i];
model2->setStatus(iColumn, basic);
}
delete [] sort;
delete [] dsort;
}
if (gap < 1.0e-3 * (static_cast<double> (numberRows + numberColumns))) {
int numberRows = model2->numberRows();
int numberColumns = model2->numberColumns();
// just primal values pass
double saveScale = model2->objectiveScale();
model2->setObjectiveScale(1.0e-3);
model2->primal(2);
model2->setObjectiveScale(saveScale);
// save primal solution and copy back dual
CoinMemcpyN(model2->primalRowSolution(),
numberRows, rowPrimal);
CoinMemcpyN(rowDual,
numberRows, model2->dualRowSolution());
CoinMemcpyN(model2->primalColumnSolution(),
numberColumns, columnPrimal);
CoinMemcpyN(columnDual,
numberColumns, model2->dualColumnSolution());
//model2->primal(1);
// clean up reduced costs and flag variables
{
double * dj = model2->dualColumnSolution();
double * cost = model2->objective();
double * saveCost = new double[numberColumns];
CoinMemcpyN(cost, numberColumns, saveCost);
double * saveLower = new double[numberColumns];
double * lower = model2->columnLower();
CoinMemcpyN(lower, numberColumns, saveLower);
double * saveUpper = new double[numberColumns];
double * upper = model2->columnUpper();
CoinMemcpyN(upper, numberColumns, saveUpper);
int i;
double tolerance = 10.0 * dualTolerance_;
for ( i = 0; i < numberColumns; i++) {
if (model2->getStatus(i) == basic) {
dj[i] = 0.0;
} else if (model2->getStatus(i) == atLowerBound) {
if (optimizationDirection_ * dj[i] < tolerance) {
if (optimizationDirection_ * dj[i] < 0.0) {
//if (dj[i]<-1.0e-3)
//printf("bad dj at lb %d %g\n",i,dj[i]);
cost[i] -= dj[i];
dj[i] = 0.0;
}
} else {
upper[i] = lower[i];
}
} else if (model2->getStatus(i) == atUpperBound) {
if (optimizationDirection_ * dj[i] > tolerance) {
if (optimizationDirection_ * dj[i] > 0.0) {
//if (dj[i]>1.0e-3)
//printf("bad dj at ub %d %g\n",i,dj[i]);
cost[i] -= dj[i];
dj[i] = 0.0;
}
} else {
lower[i] = upper[i];
}
}
}
// just dual values pass
//model2->setLogLevel(63);
//model2->setFactorizationFrequency(1);
model2->dual(2);
CoinMemcpyN(saveCost, numberColumns, cost);
delete [] saveCost;
CoinMemcpyN(saveLower, numberColumns, lower);
delete [] saveLower;
CoinMemcpyN(saveUpper, numberColumns, upper);
delete [] saveUpper;
}
// and finish
// move solutions
CoinMemcpyN(rowPrimal,
numberRows, model2->primalRowSolution());
CoinMemcpyN(columnPrimal,
numberColumns, model2->primalColumnSolution());
}
// double saveScale = model2->objectiveScale();
// model2->setObjectiveScale(1.0e-3);
// model2->primal(2);
// model2->setObjectiveScale(saveScale);
model2->primal(1);
} else if (barrierStatus == 4 && crossover) {
// memory problems
model2->setPerturbation(savePerturbation);
model2->createStatus();
model2->dual();
}
model2->setMaximumIterations(saveMaxIts);
delete [] rowPrimal;
delete [] columnPrimal;
delete [] rowDual;
delete [] columnDual;
if (saveLower) {
pinfo2.postsolve(true);
delete model2;
model2 = saveModel2;
int numberRows = model2->numberRows();
int numberColumns = model2->numberColumns();
CoinMemcpyN(saveLower, numberColumns, model2->columnLower());
CoinMemcpyN(saveLower + numberColumns, numberRows, model2->rowLower());
delete [] saveLower;
CoinMemcpyN(saveUpper, numberColumns, model2->columnUpper());
CoinMemcpyN(saveUpper + numberColumns, numberRows, model2->rowUpper());
delete [] saveUpper;
model2->primal(1);
}
model2->setPerturbation(savePerturbation);
return model2->status();
}
/* For strong branching. On input lower and upper are new bounds
while on output they are objective function values (>1.0e50 infeasible).
Return code is 0 if nothing interesting, -1 if infeasible both
ways and +1 if infeasible one way (check values to see which one(s))
*/
int ClpSimplex::strongBranching(int numberVariables, const int * variables,
double * newLower, double * newUpper,
double ** outputSolution,
int * outputStatus, int * outputIterations,
bool stopOnFirstInfeasible,
bool alwaysFinish,
int startFinishOptions)
{
return static_cast<ClpSimplexDual *> (this)->strongBranching(numberVariables, variables,
newLower, newUpper, outputSolution,
outputStatus, outputIterations,
stopOnFirstInfeasible,
alwaysFinish, startFinishOptions);
}
#endif
/* Borrow model. This is so we dont have to copy large amounts
of data around. It assumes a derived class wants to overwrite
an empty model with a real one - while it does an algorithm.
This is same as ClpModel one, but sets scaling on etc. */
void
ClpSimplex::borrowModel(ClpModel & otherModel)
{
ClpModel::borrowModel(otherModel);
createStatus();
//ClpDualRowSteepest steep1;
//setDualRowPivotAlgorithm(steep1);
//ClpPrimalColumnSteepest steep2;
//setPrimalColumnPivotAlgorithm(steep2);
}
void
ClpSimplex::borrowModel(ClpSimplex & otherModel)
{
ClpModel::borrowModel(otherModel);
createStatus();
dualBound_ = otherModel.dualBound_;
dualTolerance_ = otherModel.dualTolerance_;
primalTolerance_ = otherModel.primalTolerance_;
delete dualRowPivot_;
dualRowPivot_ = otherModel.dualRowPivot_->clone(true);
dualRowPivot_->setModel(this);
delete primalColumnPivot_;
primalColumnPivot_ = otherModel.primalColumnPivot_->clone(true);
primalColumnPivot_->setModel(this);
perturbation_ = otherModel.perturbation_;
moreSpecialOptions_ = otherModel.moreSpecialOptions_;
automaticScale_ = otherModel.automaticScale_;
maximumPerturbationSize_ = otherModel.maximumPerturbationSize_;
perturbationArray_ = otherModel.perturbationArray_;
}
/// Saves scalars for ClpSimplex
typedef struct {
double optimizationDirection;
double dblParam[ClpLastDblParam];
double objectiveValue;
double dualBound;
double dualTolerance;
double primalTolerance;
double sumDualInfeasibilities;
double sumPrimalInfeasibilities;
double infeasibilityCost;
int numberRows;
int numberColumns;
int intParam[ClpLastIntParam];
int numberIterations;
int problemStatus;
int maximumIterations;
int lengthNames;
int numberDualInfeasibilities;
int numberDualInfeasibilitiesWithoutFree;
int numberPrimalInfeasibilities;
int numberRefinements;
int scalingFlag;
int algorithm;
unsigned int specialOptions;
int dualPivotChoice;
int primalPivotChoice;
int matrixStorageChoice;
} Clp_scalars;
#ifndef SLIM_NOIO
int outDoubleArray(double * array, int length, FILE * fp)
{
size_t numberWritten;
if (array && length) {
numberWritten = fwrite(&length, sizeof(int), 1, fp);
if (numberWritten != 1)
return 1;
numberWritten = fwrite(array, sizeof(double), length, fp);
if (numberWritten != static_cast<size_t>(length))
return 1;
} else {
length = 0;
numberWritten = fwrite(&length, sizeof(int), 1, fp);
if (numberWritten != 1)
return 1;
}
return 0;
}
// Save model to file, returns 0 if success
int
ClpSimplex::saveModel(const char * fileName)
{
FILE * fp = fopen(fileName, "wb");
if (fp) {
Clp_scalars scalars;
size_t numberWritten;
// Fill in scalars
scalars.optimizationDirection = optimizationDirection_;
CoinMemcpyN( dblParam_, ClpLastDblParam, scalars.dblParam);
scalars.objectiveValue = objectiveValue_;
scalars.dualBound = dualBound_;
scalars.dualTolerance = dualTolerance_;
scalars.primalTolerance = primalTolerance_;
scalars.sumDualInfeasibilities = sumDualInfeasibilities_;
scalars.sumPrimalInfeasibilities = sumPrimalInfeasibilities_;
scalars.infeasibilityCost = infeasibilityCost_;
scalars.numberRows = numberRows_;
scalars.numberColumns = numberColumns_;
CoinMemcpyN( intParam_, ClpLastIntParam, scalars.intParam);
scalars.numberIterations = numberIterations_;
scalars.problemStatus = problemStatus_;
scalars.maximumIterations = maximumIterations();
scalars.lengthNames = lengthNames_;
scalars.numberDualInfeasibilities = numberDualInfeasibilities_;
scalars.numberDualInfeasibilitiesWithoutFree
= numberDualInfeasibilitiesWithoutFree_;
scalars.numberPrimalInfeasibilities = numberPrimalInfeasibilities_;
scalars.numberRefinements = numberRefinements_;
scalars.scalingFlag = scalingFlag_;
scalars.algorithm = algorithm_;
scalars.specialOptions = specialOptions_;
scalars.dualPivotChoice = dualRowPivot_->type();
scalars.primalPivotChoice = primalColumnPivot_->type();
scalars.matrixStorageChoice = matrix_->type();
// put out scalars
numberWritten = fwrite(&scalars, sizeof(Clp_scalars), 1, fp);
if (numberWritten != 1)
return 1;
size_t length;
#ifndef CLP_NO_STD
int i;
// strings
for (i = 0; i < ClpLastStrParam; i++) {
length = strParam_[i].size();
numberWritten = fwrite(&length, sizeof(int), 1, fp);
if (numberWritten != 1)
return 1;
if (length) {
numberWritten = fwrite(strParam_[i].c_str(), length, 1, fp);
if (numberWritten != 1)
return 1;
}
}
#endif
// arrays - in no particular order
if (outDoubleArray(rowActivity_, numberRows_, fp))
return 1;
if (outDoubleArray(columnActivity_, numberColumns_, fp))
return 1;
if (outDoubleArray(dual_, numberRows_, fp))
return 1;
if (outDoubleArray(reducedCost_, numberColumns_, fp))
return 1;
if (outDoubleArray(rowLower_, numberRows_, fp))
return 1;
if (outDoubleArray(rowUpper_, numberRows_, fp))
return 1;
if (outDoubleArray(objective(), numberColumns_, fp))
return 1;
if (outDoubleArray(rowObjective_, numberRows_, fp))
return 1;
if (outDoubleArray(columnLower_, numberColumns_, fp))
return 1;
if (outDoubleArray(columnUpper_, numberColumns_, fp))
return 1;
if (ray_) {
if (problemStatus_ == 1) {
if (outDoubleArray(ray_, numberRows_, fp))
return 1;
} else if (problemStatus_ == 2) {
if (outDoubleArray(ray_, numberColumns_, fp))
return 1;
} else {
if (outDoubleArray(NULL, 0, fp))
return 1;
}
} else {
if (outDoubleArray(NULL, 0, fp))
return 1;
}
if (status_ && (numberRows_ + numberColumns_) > 0) {
length = numberRows_ + numberColumns_;
numberWritten = fwrite(&length, sizeof(int), 1, fp);
if (numberWritten != 1)
return 1;
numberWritten = fwrite(status_, sizeof(char), length, fp);
if (numberWritten != length)
return 1;
} else {
length = 0;
numberWritten = fwrite(&length, sizeof(int), 1, fp);
if (numberWritten != 1)
return 1;
}
#ifndef CLP_NO_STD
if (lengthNames_) {
char * array =
new char[CoinMax(numberRows_, numberColumns_)*(lengthNames_+1)];
char * put = array;
CoinAssert (numberRows_ == static_cast<int> (rowNames_.size()));
for (i = 0; i < numberRows_; i++) {
assert(static_cast<int>(rowNames_[i].size()) <= lengthNames_);
strcpy(put, rowNames_[i].c_str());
put += lengthNames_ + 1;
}
numberWritten = fwrite(array, lengthNames_ + 1, numberRows_, fp);
if (numberWritten != static_cast<size_t>(numberRows_))
return 1;
put = array;
CoinAssert (numberColumns_ == static_cast<int> (columnNames_.size()));
for (i = 0; i < numberColumns_; i++) {
assert(static_cast<int> (columnNames_[i].size()) <= lengthNames_);
strcpy(put, columnNames_[i].c_str());
put += lengthNames_ + 1;
}
numberWritten = fwrite(array, lengthNames_ + 1, numberColumns_, fp);
if (numberWritten != static_cast<size_t>(numberColumns_))
return 1;
delete [] array;
}
#endif
// integers
if (integerType_) {
int marker = 1;
numberWritten = fwrite(&marker, sizeof(int), 1, fp);
numberWritten = fwrite(integerType_, 1, numberColumns_, fp);
if (numberWritten != static_cast<size_t>(numberColumns_))
return 1;
} else {
int marker = 0;
numberWritten = fwrite(&marker, sizeof(int), 1, fp);
}
// just standard type at present
assert (matrix_->type() == 1);
CoinAssert (matrix_->getNumCols() == numberColumns_);
CoinAssert (matrix_->getNumRows() == numberRows_);
// we are going to save with gaps
length = matrix_->getVectorStarts()[numberColumns_-1]
+ matrix_->getVectorLengths()[numberColumns_-1];
numberWritten = fwrite(&length, sizeof(int), 1, fp);
if (numberWritten != 1)
return 1;
numberWritten = fwrite(matrix_->getElements(),
sizeof(double), length, fp);
if (numberWritten != length)
return 1;
numberWritten = fwrite(matrix_->getIndices(),
sizeof(int), length, fp);
if (numberWritten != length)
return 1;
numberWritten = fwrite(matrix_->getVectorStarts(),
sizeof(int), numberColumns_ + 1, fp);
if (numberWritten != static_cast<size_t>(numberColumns_) + 1)
return 1;
numberWritten = fwrite(matrix_->getVectorLengths(),
sizeof(int), numberColumns_, fp);
if (numberWritten != static_cast<size_t>(numberColumns_))
return 1;
// finished
fclose(fp);
return 0;
} else {
return -1;
}
}
int inDoubleArray(double * &array, int length, FILE * fp)
{
size_t numberRead;
int length2;
numberRead = fread(&length2, sizeof(int), 1, fp);
if (numberRead != 1)
return 1;
if (length2) {
// lengths must match
if (length != length2)
return 2;
array = new double[length];
numberRead = fread(array, sizeof(double), length, fp);
if (numberRead != static_cast<size_t>(length))
return 1;
}
return 0;
}
/* Restore model from file, returns 0 if success,
deletes current model */
int
ClpSimplex::restoreModel(const char * fileName)
{
FILE * fp = fopen(fileName, "rb");
if (fp) {
// Get rid of current model
// save event handler in case already set
ClpEventHandler * handler = eventHandler_->clone();
ClpModel::gutsOfDelete(0);
eventHandler_ = handler;
gutsOfDelete(0);
int i;
for (i = 0; i < 6; i++) {
rowArray_[i] = NULL;
columnArray_[i] = NULL;
}
// get an empty factorization so we can set tolerances etc
getEmptyFactorization();
// Say sparse
factorization_->sparseThreshold(1);
Clp_scalars scalars;
size_t numberRead;
// get scalars
numberRead = fread(&scalars, sizeof(Clp_scalars), 1, fp);
if (numberRead != 1)
return 1;
// Fill in scalars
optimizationDirection_ = scalars.optimizationDirection;
CoinMemcpyN( scalars.dblParam, ClpLastDblParam, dblParam_);
objectiveValue_ = scalars.objectiveValue;
dualBound_ = scalars.dualBound;
dualTolerance_ = scalars.dualTolerance;
primalTolerance_ = scalars.primalTolerance;
sumDualInfeasibilities_ = scalars.sumDualInfeasibilities;
sumPrimalInfeasibilities_ = scalars.sumPrimalInfeasibilities;
infeasibilityCost_ = scalars.infeasibilityCost;
numberRows_ = scalars.numberRows;
numberColumns_ = scalars.numberColumns;
CoinMemcpyN( scalars.intParam, ClpLastIntParam, intParam_);
numberIterations_ = scalars.numberIterations;
problemStatus_ = scalars.problemStatus;
setMaximumIterations(scalars.maximumIterations);
lengthNames_ = scalars.lengthNames;
numberDualInfeasibilities_ = scalars.numberDualInfeasibilities;
numberDualInfeasibilitiesWithoutFree_
= scalars.numberDualInfeasibilitiesWithoutFree;
numberPrimalInfeasibilities_ = scalars.numberPrimalInfeasibilities;
numberRefinements_ = scalars.numberRefinements;
scalingFlag_ = scalars.scalingFlag;
algorithm_ = scalars.algorithm;
specialOptions_ = scalars.specialOptions;
// strings
CoinBigIndex length;
#ifndef CLP_NO_STD
for (i = 0; i < ClpLastStrParam; i++) {
numberRead = fread(&length, sizeof(int), 1, fp);
if (numberRead != 1)
return 1;
if (length) {
char * array = new char[length+1];
numberRead = fread(array, length, 1, fp);
if (numberRead != 1)
return 1;
array[length] = '\0';
strParam_[i] = array;
delete [] array;
}
}
#endif
// arrays - in no particular order
if (inDoubleArray(rowActivity_, numberRows_, fp))
return 1;
if (inDoubleArray(columnActivity_, numberColumns_, fp))
return 1;
if (inDoubleArray(dual_, numberRows_, fp))
return 1;
if (inDoubleArray(reducedCost_, numberColumns_, fp))
return 1;
if (inDoubleArray(rowLower_, numberRows_, fp))
return 1;
if (inDoubleArray(rowUpper_, numberRows_, fp))
return 1;
double * objective = NULL;
if (inDoubleArray(objective, numberColumns_, fp))
return 1;
delete objective_;
objective_ = new ClpLinearObjective(objective, numberColumns_);
delete [] objective;
if (inDoubleArray(rowObjective_, numberRows_, fp))
return 1;
if (inDoubleArray(columnLower_, numberColumns_, fp))
return 1;
if (inDoubleArray(columnUpper_, numberColumns_, fp))
return 1;
if (problemStatus_ == 1) {
if (inDoubleArray(ray_, numberRows_, fp))
return 1;
} else if (problemStatus_ == 2) {
if (inDoubleArray(ray_, numberColumns_, fp))
return 1;
} else {
// ray should be null
numberRead = fread(&length, sizeof(int), 1, fp);
if (numberRead != 1)
return 1;
if (length)
return 2;
}
delete [] status_;
status_ = NULL;
// status region
numberRead = fread(&length, sizeof(int), 1, fp);
if (numberRead != 1)
return 1;
if (length) {
if (length != numberRows_ + numberColumns_)
return 1;
status_ = new char unsigned[length];
numberRead = fread(status_, sizeof(char), length, fp);
if (numberRead != static_cast<size_t>(length))
return 1;
}
#ifndef CLP_NO_STD
if (lengthNames_) {
char * array =
new char[CoinMax(numberRows_, numberColumns_)*(lengthNames_+1)];
char * get = array;
numberRead = fread(array, lengthNames_ + 1, numberRows_, fp);
if (numberRead != static_cast<size_t>(numberRows_))
return 1;
rowNames_ = std::vector<std::string> ();
rowNames_.resize(numberRows_);
for (i = 0; i < numberRows_; i++) {
rowNames_.push_back(get);
get += lengthNames_ + 1;
}
get = array;
numberRead = fread(array, lengthNames_ + 1, numberColumns_, fp);
if (numberRead != static_cast<size_t>(numberColumns_))
return 1;
columnNames_ = std::vector<std::string> ();
columnNames_.resize(numberColumns_);
for (i = 0; i < numberColumns_; i++) {
columnNames_.push_back(get);
get += lengthNames_ + 1;
}
delete [] array;
}
#endif
// integers
int ifInteger;
delete [] integerType_;
numberRead = fread(&ifInteger, sizeof(int), 1, fp);
// But try and stay compatible with previous version
bool alreadyGotLength = false;
if (numberRead != 1)
return 1;
if (ifInteger == 1) {
integerType_ = new char [numberColumns_];
numberRead = fread(integerType_, 1, numberColumns_, fp);
if (numberRead != static_cast<size_t>(numberColumns_))
return 1;
} else {
integerType_ = NULL;
if (ifInteger) {
// probably old style save
alreadyGotLength = true;
length = ifInteger;
}
}
// Pivot choices
assert(scalars.dualPivotChoice > 0 && (scalars.dualPivotChoice & 63) < 3);
delete dualRowPivot_;
switch ((scalars.dualPivotChoice & 63)) {
default:
printf("Need another dualPivot case %d\n", scalars.dualPivotChoice & 63);
case 1:
// Dantzig
dualRowPivot_ = new ClpDualRowDantzig();
break;
case 2:
// Steepest - use mode
dualRowPivot_ = new ClpDualRowSteepest(scalars.dualPivotChoice >> 6);
break;
}
assert(scalars.primalPivotChoice > 0 && (scalars.primalPivotChoice & 63) < 3);
delete primalColumnPivot_;
switch ((scalars.primalPivotChoice & 63)) {
default:
printf("Need another primalPivot case %d\n",
scalars.primalPivotChoice & 63);
case 1:
// Dantzig
primalColumnPivot_ = new ClpPrimalColumnDantzig();
break;
case 2:
// Steepest - use mode
primalColumnPivot_
= new ClpPrimalColumnSteepest(scalars.primalPivotChoice >> 6);
break;
}
assert(scalars.matrixStorageChoice == 1);
delete matrix_;
// get arrays
if (!alreadyGotLength) {
numberRead = fread(&length, sizeof(int), 1, fp);
if (numberRead != 1)
return 1;
}
double * elements = new double[length];
int * indices = new int[length];
CoinBigIndex * starts = new CoinBigIndex[numberColumns_+1];
int * lengths = new int[numberColumns_];
numberRead = fread(elements, sizeof(double), length, fp);
if (numberRead != static_cast<size_t>(length))
return 1;
numberRead = fread(indices, sizeof(int), length, fp);
if (numberRead != static_cast<size_t>(length))
return 1;
numberRead = fread(starts, sizeof(int), numberColumns_ + 1, fp);
if (numberRead != static_cast<size_t>(numberColumns_) + 1)
return 1;
numberRead = fread(lengths, sizeof(int), numberColumns_, fp);
if (numberRead != static_cast<size_t>(numberColumns_))
return 1;
// assign matrix
CoinPackedMatrix * matrix = new CoinPackedMatrix();
matrix->setExtraGap(0.0);
matrix->setExtraMajor(0.0);
// Pack down
length = 0;
for (i = 0; i < numberColumns_; i++) {
int start = starts[i];
starts[i] = length;
for (CoinBigIndex j = start; j < start + lengths[i]; j++) {
elements[length] = elements[j];
indices[length++] = indices[j];
}
lengths[i] = length - starts[i];
}
starts[numberColumns_] = length;
matrix->assignMatrix(true, numberRows_, numberColumns_,
length, elements, indices, starts, lengths);
// and transfer to Clp
matrix_ = new ClpPackedMatrix(matrix);
// finished
fclose(fp);
return 0;
} else {
return -1;
}
return 0;
}
#endif
// value of incoming variable (in Dual)
double
ClpSimplex::valueIncomingDual() const
{
// Need value of incoming for list of infeasibilities as may be infeasible
double valueIncoming = (dualOut_ / alpha_) * directionOut_;
if (directionIn_ == -1)
valueIncoming = upperIn_ - valueIncoming;
else
valueIncoming = lowerIn_ - valueIncoming;
return valueIncoming;
}
// Sanity check on input data - returns true if okay
bool
ClpSimplex::sanityCheck()
{
// bad if empty
if (!numberColumns_ || ((!numberRows_ || !matrix_->getNumElements()) && objective_->type() < 2)) {
int infeasNumber[2];
double infeasSum[2];
problemStatus_ = emptyProblem(infeasNumber, infeasSum, false);
numberDualInfeasibilities_ = infeasNumber[0];
sumDualInfeasibilities_ = infeasSum[0];
numberPrimalInfeasibilities_ = infeasNumber[1];
sumPrimalInfeasibilities_ = infeasSum[1];
return false;
}
int numberBad ;
double largestBound, smallestBound, minimumGap;
double smallestObj, largestObj;
int firstBad;
int modifiedBounds = 0;
int i;
numberBad = 0;
firstBad = -1;
minimumGap = 1.0e100;
smallestBound = 1.0e100;
largestBound = 0.0;
smallestObj = 1.0e100;
largestObj = 0.0;
// If bounds are too close - fix
double fixTolerance = primalTolerance_;
if (fixTolerance < 2.0e-8)
fixTolerance *= 1.1;
for (i = numberColumns_; i < numberColumns_ + numberRows_; i++) {
double value;
value = fabs(cost_[i]);
if (value > 1.0e50) {
numberBad++;
if (firstBad < 0)
firstBad = i;
} else if (value) {
if (value > largestObj)
largestObj = value;
if (value < smallestObj)
smallestObj = value;
}
value = upper_[i] - lower_[i];
if (value < -primalTolerance_) {
numberBad++;
if (firstBad < 0)
firstBad = i;
} else if (value <= fixTolerance) {
if (value) {
// modify
upper_[i] = lower_[i];
modifiedBounds++;
}
} else {
if (value < minimumGap)
minimumGap = value;
}
if (lower_[i] > -1.0e100 && lower_[i]) {
value = fabs(lower_[i]);
if (value > largestBound)
largestBound = value;
if (value < smallestBound)
smallestBound = value;
}
if (upper_[i] < 1.0e100 && upper_[i]) {
value = fabs(upper_[i]);
if (value > largestBound)
largestBound = value;
if (value < smallestBound)
smallestBound = value;
}
}
if (largestBound)
handler_->message(CLP_RIMSTATISTICS3, messages_)
<< smallestBound
<< largestBound
<< minimumGap
<< CoinMessageEol;
minimumGap = 1.0e100;
smallestBound = 1.0e100;
largestBound = 0.0;
for (i = 0; i < numberColumns_; i++) {
double value;
value = fabs(cost_[i]);
if (value > 1.0e50) {
numberBad++;
if (firstBad < 0)
firstBad = i;
} else if (value) {
if (value > largestObj)
largestObj = value;
if (value < smallestObj)
smallestObj = value;
}
value = upper_[i] - lower_[i];
if (value < -primalTolerance_) {
numberBad++;
if (firstBad < 0)
firstBad = i;
} else if (value <= fixTolerance) {
if (value) {
// modify
upper_[i] = lower_[i];
modifiedBounds++;
}
} else {
if (value < minimumGap)
minimumGap = value;
}
if (lower_[i] > -1.0e100 && lower_[i]) {
value = fabs(lower_[i]);
if (value > largestBound)
largestBound = value;
if (value < smallestBound)
smallestBound = value;
}
if (upper_[i] < 1.0e100 && upper_[i]) {
value = fabs(upper_[i]);
if (value > largestBound)
largestBound = value;
if (value < smallestBound)
smallestBound = value;
}
}
char rowcol[] = {'R', 'C'};
if (numberBad) {
handler_->message(CLP_BAD_BOUNDS, messages_)
<< numberBad
<< rowcol[isColumn(firstBad)] << sequenceWithin(firstBad)
<< CoinMessageEol;
problemStatus_ = 4;
return false;
}
if (modifiedBounds)
handler_->message(CLP_MODIFIEDBOUNDS, messages_)
<< modifiedBounds
<< CoinMessageEol;
handler_->message(CLP_RIMSTATISTICS1, messages_)
<< smallestObj
<< largestObj
<< CoinMessageEol;
if (largestBound)
handler_->message(CLP_RIMSTATISTICS2, messages_)
<< smallestBound
<< largestBound
<< minimumGap
<< CoinMessageEol;
return true;
}
// Set up status array (for OsiClp)
void
ClpSimplex::createStatus()
{
if(!status_)
status_ = new unsigned char [numberColumns_+numberRows_];
memset(status_, 0, (numberColumns_ + numberRows_)*sizeof(char));
int i;
// set column status to one nearest zero
for (i = 0; i < numberColumns_; i++) {
#if 0
if (columnLower_[i] >= 0.0) {
setColumnStatus(i, atLowerBound);
} else if (columnUpper_[i] <= 0.0) {
setColumnStatus(i, atUpperBound);
} else if (columnLower_[i] < -1.0e20 && columnUpper_[i] > 1.0e20) {
// free
setColumnStatus(i, isFree);
} else if (fabs(columnLower_[i]) < fabs(columnUpper_[i])) {
setColumnStatus(i, atLowerBound);
} else {
setColumnStatus(i, atUpperBound);
}
#else
setColumnStatus(i, atLowerBound);
#endif
}
for (i = 0; i < numberRows_; i++) {
setRowStatus(i, basic);
}
}
/* Sets up all slack basis and resets solution to
as it was after initial load or readMps */
void ClpSimplex::allSlackBasis(bool resetSolution)
{
createStatus();
if (resetSolution) {
// put back to as it was originally
int i;
// set column status to one nearest zero
// But set value to zero if lb <0.0 and ub>0.0
for (i = 0; i < numberColumns_; i++) {
if (columnLower_[i] >= 0.0) {
columnActivity_[i] = columnLower_[i];
setColumnStatus(i, atLowerBound);
} else if (columnUpper_[i] <= 0.0) {
columnActivity_[i] = columnUpper_[i];
setColumnStatus(i, atUpperBound);
} else if (columnLower_[i] < -1.0e20 && columnUpper_[i] > 1.0e20) {
// free
columnActivity_[i] = 0.0;
setColumnStatus(i, isFree);
} else if (fabs(columnLower_[i]) < fabs(columnUpper_[i])) {
columnActivity_[i] = 0.0;
setColumnStatus(i, atLowerBound);
} else {
columnActivity_[i] = 0.0;
setColumnStatus(i, atUpperBound);
}
}
if (solution_) {
// do that as well
if (!columnScale_) {
for (i = 0; i < numberColumns_; i++) {
solution_[i] = columnActivity_[i];
}
} else {
double * inverseColumnScale = columnScale_ + numberColumns_;
for (i = 0; i < numberColumns_; i++) {
solution_[i] = columnActivity_[i] * (rhsScale_ * inverseColumnScale[i]);
}
}
}
}
}
/* Loads a problem (the constraints on the
rows are given by lower and upper bounds). If a pointer is 0 then the
following values are the default:
<ul>
<li> <code>colub</code>: all columns have upper bound infinity
<li> <code>collb</code>: all columns have lower bound 0
<li> <code>rowub</code>: all rows have upper bound infinity
<li> <code>rowlb</code>: all rows have lower bound -infinity
<li> <code>obj</code>: all variables have 0 objective coefficient
</ul>
*/
void
ClpSimplex::loadProblem ( const ClpMatrixBase& matrix,
const double* collb, const double* colub,
const double* obj,
const double* rowlb, const double* rowub,
const double * rowObjective)
{
ClpModel::loadProblem(matrix, collb, colub, obj, rowlb, rowub,
rowObjective);
createStatus();
}
void
ClpSimplex::loadProblem ( const CoinPackedMatrix& matrix,
const double* collb, const double* colub,
const double* obj,
const double* rowlb, const double* rowub,
const double * rowObjective)
{
ClpModel::loadProblem(matrix, collb, colub, obj, rowlb, rowub,
rowObjective);
createStatus();
}
/* Just like the other loadProblem() method except that the matrix is
given in a standard column major ordered format (without gaps). */
void
ClpSimplex::loadProblem ( const int numcols, const int numrows,
const CoinBigIndex* start, const int* index,
const double* value,
const double* collb, const double* colub,
const double* obj,
const double* rowlb, const double* rowub,
const double * rowObjective)
{
ClpModel::loadProblem(numcols, numrows, start, index, value,
collb, colub, obj, rowlb, rowub,
rowObjective);
createStatus();
}
#ifndef SLIM_NOIO
// This loads a model from a coinModel object - returns number of errors
int
ClpSimplex::loadProblem ( CoinModel & modelObject, bool /*keepSolution*/)
{
unsigned char * status = NULL;
double * psol = NULL;
double * dsol = NULL;
if (status_ && numberRows_ && numberRows_ == modelObject.numberRows() &&
numberColumns_ == modelObject.numberColumns()) {
status = new unsigned char [numberRows_+numberColumns_];
CoinMemcpyN(status_, numberRows_ + numberColumns_, status);
psol = new double [numberRows_+numberColumns_];
CoinMemcpyN(columnActivity_, numberColumns_, psol);
CoinMemcpyN(rowActivity_, numberRows_, psol + numberColumns_);
dsol = new double [numberRows_+numberColumns_];
CoinMemcpyN(reducedCost_, numberColumns_, dsol);
CoinMemcpyN(dual_, numberRows_, dsol + numberColumns_);
}
int returnCode = ClpModel::loadProblem(modelObject);
const int * integerType = modelObject.integerTypeArray();
if (integerType) {
for (int iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (integerType[iColumn])
setInteger(iColumn);
}
}
createStatus();
if (status) {
// copy back
CoinMemcpyN(status, numberRows_ + numberColumns_, status_);
CoinMemcpyN(psol, numberColumns_, columnActivity_);
CoinMemcpyN(psol + numberColumns_, numberRows_, rowActivity_);
CoinMemcpyN(dsol, numberColumns_, reducedCost_);
CoinMemcpyN(dsol + numberColumns_, numberRows_, dual_);
delete [] status;
delete [] psol;
delete [] dsol;
}
optimizationDirection_ = modelObject.optimizationDirection();
return returnCode;
}
#endif
void
ClpSimplex::loadProblem ( const int numcols, const int numrows,
const CoinBigIndex* start, const int* index,
const double* value, const int * length,
const double* collb, const double* colub,
const double* obj,
const double* rowlb, const double* rowub,
const double * rowObjective)
{
ClpModel::loadProblem(numcols, numrows, start, index, value, length,
collb, colub, obj, rowlb, rowub,
rowObjective);
createStatus();
}
#ifndef SLIM_NOIO
// Read an mps file from the given filename
int
ClpSimplex::readMps(const char *filename,
bool keepNames,
bool ignoreErrors)
{
int status = ClpModel::readMps(filename, keepNames, ignoreErrors);
createStatus();
return status;
}
// Read GMPL files from the given filenames
int
ClpSimplex::readGMPL(const char *filename, const char * dataName,
bool keepNames)
{
int status = ClpModel::readGMPL(filename, dataName, keepNames);
createStatus();
return status;
}
// Read file in LP format from file with name filename.
int
ClpSimplex::readLp(const char *filename, const double epsilon )
{
FILE *fp;
if (strcmp(filename,"-"))
fp = fopen(filename, "r");
else
fp = stdin;
if(!fp) {
printf("### ERROR: ClpSimplex::readLp(): Unable to open file %s for reading\n",
filename);
return(1);
}
CoinLpIO m;
m.readLp(fp, epsilon);
fclose(fp);
// set problem name
setStrParam(ClpProbName, m.getProblemName());
// no errors
loadProblem(*m.getMatrixByRow(), m.getColLower(), m.getColUpper(),
m.getObjCoefficients(), m.getRowLower(), m.getRowUpper());
if (m.integerColumns()) {
integerType_ = new char[numberColumns_];
CoinMemcpyN(m.integerColumns(), numberColumns_, integerType_);
} else {
integerType_ = NULL;
}
createStatus();
unsigned int maxLength = 0;
int iRow;
rowNames_ = std::vector<std::string> ();
columnNames_ = std::vector<std::string> ();
rowNames_.reserve(numberRows_);
for (iRow = 0; iRow < numberRows_; iRow++) {
const char * name = m.rowName(iRow);
if (name) {
maxLength = CoinMax(maxLength, static_cast<unsigned int> (strlen(name)));
rowNames_.push_back(name);
} else {
rowNames_.push_back("");
}
}
int iColumn;
columnNames_.reserve(numberColumns_);
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
const char * name = m.columnName(iColumn);
if (name) {
maxLength = CoinMax(maxLength, static_cast<unsigned int> (strlen(name)));
columnNames_.push_back(name);
} else {
columnNames_.push_back("");
}
}
lengthNames_ = static_cast<int> (maxLength);
return 0;
}
#endif
// Just check solution (for external use)
void
ClpSimplex::checkSolution(int setToBounds)
{
if (setToBounds) {
// Set all ones that look at bounds to bounds
bool changed = false;
int i;
for (i = 0; i < numberRows_; i++) {
double newValue = 0.0;
switch(getRowStatus(i)) {
case basic:
newValue = rowActivity_[i];
break;
case atUpperBound:
newValue = rowUpper_[i];
if (newValue > largeValue_) {
if (rowLower_[i] > -largeValue_) {
newValue = rowLower_[i];
setRowStatus(i, atLowerBound);
} else {
// say free
setRowStatus(i, isFree);
newValue = 0.0;
}
}
break;
case ClpSimplex::isFixed:
case atLowerBound:
newValue = rowLower_[i];
if (newValue < -largeValue_) {
if (rowUpper_[i] < largeValue_) {
newValue = rowUpper_[i];
setRowStatus(i, atUpperBound);
} else {
// say free
setRowStatus(i, isFree);
newValue = 0.0;
}
}
break;
case isFree:
newValue = rowActivity_[i];
break;
// not really free - fall through to superbasic
case superBasic:
if (rowUpper_[i] > largeValue_) {
if (rowLower_[i] > -largeValue_) {
newValue = rowLower_[i];
setRowStatus(i, atLowerBound);
} else {
// say free
setRowStatus(i, isFree);
newValue = 0.0;
}
} else {
if (rowLower_[i] > -largeValue_) {
// set to nearest
if (fabs(newValue - rowLower_[i])
< fabs(newValue - rowUpper_[i])) {
newValue = rowLower_[i];
setRowStatus(i, atLowerBound);
} else {
newValue = rowUpper_[i];
setRowStatus(i, atUpperBound);
}
} else {
newValue = rowUpper_[i];
setRowStatus(i, atUpperBound);
}
}
break;
}
if (fabs(newValue - rowActivity_[i]) > 1.0e-12) {
changed = true;
rowActivity_[i] = newValue;
}
}
for (i = 0; i < numberColumns_; i++) {
double newValue = 0.0;
switch(getColumnStatus(i)) {
case basic:
newValue = columnActivity_[i];
break;
case atUpperBound:
newValue = columnUpper_[i];
if (newValue > largeValue_) {
if (columnLower_[i] > -largeValue_) {
newValue = columnLower_[i];
setColumnStatus(i, atLowerBound);
} else {
// say free
setColumnStatus(i, isFree);
newValue = 0.0;
}
}
break;
case ClpSimplex::isFixed:
case atLowerBound:
newValue = columnLower_[i];
if (newValue < -largeValue_) {
if (columnUpper_[i] < largeValue_) {
newValue = columnUpper_[i];
setColumnStatus(i, atUpperBound);
} else {
// say free
setColumnStatus(i, isFree);
newValue = 0.0;
}
}
break;
case isFree:
newValue = columnActivity_[i];
break;
// not really free - fall through to superbasic
case superBasic:
if (columnUpper_[i] > largeValue_) {
if (columnLower_[i] > -largeValue_) {
newValue = columnLower_[i];
setColumnStatus(i, atLowerBound);
} else {
// say free
setColumnStatus(i, isFree);
newValue = 0.0;
}
} else {
if (columnLower_[i] > -largeValue_) {
// set to nearest
if (fabs(newValue - columnLower_[i])
< fabs(newValue - columnUpper_[i])) {
newValue = columnLower_[i];
setColumnStatus(i, atLowerBound);
} else {
newValue = columnUpper_[i];
setColumnStatus(i, atUpperBound);
}
} else {
newValue = columnUpper_[i];
setColumnStatus(i, atUpperBound);
}
}
break;
}
if (fabs(newValue - columnActivity_[i]) > 1.0e-12) {
changed = true;
columnActivity_[i] = newValue;
}
}
if (!changed && setToBounds == 1)
// no need to do anything
setToBounds = 0;
}
if (!setToBounds) {
// Just use column solution
CoinZeroN(rowActivity_, numberRows_);
matrix()->times(columnActivity_, rowActivity_) ;
// put in standard form
createRim(7 + 8 + 16 + 32);
dualTolerance_ = dblParam_[ClpDualTolerance];
primalTolerance_ = dblParam_[ClpPrimalTolerance];
checkPrimalSolution( rowActivityWork_, columnActivityWork_);
checkDualSolution();
} else {
startup(0, 0);
gutsOfSolution(NULL, NULL);
}
if (!numberDualInfeasibilities_ &&
!numberPrimalInfeasibilities_)
problemStatus_ = 0;
else
problemStatus_ = -1;
#ifdef CLP_DEBUG
int i;
double value = 0.0;
for (i = 0; i < numberRows_ + numberColumns_; i++)
value += dj_[i] * solution_[i];
printf("dual value %g, primal %g\n", value, objectiveValue());
#endif
// release extra memory
deleteRim(0);
}
// Check unscaled primal solution but allow for rounding error
void
ClpSimplex::checkUnscaledSolution()
{
if (problemStatus_==1 && matrix_->getNumElements()) {
const double * element = matrix_->getElements();
const CoinBigIndex * columnStart = matrix_->getVectorStarts();
const int * columnLength = matrix_->getVectorLengths();
const int * row = matrix_->getIndices();
memset(rowActivity_,0,numberRows_*sizeof(double));
double * sum = new double [numberRows_+100000];
memset(sum,0,numberRows_*sizeof(double));
// clean column activity
for (int i=0;i<numberColumns_;i++) {
double value = columnActivity_[i];
value = CoinMax(value,columnLower_[i]);
value = CoinMin(value,columnUpper_[i]);
//columnActivity_[i]=value;
if (value) {
for (CoinBigIndex j=columnStart[i];
j<columnStart[i]+columnLength[i];j++) {
double value2 = value*element[j];
int iRow = row[j];
assert (iRow>=0&&iRow<numberRows_);
rowActivity_[iRow] += value2;
sum[iRow]+=fabs(value2);
}
}
}
sumPrimalInfeasibilities_ = 0.0;
numberPrimalInfeasibilities_ = 0;
double sumPrimalInfeasibilities2 = 0.0;
int numberPrimalInfeasibilities2 = 0;
double fudgeFactor = 1.0e-12;
double fudgeFactor2 = 1.0e-12;
double tolerance = primalTolerance_;
for (int i=0;i<numberRows_;i++) {
double useTolerance = CoinMax(tolerance,fudgeFactor*sum[i]);
double value = rowActivity_[i];
useTolerance = CoinMax(useTolerance,fudgeFactor2*fabs(value));
if (value>rowUpper_[i]) {
sumPrimalInfeasibilities2 += value - rowUpper_[i];
numberPrimalInfeasibilities2++;
if (value>rowUpper_[i]+useTolerance) {
sumPrimalInfeasibilities_ += value - (rowUpper_[i]+useTolerance);
numberPrimalInfeasibilities_++;
}
} else if (value<rowLower_[i]) {
sumPrimalInfeasibilities2 -= value - rowLower_[i];
numberPrimalInfeasibilities2++;
if (value<rowLower_[i]-useTolerance) {
sumPrimalInfeasibilities_ -= value - (rowLower_[i]-useTolerance);
numberPrimalInfeasibilities_++;
}
}
}
char line[1000];
if (!numberPrimalInfeasibilities2) {
sprintf(line,"%d unscaled row infeasibilities - summing to %g",
numberPrimalInfeasibilities2,
sumPrimalInfeasibilities2);
handler_->message(CLP_GENERAL, messages_)
<< line
<< CoinMessageEol;
}
if (!numberPrimalInfeasibilities_) {
if (!numberDualInfeasibilities_)
problemStatus_=0;
} else {
sprintf(line,"%d relaxed row infeasibilities - summing to %g",
numberPrimalInfeasibilities_,
sumPrimalInfeasibilities_);
handler_->message(CLP_GENERAL, messages_)
<< line
<< CoinMessageEol;
}
delete [] sum;
}
}
/* Crash - at present just aimed at dual, returns
-2 if dual preferred and crash basis created
-1 if dual preferred and all slack basis preferred
0 if basis going in was not all slack
1 if primal preferred and all slack basis preferred
2 if primal preferred and crash basis created.
if gap between bounds <="gap" variables can be flipped
( If pivot -1 then can be made super basic!)
If "pivot" is
-1 No pivoting - always primal
0 No pivoting (so will just be choice of algorithm)
1 Simple pivoting e.g. gub
2 Mini iterations
3 Just throw all free variables in basis
*/
int
ClpSimplex::crash(double gap, int pivot)
{
//CoinAssert(!rowObjective_); // not coded
int iColumn;
int numberBad = 0;
int numberBasic = 0;
double dualTolerance = dblParam_[ClpDualTolerance];
//double primalTolerance=dblParam_[ClpPrimalTolerance];
int returnCode = 0;
// If no basis then make all slack one
if (!status_)
createStatus();
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (getColumnStatus(iColumn) == basic)
numberBasic++;
}
if (!numberBasic || pivot == 3) {
#if 0
int nFree=0;
for (int iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (columnLower_[iColumn] < -1.0e20 && columnUpper_[iColumn] > 1.0e20)
nFree++;
}
if (nFree>numberColumns_/10)
pivot=3;
#endif
if (pivot == 3) {
// Get column copy
CoinPackedMatrix * columnCopy = matrix();
const int * row = columnCopy->getIndices();
const CoinBigIndex * columnStart = columnCopy->getVectorStarts();
const int * columnLength = columnCopy->getVectorLengths();
const double * element = columnCopy->getElements();
int nFree=0;
for (int iColumn = 0; iColumn < numberColumns_; iColumn++) {
if (columnLower_[iColumn] < -1.0e20 && columnUpper_[iColumn] > 1.0e20) {
// find triangular row
int kRow=-1;
double largest=0.0;
double largestOther=0.0;
for (CoinBigIndex j=columnStart[iColumn];
j<columnStart[iColumn]+columnLength[iColumn];j++) {
int iSequence=row[j]+numberColumns_;
if (!flagged(iSequence)) {
if (fabs(element[j])>largest) {
kRow=row[j];
largest=fabs(element[j]);
}
} else {
if (fabs(element[j])>largestOther) {
largestOther=fabs(element[j]);
}
}
}
if (kRow>=0&&largest*2.5>=largestOther) {
nFree++;
setColumnStatus(iColumn, basic);
if (fabs(rowLower_[kRow]) < fabs(rowUpper_[kRow]))
setRowStatus(kRow, atLowerBound);
else
setRowStatus(kRow, atUpperBound);
for (CoinBigIndex j=columnStart[iColumn];
j<columnStart[iColumn]+columnLength[iColumn];j++) {
int iSequence=row[j]+numberColumns_;
setFlagged(iSequence);
}
}
}
}
if (nFree) {
for (int i=0;i<numberRows_;i++)
clearFlagged(i);
printf("%d free variables put in basis\n",nFree);
return 0;
}
}
// all slack
double * dj = new double [numberColumns_];
double * solution = columnActivity_;
const double * linearObjective = objective();
//double objectiveValue=0.0;
int iColumn;
double direction = optimizationDirection_;
// direction is actually scale out not scale in
if (direction)
direction = 1.0 / direction;
for (iColumn = 0; iColumn < numberColumns_; iColumn++)
dj[iColumn] = direction * linearObjective[iColumn];
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
// assume natural place is closest to zero
double lowerBound = columnLower_[iColumn];
double upperBound = columnUpper_[iColumn];
if (lowerBound > -1.0e20 || upperBound < 1.0e20) {
bool atLower;
if (fabs(upperBound) < fabs(lowerBound)) {
atLower = false;
setColumnStatus(iColumn, atUpperBound);
solution[iColumn] = upperBound;
} else {
atLower = true;
setColumnStatus(iColumn, atLowerBound);
solution[iColumn] = lowerBound;
}
if (dj[iColumn] < -dualTolerance_) {
// should be at upper bound
if (atLower) {
// can we flip
if (upperBound - lowerBound <= gap) {
columnActivity_[iColumn] = upperBound;
setColumnStatus(iColumn, atUpperBound);
} else if (pivot < 0) {
// set superbasic
columnActivity_[iColumn] = lowerBound + gap;
setColumnStatus(iColumn, superBasic);
} else if (dj[iColumn] < -dualTolerance) {
numberBad++;
}
}
} else if (dj[iColumn] > dualTolerance_) {
// should be at lower bound
if (!atLower) {
// can we flip
if (upperBound - lowerBound <= gap) {
columnActivity_[iColumn] = lowerBound;
setColumnStatus(iColumn, atLowerBound);
} else if (pivot < 0) {
// set superbasic
columnActivity_[iColumn] = upperBound - gap;
setColumnStatus(iColumn, superBasic);
} else if (dj[iColumn] > dualTolerance) {
numberBad++;
}
}
}
} else {
// free
setColumnStatus(iColumn, isFree);
if (fabs(dj[iColumn]) > dualTolerance)
numberBad++;
}
}
if (numberBad || pivot) {
if (pivot <= 0) {
delete [] dj;
returnCode = 1;
} else {
// see if can be made dual feasible with gubs etc
double * pi = new double[numberRows_];
memset (pi, 0, numberRows_ * sizeof(double));
int * way = new int[numberColumns_];
int numberIn = 0;
// Get column copy
CoinPackedMatrix * columnCopy = matrix();
// Get a row copy in standard format
CoinPackedMatrix copy;
copy.setExtraGap(0.0);
copy.setExtraMajor(0.0);
copy.reverseOrderedCopyOf(*columnCopy);
// get matrix data pointers
const int * column = copy.getIndices();
const CoinBigIndex * rowStart = copy.getVectorStarts();
const int * rowLength = copy.getVectorLengths();
const double * elementByRow = copy.getElements();
//const int * row = columnCopy->getIndices();
//const CoinBigIndex * columnStart = columnCopy->getVectorStarts();
//const int * columnLength = columnCopy->getVectorLengths();
//const double * element = columnCopy->getElements();
// if equality row and bounds mean artificial in basis bad
// then do anyway
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
// - if we want to reduce dj, + if we want to increase
int thisWay = 100;
double lowerBound = columnLower_[iColumn];
double upperBound = columnUpper_[iColumn];
if (upperBound > lowerBound) {
switch(getColumnStatus(iColumn)) {
case basic:
thisWay = 0;
case ClpSimplex::isFixed:
break;
case isFree:
case superBasic:
if (dj[iColumn] < -dualTolerance)
thisWay = 1;
else if (dj[iColumn] > dualTolerance)
thisWay = -1;
else
thisWay = 0;
break;
case atUpperBound:
if (dj[iColumn] > dualTolerance)
thisWay = -1;
else if (dj[iColumn] < -dualTolerance)
thisWay = -3;
else
thisWay = -2;
break;
case atLowerBound:
if (dj[iColumn] < -dualTolerance)
thisWay = 1;
else if (dj[iColumn] > dualTolerance)
thisWay = 3;
else
thisWay = 2;
break;
}
}
way[iColumn] = thisWay;
}
/*if (!numberBad)
printf("Was dual feasible before passes - rows %d\n",
numberRows_);*/
int lastNumberIn = -100000;
int numberPasses = 5;
while (numberIn > lastNumberIn + numberRows_ / 100) {
lastNumberIn = numberIn;
// we need to maximize chance of doing good
int iRow;
for (iRow = 0; iRow < numberRows_; iRow++) {
double lowerBound = rowLower_[iRow];
double upperBound = rowUpper_[iRow];
if (getRowStatus(iRow) == basic) {
// see if we can find a column to pivot on
int j;
// down is amount pi can go down
double maximumDown = COIN_DBL_MAX;
double maximumUp = COIN_DBL_MAX;
double minimumDown = 0.0;
double minimumUp = 0.0;
int iUp = -1;
int iDown = -1;
int iUpB = -1;
int iDownB = -1;
if (lowerBound < -1.0e20)
maximumUp = -1.0;
if (upperBound > 1.0e20)
maximumDown = -1.0;
for (j = rowStart[iRow]; j < rowStart[iRow] + rowLength[iRow]; j++) {
int iColumn = column[j];
double value = elementByRow[j];
double djValue = dj[iColumn];
/* way -
-3 - okay at upper bound with negative dj
-2 - marginal at upper bound with zero dj - can only decrease
-1 - bad at upper bound
0 - we can never pivot on this row
1 - bad at lower bound
2 - marginal at lower bound with zero dj - can only increase
3 - okay at lower bound with positive dj
100 - fine we can just ignore
*/
if (way[iColumn] != 100) {
switch(way[iColumn]) {
case -3:
if (value > 0.0) {
if (maximumDown * value > -djValue) {
maximumDown = -djValue / value;
iDown = iColumn;
}
} else {
if (-maximumUp * value > -djValue) {
maximumUp = djValue / value;
iUp = iColumn;
}
}
break;
case -2:
if (value > 0.0) {
maximumDown = 0.0;
} else {
maximumUp = 0.0;
}
break;
case -1:
// see if could be satisfied
// dj value > 0
if (value > 0.0) {
maximumDown = 0.0;
if (maximumUp * value < djValue - dualTolerance) {
maximumUp = 0.0; // would improve but not enough
} else {
if (minimumUp * value < djValue) {
minimumUp = djValue / value;
iUpB = iColumn;
}
}
} else {
maximumUp = 0.0;
if (-maximumDown * value < djValue - dualTolerance) {
maximumDown = 0.0; // would improve but not enough
} else {
if (-minimumDown * value < djValue) {
minimumDown = -djValue / value;
iDownB = iColumn;
}
}
}
break;
case 0:
maximumDown = -1.0;
maximumUp = -1.0;
break;
case 1:
// see if could be satisfied
// dj value < 0
if (value > 0.0) {
maximumUp = 0.0;
if (maximumDown * value < -djValue - dualTolerance) {
maximumDown = 0.0; // would improve but not enough
} else {
if (minimumDown * value < -djValue) {
minimumDown = -djValue / value;
iDownB = iColumn;
}
}
} else {
maximumDown = 0.0;
if (-maximumUp * value < -djValue - dualTolerance) {
maximumUp = 0.0; // would improve but not enough
} else {
if (-minimumUp * value < -djValue) {
minimumUp = djValue / value;
iUpB = iColumn;
}
}
}
break;
case 2:
if (value > 0.0) {
maximumUp = 0.0;
} else {
maximumDown = 0.0;
}
break;
case 3:
if (value > 0.0) {
if (maximumUp * value > djValue) {
maximumUp = djValue / value;
iUp = iColumn;
}
} else {
if (-maximumDown * value > djValue) {
maximumDown = -djValue / value;
iDown = iColumn;
}
}
break;
default:
break;
}
}
}
if (iUpB >= 0)
iUp = iUpB;
if (maximumUp <= dualTolerance || maximumUp < minimumUp)
iUp = -1;
if (iDownB >= 0)
iDown = iDownB;
if (maximumDown <= dualTolerance || maximumDown < minimumDown)
iDown = -1;
if (iUp >= 0 || iDown >= 0) {
// do something
if (iUp >= 0 && iDown >= 0) {
if (maximumDown > maximumUp)
iUp = -1;
}
double change;
int kColumn;
if (iUp >= 0) {
kColumn = iUp;
change = maximumUp;
// just do minimum if was dual infeasible
// ? only if maximum large?
if (minimumUp > 0.0)
change = minimumUp;
setRowStatus(iRow, atUpperBound);
} else {
kColumn = iDown;
change = -maximumDown;
// just do minimum if was dual infeasible
// ? only if maximum large?
if (minimumDown > 0.0)
change = -minimumDown;
setRowStatus(iRow, atLowerBound);
}
assert (fabs(change) < 1.0e200);
setColumnStatus(kColumn, basic);
numberIn++;
pi[iRow] = change;
for (j = rowStart[iRow]; j < rowStart[iRow] + rowLength[iRow]; j++) {
int iColumn = column[j];
double value = elementByRow[j];
double djValue = dj[iColumn] - change * value;
dj[iColumn] = djValue;
if (abs(way[iColumn]) == 1) {
numberBad--;
/*if (!numberBad)
printf("Became dual feasible at row %d out of %d\n",
iRow, numberRows_);*/
lastNumberIn = -1000000;
}
int thisWay = 100;
double lowerBound = columnLower_[iColumn];
double upperBound = columnUpper_[iColumn];
if (upperBound > lowerBound) {
switch(getColumnStatus(iColumn)) {
case basic:
thisWay = 0;
case isFixed:
break;
case isFree:
case superBasic:
if (djValue < -dualTolerance)
thisWay = 1;
else if (djValue > dualTolerance)
thisWay = -1;
else {
thisWay = 0;
}
break;
case atUpperBound:
if (djValue > dualTolerance) {
thisWay = -1;
} else if (djValue < -dualTolerance)
thisWay = -3;
else
thisWay = -2;
break;
case atLowerBound:
if (djValue < -dualTolerance) {
thisWay = 1;
} else if (djValue > dualTolerance)
thisWay = 3;
else
thisWay = 2;
break;
}
}
way[iColumn] = thisWay;
}
}
}
}
if (numberIn == lastNumberIn || numberBad || pivot < 2)
break;
if (!(--numberPasses))
break;
//printf("%d put in so far\n",numberIn);
}
// last attempt to flip
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
double lowerBound = columnLower_[iColumn];
double upperBound = columnUpper_[iColumn];
if (upperBound - lowerBound <= gap && upperBound > lowerBound) {
double djValue = dj[iColumn];
switch(getColumnStatus(iColumn)) {
case basic:
case ClpSimplex::isFixed:
break;
case isFree:
case superBasic:
break;
case atUpperBound:
if (djValue > dualTolerance) {
setColumnStatus(iColumn, atUpperBound);
solution[iColumn] = upperBound;
}
break;
case atLowerBound:
if (djValue < -dualTolerance) {
setColumnStatus(iColumn, atUpperBound);
solution[iColumn] = upperBound;
}
break;
}
}
}
delete [] pi;
delete [] dj;
delete [] way;
handler_->message(CLP_CRASH, messages_)
<< numberIn
<< numberBad
<< CoinMessageEol;
// could do elsewhere and could even if something done
// see if some MUST be in basis
if (!numberIn) {
}
returnCode = -1;
}
} else {
delete [] dj;
returnCode = -1;
}
//cleanStatus();
}
return returnCode;
}
/* Pivot in a variable and out a variable. Returns 0 if okay,
1 if inaccuracy forced re-factorization, -1 if would be singular.
Also updates primal/dual infeasibilities.
Assumes sequenceIn_ and pivotRow_ set and also directionIn and Out.
*/
int ClpSimplex::pivot()
{
// scaling not allowed
assert (!scalingFlag_);
// assume In_ and Out_ are correct and directionOut_ set
// (or In_ if flip
lowerIn_ = lower_[sequenceIn_];
valueIn_ = solution_[sequenceIn_];
upperIn_ = upper_[sequenceIn_];
dualIn_ = dj_[sequenceIn_];
lowerOut_ = lower_[sequenceOut_];
valueOut_ = solution_[sequenceOut_];
upperOut_ = upper_[sequenceOut_];
// for now assume primal is feasible (or in dual)
dualOut_ = dj_[sequenceOut_];
assert(fabs(dualOut_) < 1.0e-5);
bool roundAgain = true;
int returnCode = 0;
bool updateSolution = true;
while (roundAgain) {
roundAgain = false;
unpack(rowArray_[1]);
factorization_->updateColumnFT(rowArray_[2], rowArray_[1]);
alpha_ = 0.0;
int i;
int * index = rowArray_[1]->getIndices();
int number = rowArray_[1]->getNumElements();
double * element = rowArray_[1]->denseVector();
assert ( !rowArray_[3]->getNumElements());
double * saveSolution = rowArray_[3]->denseVector();
for (i = 0; i < number; i++) {
int ii = index[i];
if ( pivotVariable_[ii] == sequenceOut_) {
pivotRow_ = ii;
alpha_ = element[pivotRow_];
break;
}
}
if (fabs(alpha_) < 1.0e-8) {
// be on safe side and clear arrays
rowArray_[0]->clear();
rowArray_[1]->clear();
return -1; // will be singular
}
// we are going to subtract movement from current basic
double movement;
// see where incoming will go to
if (sequenceOut_ < 0 || sequenceIn_ == sequenceOut_) {
// flip so go to bound
movement = ((directionIn_ > 0) ? upperIn_ : lowerIn_) - valueIn_;
} else {
// get where outgoing needs to get to
double outValue = (directionOut_ < 0) ? upperOut_ : lowerOut_;
// solutionOut_ - movement*alpha_ == outValue
movement = (valueOut_ - outValue) / alpha_;
// set directionIn_ correctly
directionIn_ = (movement > 0) ? 1 : -1;
}
theta_ = movement;
double oldValueIn = valueIn_;
// update primal solution
for (i = 0; i < number; i++) {
int ii = index[i];
// get column
int ij = pivotVariable_[ii];
double value = element[ii];
saveSolution[ii] = solution_[ij];
solution_[ij] -= movement * value;
}
//rowArray_[1]->setNumElements(0);
// see where something went to
#ifndef NDEBUG
CoinRelFltEq eq(1.0e-7);
#endif
if (sequenceOut_ < 0) {
if (directionIn_ < 0) {
assert (eq(solution_[sequenceIn_], upperIn_));
solution_[sequenceIn_] = upperIn_;
} else {
assert (eq(solution_[sequenceIn_], lowerIn_));
solution_[sequenceIn_] = lowerIn_;
}
} else {
if (directionOut_ < 0) {
assert (eq(solution_[sequenceOut_], upperOut_));
solution_[sequenceOut_] = upperOut_;
} else {
assert (eq(solution_[sequenceOut_], lowerOut_));
solution_[sequenceOut_] = lowerOut_;
}
valueOut_ = solution_[sequenceOut_];
solution_[sequenceIn_] = valueIn_ + movement;
}
valueIn_ = solution_[sequenceIn_];
double objectiveChange = dualIn_ * movement;
// update duals
if (pivotRow_ >= 0) {
if (fabs(alpha_) < 1.0e-8) {
// be on safe side and clear arrays
rowArray_[0]->clear();
rowArray_[1]->clear();
return -1; // will be singular
}
double multiplier = dualIn_ / alpha_;
rowArray_[0]->insert(pivotRow_, multiplier);
factorization_->updateColumnTranspose(rowArray_[2], rowArray_[0]);
// put row of tableau in rowArray[0] and columnArray[0]
matrix_->transposeTimes(this, -1.0,
rowArray_[0], columnArray_[1], columnArray_[0]);
// update column djs
int i;
int * index = columnArray_[0]->getIndices();
int number = columnArray_[0]->getNumElements();
double * element = columnArray_[0]->denseVector();
for (i = 0; i < number; i++) {
int ii = index[i];
dj_[ii] += element[ii];
reducedCost_[ii] = dj_[ii];
element[ii] = 0.0;
}
columnArray_[0]->setNumElements(0);
// and row djs
index = rowArray_[0]->getIndices();
number = rowArray_[0]->getNumElements();
element = rowArray_[0]->denseVector();
for (i = 0; i < number; i++) {
int ii = index[i];
dj_[ii+numberColumns_] += element[ii];
dual_[ii] = dj_[ii+numberColumns_];
element[ii] = 0.0;
}
rowArray_[0]->setNumElements(0);
// check incoming
assert (fabs(dj_[sequenceIn_]) < 1.0e-6 || CoinAbs(solveType_) == 2);
}
// if stable replace in basis
int updateStatus = factorization_->replaceColumn(this,
rowArray_[2],
rowArray_[1],
pivotRow_,
alpha_);
bool takePivot = true;
// See if Factorization updated
if (updateStatus) {
updateSolution = false;
returnCode = 1;
}
// if no pivots, bad update but reasonable alpha - take and invert
if (updateStatus == 2 &&
lastGoodIteration_ == numberIterations_ && fabs(alpha_) > 1.0e-5)
updateStatus = 4;
if (updateStatus == 1 || updateStatus == 4 || fabs(alpha_) < 1.0e-6) {
// slight error
if (factorization_->pivots() > 5 || updateStatus == 4) {
returnCode = 1;
}
} else if (updateStatus == 2) {
// major error - put back solution
valueIn_ = oldValueIn;
solution_[sequenceIn_] = valueIn_;
int * index = rowArray_[1]->getIndices();
int number = rowArray_[1]->getNumElements();
for (i = 0; i < number; i++) {
int ii = index[i];
// get column
int ij = pivotVariable_[ii];
solution_[ij] = saveSolution[ii];
}
if (sequenceOut_ >= 0)
valueOut_ = solution_[sequenceOut_];
takePivot = false;
if (factorization_->pivots()) {
// refactorize here
int factorStatus = internalFactorize(1);
if (factorStatus) {
printf("help in user pivot\n");
abort();
}
gutsOfSolution(NULL, NULL);
valueIn_ = solution_[sequenceIn_];
if (sequenceOut_ >= 0)
valueOut_ = solution_[sequenceOut_];
roundAgain = true;
} else {
returnCode = -1;
}
} else if (updateStatus == 3) {
// out of memory
// increase space if not many iterations
if (factorization_->pivots() <
0.5 * factorization_->maximumPivots() &&
factorization_->pivots() < 200)
factorization_->areaFactor(
factorization_->areaFactor() * 1.1);
returnCode = 1; // factorize now
}
{
// clear saveSolution
int * index = rowArray_[1]->getIndices();
int number = rowArray_[1]->getNumElements();
for (i = 0; i < number; i++) {
int ii = index[i];
saveSolution[ii] = 0.0;
}
}
rowArray_[1]->clear();
if (takePivot) {
int save = algorithm_;
// make simple so always primal
algorithm_ = 1;
housekeeping(objectiveChange);
algorithm_ = save;
}
}
if (returnCode == 1) {
// refactorize here
int factorStatus = internalFactorize(1);
if (factorStatus) {
printf("help in user pivot\n");
abort();
}
updateSolution = true;
}
if (updateSolution) {
// just for now - recompute anyway
gutsOfSolution(NULL, NULL);
}
return returnCode;
}
/* Pivot in a variable and choose an outgoing one. Assumes primal
feasible - will not go through a bound. Returns step length in theta
Returns ray in ray_ (or NULL if no pivot)
Return codes as before but -1 means no acceptable pivot
*/
int ClpSimplex::primalPivotResult()
{
assert (sequenceIn_ >= 0);
valueIn_ = solution_[sequenceIn_];
lowerIn_ = lower_[sequenceIn_];
upperIn_ = upper_[sequenceIn_];
dualIn_ = dj_[sequenceIn_];
if (!nonLinearCost_)
nonLinearCost_ = new ClpNonLinearCost(this);
int returnCode = static_cast<ClpSimplexPrimal *> (this)->pivotResult();
if (returnCode < 0 && returnCode > -4) {
return 0;
} else {
COIN_DETAIL_PRINT(printf("Return code of %d from ClpSimplexPrimal::pivotResult\n",
returnCode));
return -1;
}
}
// Factorization frequency
int
ClpSimplex::factorizationFrequency() const
{
if (factorization_)
return factorization_->maximumPivots();
else
return -1;
}
void
ClpSimplex::setFactorizationFrequency(int value)
{
if (factorization_)
factorization_->maximumPivots(value);
}
// Common bits of coding for dual and primal
int
ClpSimplex::startup(int ifValuesPass, int startFinishOptions)
{
// Get rid of some arrays and empty factorization
int useFactorization = false;
if ((startFinishOptions & 2) != 0 && (whatsChanged_&(2 + 512)) == 2 + 512)
useFactorization = true; // Keep factorization if possible
#if 0
// seems to be needed if rows deleted later in CbcModel!
if (!solution_ && scaledMatrix_) {
// get rid of scaled matrix
if (scaledMatrix_->getNumRows() != numberRows_) {
delete scaledMatrix_;
scaledMatrix_ = NULL;
}
}
#endif
// sanity check
// bad if empty (trap here to avoid using bad matrix_)
#if 0
// but also check bounds
{
int badProblem = 0;
int i;
for (i = 0; i < numberColumns_; i++) {
if (columnLower_[i] > columnUpper_[i])
badProblem++;
}
for (i = 0; i < numberRows_; i++) {
if (rowLower_[i] > rowUpper_[i])
badProblem++;
}
if (badProblem) {
numberDualInfeasibilities_ = 0;
sumDualInfeasibilities_ = 0.0;
numberPrimalInfeasibilities_ = badProblem;
sumPrimalInfeasibilities_ = badProblem;
secondaryStatus_ = 6; // so user can see something odd
problemStatus_ = 1;
bool printIt = (specialOptions_ & 32768) == 0 ? true : false; // no message if from Osi
if (printIt)
handler_->message(CLP_INFEASIBLE, messages_)
<< CoinMessageEol;
return 2;
}
}
#endif
if (!matrix_ || (!matrix_->getNumElements() && objective_->type() < 2)) {
int infeasNumber[2];
double infeasSum[2];
bool printIt = (specialOptions_ & 32768) == 0 ? true : false; // no message if from Osi
problemStatus_ = emptyProblem(infeasNumber, infeasSum, printIt);
if ((startFinishOptions & 1) != 0) {
// User may expect user data - fill in as required
if (numberRows_) {
if (!pivotVariable_)
pivotVariable_ = new int [numberRows_];
CoinIotaN(pivotVariable_, numberRows_, numberColumns_);
}
}
numberDualInfeasibilities_ = infeasNumber[0];
sumDualInfeasibilities_ = infeasSum[0];
numberPrimalInfeasibilities_ = infeasNumber[1];
sumPrimalInfeasibilities_ = infeasSum[1];
return 2;
}
pivotRow_ = -1;
sequenceIn_ = -1;
sequenceOut_ = -1;
secondaryStatus_ = 0;
primalTolerance_ = dblParam_[ClpPrimalTolerance];
dualTolerance_ = dblParam_[ClpDualTolerance];
if (problemStatus_ != 10)
numberIterations_ = 0;
// put in standard form (and make row copy)
// create modifiable copies of model rim and do optional scaling
bool goodMatrix = createRim(7 + 8 + 16 + 32, true, startFinishOptions);
if (goodMatrix) {
// switch off factorization if bad
if (pivotVariable_[0] < 0)
useFactorization = false;
// Model looks okay
// Do initial factorization
// and set certain stuff
// We can either set increasing rows so ...IsBasic gives pivot row
// or we can just increment iBasic one by one
// for now let ...iBasic give pivot row
int saveThreshold = factorization_->denseThreshold();
if (!useFactorization || factorization_->numberRows() != numberRows_) {
useFactorization = false;
factorization_->setDefaultValues();
// Switch off dense (unless special option set)
if ((specialOptions_ & 8) == 0)
factorization_->setDenseThreshold(0);
}
// If values pass then perturb (otherwise may be optimal so leave a bit)
if (ifValuesPass) {
// do perturbation if asked for
if (perturbation_ < 100) {
if (algorithm_ > 0 && (objective_->type() < 2 || !objective_->activated())) {
#ifndef FEB_TRY
//static_cast<ClpSimplexPrimal *> (this)->perturb(0);
#endif
} else if (algorithm_ < 0) {
static_cast<ClpSimplexDual *> (this)->perturb();
}
}
}
// for primal we will change bounds using infeasibilityCost_
if (nonLinearCost_ == NULL && algorithm_ > 0) {
// get a valid nonlinear cost function
nonLinearCost_ = new ClpNonLinearCost(this);
}
// loop round to clean up solution if values pass
int numberThrownOut = -1;
int totalNumberThrownOut = 0;
problemStatus_ = -1;
// see if we are re-using factorization
if (!useFactorization) {
while(numberThrownOut) {
int status = internalFactorize(ifValuesPass ? 10 : 0);
if (status < 0)
return 1; // some error
else
numberThrownOut = status;
// for this we need clean basis so it is after factorize
if (!numberThrownOut || numberThrownOut == numberRows_ + 1) {
// solution will be done again - skip if absolutely sure
if ((specialOptions_ & 512) == 0 || numberThrownOut == numberRows_ + 1) {
//int saveFirstFree=firstFree_;
numberThrownOut = gutsOfSolution( NULL, NULL,
ifValuesPass != 0);
//firstFree_=saveFirstFree;
if (largestPrimalError_ > 10.0 && !ifValuesPass && !numberThrownOut) {
// throw out up to 1000 structurals
int iRow;
int * sort = new int[numberRows_];
double * array = rowArray_[0]->denseVector();
memset(array, 0, numberRows_ * sizeof(double));
times(-1.0, columnActivityWork_, array);
int numberBasic = 0;
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
if (iPivot < numberColumns_) {
// column
double difference = fabs(array[iRow] + rowActivityWork_[iRow]);
if (difference > 1.0e-4) {
sort[numberThrownOut] = iPivot;
array[numberThrownOut++] = difference;
if (getStatus(iPivot) == basic)
numberBasic++;
}
}
}
if (!numberBasic) {
allSlackBasis(true);
numberThrownOut = 1; // force another go
} else {
CoinSort_2(array, array + numberThrownOut, sort);
numberThrownOut = CoinMin(1000, numberThrownOut);
for (iRow = 0; iRow < numberThrownOut; iRow++) {
int iColumn = sort[iRow];
setColumnStatus(iColumn, superBasic);
if (fabs(solution_[iColumn]) > 1.0e10) {
if (upper_[iColumn] < 0.0) {
solution_[iColumn] = upper_[iColumn];
} else if (lower_[iColumn] > 0.0) {
solution_[iColumn] = lower_[iColumn];
} else {
solution_[iColumn] = 0.0;
}
}
}
}
CoinZeroN(array, numberRows_);
delete [] sort;
}
} else {
// make sure not optimal at once
numberPrimalInfeasibilities_ = 1;
numberThrownOut = 0;
}
} else {
matrix_->rhsOffset(this, true); // redo rhs offset
}
totalNumberThrownOut += numberThrownOut;
}
} else {
// using previous factorization - we assume fine
if ((moreSpecialOptions_ & 8) == 0) {
// but we need to say not optimal
numberPrimalInfeasibilities_ = 1;
numberDualInfeasibilities_ = 1;
}
matrix_->rhsOffset(this, true); // redo rhs offset
}
if (totalNumberThrownOut)
handler_->message(CLP_SINGULARITIES, messages_)
<< totalNumberThrownOut
<< CoinMessageEol;
// Switch back dense
factorization_->setDenseThreshold(saveThreshold);
if (!numberPrimalInfeasibilities_ && !numberDualInfeasibilities_
&& !ifValuesPass &&
(!nonLinearCost_ || !nonLinearCost_->numberInfeasibilities()))
problemStatus_ = 0;
else
assert(problemStatus_ == -1);
// number of times we have declared optimality
numberTimesOptimal_ = 0;
if (disasterArea_)
disasterArea_->intoSimplex();
return 0;
} else {
// bad matrix
return 2;
}
}
/* Get a clean factorization - i.e. throw out singularities
may do more later */
int
ClpSimplex::cleanFactorization(int ifValuesPass)
{
int status = internalFactorize(ifValuesPass ? 10 : 0);
if (status < 0) {
return 1; // some error
} else {
firstFree_=0;
return 0;
}
}
void
ClpSimplex::finish(int startFinishOptions)
{
// Get rid of some arrays and empty factorization
int getRidOfData = 1;
if (upper_ && ((startFinishOptions & 1) != 0 || problemStatus_ == 10)) {
getRidOfData = 0; // Keep stuff
// mark all as current
whatsChanged_ = 0x3ffffff;
} else {
whatsChanged_ &= ~0xffff;
}
double saveObjValue = objectiveValue_;
deleteRim(getRidOfData);
if (matrix_->type()>=15)
objectiveValue_ = saveObjValue;
// Skip message if changing algorithms
if (problemStatus_ != 10) {
if (problemStatus_ == -1)
problemStatus_ = 4;
assert(problemStatus_ >= 0 && problemStatus_ < 6);
if (handler_->detail(CLP_SIMPLEX_FINISHED, messages_) < 100) {
handler_->message(CLP_SIMPLEX_FINISHED + problemStatus_, messages_)
<< objectiveValue()
<< CoinMessageEol;
}
}
factorization_->relaxAccuracyCheck(1.0);
// get rid of any network stuff - could do more
factorization_->cleanUp();
}
// Save data
ClpDataSave
ClpSimplex::saveData()
{
ClpDataSave saved;
saved.dualBound_ = dualBound_;
saved.infeasibilityCost_ = infeasibilityCost_;
saved.sparseThreshold_ = factorization_->sparseThreshold();
saved.pivotTolerance_ = factorization_->pivotTolerance();
saved.zeroFactorizationTolerance_ = factorization_->zeroTolerance();
saved.zeroSimplexTolerance_ = zeroTolerance_;
saved.perturbation_ = perturbation_;
saved.forceFactorization_ = forceFactorization_;
saved.acceptablePivot_ = acceptablePivot_;
saved.objectiveScale_ = objectiveScale_;
// Progress indicator
progress_.fillFromModel (this);
return saved;
}
// Restore data
void
ClpSimplex::restoreData(ClpDataSave saved)
{
//factorization_->sparseThreshold(saved.sparseThreshold_);
factorization_->pivotTolerance(saved.pivotTolerance_);
factorization_->zeroTolerance(saved.zeroFactorizationTolerance_);
zeroTolerance_ = saved.zeroSimplexTolerance_;
perturbation_ = saved.perturbation_;
infeasibilityCost_ = saved.infeasibilityCost_;
dualBound_ = saved.dualBound_;
forceFactorization_ = saved.forceFactorization_;
objectiveScale_ = saved.objectiveScale_;
acceptablePivot_ = saved.acceptablePivot_;
}
// To flag a variable (not inline to allow for column generation)
void
ClpSimplex::setFlagged( int sequence)
{
status_[sequence] |= 64;
matrix_->generalExpanded(this, 7, sequence);
lastFlaggedIteration_ = numberIterations_;
}
/* Factorizes and returns true if optimal. Used by user */
bool
ClpSimplex::statusOfProblem(bool initial)
{
// We don't want scaling
int saveFlag = scalingFlag_;
if (!rowScale_)
scalingFlag_ = 0;
bool goodMatrix = createRim(7 + 8 + 16 + 32);
if (!goodMatrix) {
problemStatus_ = 4;
scalingFlag_ = saveFlag;
return false;
}
// is factorization okay?
if (initial) {
// First time - allow singularities
int numberThrownOut = -1;
int totalNumberThrownOut = 0;
while(numberThrownOut) {
int status = internalFactorize(0);
if (status == numberRows_ + 1)
status = 0; // all slack
if (status < 0) {
deleteRim(-1);
scalingFlag_ = saveFlag;
return false; // some error
} else {
numberThrownOut = status;
}
// for this we need clean basis so it is after factorize
//if (!numberThrownOut)
//numberThrownOut = gutsOfSolution( NULL,NULL,
// false);
//else
//matrix_->rhsOffset(this,true); // redo rhs offset
totalNumberThrownOut += numberThrownOut;
}
if (totalNumberThrownOut)
handler_->message(CLP_SINGULARITIES, messages_)
<< totalNumberThrownOut
<< CoinMessageEol;
} else {
#ifndef NDEBUG
int returnCode = internalFactorize(1);
assert (!returnCode);
#else
internalFactorize(1);
#endif
}
CoinMemcpyN(rowActivity_, numberRows_, rowActivityWork_);
CoinMemcpyN(columnActivity_, numberColumns_, columnActivityWork_);
gutsOfSolution(NULL, NULL);
CoinMemcpyN(rowActivityWork_, numberRows_, rowActivity_);
CoinMemcpyN(columnActivityWork_, numberColumns_, columnActivity_);
CoinMemcpyN(dj_, numberColumns_, reducedCost_);
deleteRim(-1);
scalingFlag_ = saveFlag;
return (primalFeasible() && dualFeasible());
}
/* Return model - updates any scalars */
void
ClpSimplex::returnModel(ClpSimplex & otherModel)
{
ClpModel::returnModel(otherModel);
otherModel.bestPossibleImprovement_ = bestPossibleImprovement_;
otherModel.columnPrimalSequence_ = columnPrimalSequence_;
otherModel.zeroTolerance_ = zeroTolerance_;
otherModel.rowPrimalSequence_ = rowPrimalSequence_;
otherModel.bestObjectiveValue_ = bestObjectiveValue_;
otherModel.moreSpecialOptions_ = moreSpecialOptions_;
otherModel.baseIteration_ = baseIteration_;
otherModel.primalToleranceToGetOptimal_ = primalToleranceToGetOptimal_;
otherModel.largestPrimalError_ = largestPrimalError_;
otherModel.largestDualError_ = largestDualError_;
otherModel.alphaAccuracy_ = alphaAccuracy_;
otherModel.alpha_ = alpha_;
otherModel.theta_ = theta_;
otherModel.lowerIn_ = lowerIn_;
otherModel.valueIn_ = valueIn_;
otherModel.upperIn_ = upperIn_;
otherModel.dualIn_ = dualIn_;
otherModel.sequenceIn_ = sequenceIn_;
otherModel.directionIn_ = directionIn_;
otherModel.lowerOut_ = lowerOut_;
otherModel.valueOut_ = valueOut_;
otherModel.upperOut_ = upperOut_;
otherModel.dualOut_ = dualOut_;
otherModel.sequenceOut_ = sequenceOut_;
otherModel.directionOut_ = directionOut_;
otherModel.pivotRow_ = pivotRow_;
otherModel.algorithm_ = algorithm_;
otherModel.sumDualInfeasibilities_ = sumDualInfeasibilities_;
otherModel.numberDualInfeasibilities_ = numberDualInfeasibilities_;
otherModel.numberDualInfeasibilitiesWithoutFree_ =
numberDualInfeasibilitiesWithoutFree_;
otherModel.sumPrimalInfeasibilities_ = sumPrimalInfeasibilities_;
otherModel.numberPrimalInfeasibilities_ = numberPrimalInfeasibilities_;
otherModel.numberTimesOptimal_ = numberTimesOptimal_;
otherModel.disasterArea_ = NULL;
otherModel.sumOfRelaxedDualInfeasibilities_ = sumOfRelaxedDualInfeasibilities_;
otherModel.sumOfRelaxedPrimalInfeasibilities_ = sumOfRelaxedPrimalInfeasibilities_;
if (perturbationArray_ != otherModel.perturbationArray_)
delete [] perturbationArray_;
perturbationArray_ = NULL;
}
/* Constructs a non linear cost from list of non-linearities (columns only)
First lower of each column is taken as real lower
Last lower is taken as real upper and cost ignored
Returns nonzero if bad data e.g. lowers not monotonic
*/
int
ClpSimplex::createPiecewiseLinearCosts(const int * starts,
const double * lower, const double * gradient)
{
delete nonLinearCost_;
// Set up feasible bounds and check monotonicity
int iColumn;
int returnCode = 0;
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
int iIndex = starts[iColumn];
int end = starts[iColumn+1] - 1;
columnLower_[iColumn] = lower[iIndex];
columnUpper_[iColumn] = lower[end];
double value = columnLower_[iColumn];
iIndex++;
for (; iIndex < end; iIndex++) {
if (lower[iIndex] < value)
returnCode++; // not monotonic
value = lower[iIndex];
}
}
nonLinearCost_ = new ClpNonLinearCost(this, starts, lower, gradient);
specialOptions_ |= 2; // say keep
return returnCode;
}
/* For advanced use. When doing iterative solves things can get
nasty so on values pass if incoming solution has largest
infeasibility < incomingInfeasibility throw out variables
from basis until largest infeasibility < allowedInfeasibility
or incoming largest infeasibility.
If allowedInfeasibility>= incomingInfeasibility this is
always possible altough you may end up with an all slack basis.
Defaults are 1.0,10.0
*/
void
ClpSimplex::setValuesPassAction(double incomingInfeasibility,
double allowedInfeasibility)
{
incomingInfeasibility_ = incomingInfeasibility;
allowedInfeasibility_ = allowedInfeasibility;
CoinAssert(incomingInfeasibility_ >= 0.0);
CoinAssert(allowedInfeasibility_ >= incomingInfeasibility_);
}
// Allow initial dense factorization
void
ClpSimplex::setInitialDenseFactorization(bool onOff)
{
if (onOff)
specialOptions_ |= 8;
else
specialOptions_ &= ~8;
}
bool
ClpSimplex::initialDenseFactorization() const
{
return (specialOptions_ & 8) != 0;
}
/* This constructor modifies original ClpSimplex and stores
original stuff in created ClpSimplex. It is only to be used in
conjunction with originalModel */
ClpSimplex::ClpSimplex (ClpSimplex * wholeModel,
int numberColumns, const int * whichColumns)
{
// Set up dummy row selection list
numberRows_ = wholeModel->numberRows_;
int * whichRow = new int [numberRows_];
int iRow;
for (iRow = 0; iRow < numberRows_; iRow++)
whichRow[iRow] = iRow;
// ClpModel stuff (apart from numberColumns_)
matrix_ = wholeModel->matrix_;
rowCopy_ = wholeModel->rowCopy_;
if (wholeModel->rowCopy_) {
// note reversal of order
wholeModel->rowCopy_ = wholeModel->rowCopy_->subsetClone(numberRows_, whichRow,
numberColumns, whichColumns);
} else {
wholeModel->rowCopy_ = NULL;
}
whatsChanged_ &= ~0xffff;
CoinAssert (wholeModel->matrix_);
wholeModel->matrix_ = wholeModel->matrix_->subsetClone(numberRows_, whichRow,
numberColumns, whichColumns);
delete [] whichRow;
numberColumns_ = wholeModel->numberColumns_;
// Now ClpSimplex stuff and status_
delete wholeModel->primalColumnPivot_;
wholeModel->primalColumnPivot_ = new ClpPrimalColumnSteepest(0);
nonLinearCost_ = wholeModel->nonLinearCost_;
// Now main arrays
int iColumn;
int numberTotal = numberRows_ + numberColumns;
COIN_DETAIL_PRINT(printf("%d %d %d\n", numberTotal, numberRows_, numberColumns));
// mapping
int * mapping = new int[numberRows_+numberColumns_];
for (iColumn = 0; iColumn < numberColumns_; iColumn++)
mapping[iColumn] = -1;
for (iRow = 0; iRow < numberRows_; iRow++)
mapping[iRow+numberColumns_] = iRow + numberColumns;
// Redo costs and bounds of whole model
wholeModel->createRim(1 + 4, false);
lower_ = wholeModel->lower_;
wholeModel->lower_ = new double [numberTotal];
CoinMemcpyN(lower_ + numberColumns_, numberRows_, wholeModel->lower_ + numberColumns);
for (iColumn = 0; iColumn < numberColumns; iColumn++) {
int jColumn = whichColumns[iColumn];
wholeModel->lower_[iColumn] = lower_[jColumn];
// and pointer back
mapping[jColumn] = iColumn;
}
#ifdef CLP_DEBUG
for (iColumn = 0; iColumn < numberRows_ + numberColumns_; iColumn++)
printf("mapx %d %d\n", iColumn, mapping[iColumn]);
#endif
// Re-define pivotVariable_
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = wholeModel->pivotVariable_[iRow];
wholeModel->pivotVariable_[iRow] = mapping[iPivot];
#ifdef CLP_DEBUG
printf("p row %d, pivot %d -> %d\n", iRow, iPivot, mapping[iPivot]);
#endif
assert (wholeModel->pivotVariable_[iRow] >= 0);
}
// Reverse mapping (so extended version of whichColumns)
for (iColumn = 0; iColumn < numberColumns; iColumn++)
mapping[iColumn] = whichColumns[iColumn];
for (; iColumn < numberRows_ + numberColumns; iColumn++)
mapping[iColumn] = iColumn + (numberColumns_ - numberColumns);
#ifdef CLP_DEBUG
for (iColumn = 0; iColumn < numberRows_ + numberColumns; iColumn++)
printf("map %d %d\n", iColumn, mapping[iColumn]);
#endif
// Save mapping somewhere - doesn't matter
rowUpper_ = reinterpret_cast<double *> (mapping);
upper_ = wholeModel->upper_;
wholeModel->upper_ = new double [numberTotal];
for (iColumn = 0; iColumn < numberTotal; iColumn++) {
int jColumn = mapping[iColumn];
wholeModel->upper_[iColumn] = upper_[jColumn];
}
cost_ = wholeModel->cost_;
wholeModel->cost_ = new double [numberTotal];
for (iColumn = 0; iColumn < numberTotal; iColumn++) {
int jColumn = mapping[iColumn];
wholeModel->cost_[iColumn] = cost_[jColumn];
}
dj_ = wholeModel->dj_;
wholeModel->dj_ = new double [numberTotal];
for (iColumn = 0; iColumn < numberTotal; iColumn++) {
int jColumn = mapping[iColumn];
wholeModel->dj_[iColumn] = dj_[jColumn];
}
solution_ = wholeModel->solution_;
wholeModel->solution_ = new double [numberTotal];
for (iColumn = 0; iColumn < numberTotal; iColumn++) {
int jColumn = mapping[iColumn];
wholeModel->solution_[iColumn] = solution_[jColumn];
}
// now see what variables left out do to row solution
double * rowSolution = wholeModel->solution_ + numberColumns;
double * fullSolution = solution_;
double * sumFixed = new double[numberRows_];
memset (sumFixed, 0, numberRows_ * sizeof(double));
// zero out ones in small problem
for (iColumn = 0; iColumn < numberColumns; iColumn++) {
int jColumn = mapping[iColumn];
fullSolution[jColumn] = 0.0;
}
// Get objective offset
double originalOffset;
wholeModel->getDblParam(ClpObjOffset, originalOffset);
double offset = 0.0;
const double * cost = cost_;
for (iColumn = 0; iColumn < numberColumns_; iColumn++)
offset += fullSolution[iColumn] * cost[iColumn];
wholeModel->setDblParam(ClpObjOffset, originalOffset - offset);
setDblParam(ClpObjOffset, originalOffset);
matrix_->times(1.0, fullSolution, sumFixed, wholeModel->rowScale_, wholeModel->columnScale_);
double * lower = lower_ + numberColumns;
double * upper = upper_ + numberColumns;
double fixed = 0.0;
for (iRow = 0; iRow < numberRows_; iRow++) {
fixed += fabs(sumFixed[iRow]);
if (lower[iRow] > -1.0e50)
lower[iRow] -= sumFixed[iRow];
if (upper[iRow] < 1.0e50)
upper[iRow] -= sumFixed[iRow];
rowSolution[iRow] -= sumFixed[iRow];
}
COIN_DETAIL_PRINT(printf("offset %g sumfixed %g\n", offset, fixed));
delete [] sumFixed;
columnScale_ = wholeModel->columnScale_;
if (columnScale_) {
wholeModel->columnScale_ = new double [numberTotal];
for (iColumn = 0; iColumn < numberColumns; iColumn++) {
int jColumn = mapping[iColumn];
wholeModel->columnScale_[iColumn] = columnScale_[jColumn];
}
}
status_ = wholeModel->status_;
wholeModel->status_ = new unsigned char [numberTotal];
for (iColumn = 0; iColumn < numberTotal; iColumn++) {
int jColumn = mapping[iColumn];
wholeModel->status_[iColumn] = status_[jColumn];
}
savedSolution_ = wholeModel->savedSolution_;
if (savedSolution_) {
wholeModel->savedSolution_ = new double [numberTotal];
for (iColumn = 0; iColumn < numberTotal; iColumn++) {
int jColumn = mapping[iColumn];
wholeModel->savedSolution_[iColumn] = savedSolution_[jColumn];
}
}
saveStatus_ = wholeModel->saveStatus_;
if (saveStatus_) {
wholeModel->saveStatus_ = new unsigned char [numberTotal];
for (iColumn = 0; iColumn < numberTotal; iColumn++) {
int jColumn = mapping[iColumn];
wholeModel->saveStatus_[iColumn] = saveStatus_[jColumn];
}
}
wholeModel->numberColumns_ = numberColumns;
// Initialize weights
wholeModel->primalColumnPivot_->saveWeights(wholeModel, 2);
// Costs
wholeModel->nonLinearCost_ = new ClpNonLinearCost(wholeModel);
wholeModel->nonLinearCost_->checkInfeasibilities();
COIN_DETAIL_PRINT(printf("after contraction %d infeasibilities summing to %g\n",
nonLinearCost_->numberInfeasibilities(), nonLinearCost_->sumInfeasibilities()));
// Redo some stuff
wholeModel->reducedCostWork_ = wholeModel->dj_;
wholeModel->rowReducedCost_ = wholeModel->dj_ + wholeModel->numberColumns_;
wholeModel->columnActivityWork_ = wholeModel->solution_;
wholeModel->rowActivityWork_ = wholeModel->solution_ + wholeModel->numberColumns_;
wholeModel->objectiveWork_ = wholeModel->cost_;
wholeModel->rowObjectiveWork_ = wholeModel->cost_ + wholeModel->numberColumns_;
wholeModel->rowLowerWork_ = wholeModel->lower_ + wholeModel->numberColumns_;
wholeModel->columnLowerWork_ = wholeModel->lower_;
wholeModel->rowUpperWork_ = wholeModel->upper_ + wholeModel->numberColumns_;
wholeModel->columnUpperWork_ = wholeModel->upper_;
#ifndef NDEBUG
// Check status
ClpSimplex * xxxx = wholeModel;
int nBasic = 0;
for (iColumn = 0; iColumn < xxxx->numberRows_ + xxxx->numberColumns_; iColumn++)
if (xxxx->getStatus(iColumn) == basic)
nBasic++;
assert (nBasic == xxxx->numberRows_);
for (iRow = 0; iRow < xxxx->numberRows_; iRow++) {
int iPivot = xxxx->pivotVariable_[iRow];
assert (xxxx->getStatus(iPivot) == basic);
}
#endif
}
/* This copies back stuff from miniModel and then deletes miniModel.
Only to be used with mini constructor */
void
ClpSimplex::originalModel(ClpSimplex * miniModel)
{
int numberSmall = numberColumns_;
numberColumns_ = miniModel->numberColumns_;
int numberTotal = numberSmall + numberRows_;
// copy back
int iColumn;
int * mapping = reinterpret_cast<int *> (miniModel->rowUpper_);
#ifdef CLP_DEBUG
for (iColumn = 0; iColumn < numberRows_ + numberColumns_; iColumn++)
printf("mapb %d %d\n", iColumn, mapping[iColumn]);
#endif
// miniModel actually has full arrays
// now see what variables left out do to row solution
double * fullSolution = miniModel->solution_;
double * sumFixed = new double[numberRows_];
memset (sumFixed, 0, numberRows_ * sizeof(double));
miniModel->matrix_->times(1.0, fullSolution, sumFixed, rowScale_, miniModel->columnScale_);
for (iColumn = 0; iColumn < numberTotal; iColumn++) {
int jColumn = mapping[iColumn];
miniModel->lower_[jColumn] = lower_[iColumn];
miniModel->upper_[jColumn] = upper_[iColumn];
miniModel->cost_[jColumn] = cost_[iColumn];
miniModel->dj_[jColumn] = dj_[iColumn];
miniModel->solution_[jColumn] = solution_[iColumn];
miniModel->status_[jColumn] = status_[iColumn];
#ifdef CLP_DEBUG
printf("%d in small -> %d in original\n", iColumn, jColumn);
#endif
}
delete [] lower_;
lower_ = miniModel->lower_;
delete [] upper_;
upper_ = miniModel->upper_;
delete [] cost_;
cost_ = miniModel->cost_;
delete [] dj_;
dj_ = miniModel->dj_;
delete [] solution_;
solution_ = miniModel->solution_;
delete [] status_;
status_ = miniModel->status_;
if (columnScale_) {
for (iColumn = 0; iColumn < numberSmall; iColumn++) {
int jColumn = mapping[iColumn];
miniModel->columnScale_[jColumn] = columnScale_[iColumn];
}
delete [] columnScale_;
columnScale_ = miniModel->columnScale_;
}
if (savedSolution_) {
if (!miniModel->savedSolution_) {
miniModel->savedSolution_ = ClpCopyOfArray(solution_, numberColumns_ + numberRows_);
} else {
for (iColumn = 0; iColumn < numberTotal; iColumn++) {
int jColumn = mapping[iColumn];
miniModel->savedSolution_[jColumn] = savedSolution_[iColumn];
}
}
delete [] savedSolution_;
savedSolution_ = miniModel->savedSolution_;
}
if (saveStatus_) {
if (!miniModel->saveStatus_) {
miniModel->saveStatus_ = ClpCopyOfArray(status_, numberColumns_ + numberRows_);
} else {
for (iColumn = 0; iColumn < numberTotal; iColumn++) {
int jColumn = mapping[iColumn];
miniModel->saveStatus_[jColumn] = saveStatus_[iColumn];
}
}
delete [] saveStatus_;
saveStatus_ = miniModel->saveStatus_;
}
// Re-define pivotVariable_
int iRow;
for (iRow = 0; iRow < numberRows_; iRow++) {
int iPivot = pivotVariable_[iRow];
#ifdef CLP_DEBUG
printf("pb row %d, pivot %d -> %d\n", iRow, iPivot, mapping[iPivot]);
#endif
pivotVariable_[iRow] = mapping[iPivot];
assert (pivotVariable_[iRow] >= 0);
}
// delete stuff and move back
delete matrix_;
delete rowCopy_;
delete primalColumnPivot_;
delete nonLinearCost_;
matrix_ = miniModel->matrix_;
rowCopy_ = miniModel->rowCopy_;
nonLinearCost_ = miniModel->nonLinearCost_;
double originalOffset;
miniModel->getDblParam(ClpObjOffset, originalOffset);
setDblParam(ClpObjOffset, originalOffset);
// Redo some stuff
reducedCostWork_ = dj_;
rowReducedCost_ = dj_ + numberColumns_;
columnActivityWork_ = solution_;
rowActivityWork_ = solution_ + numberColumns_;
objectiveWork_ = cost_;
rowObjectiveWork_ = cost_ + numberColumns_;
rowLowerWork_ = lower_ + numberColumns_;
columnLowerWork_ = lower_;
rowUpperWork_ = upper_ + numberColumns_;
columnUpperWork_ = upper_;
// Cleanup
for (iRow = 0; iRow < numberRows_; iRow++) {
double value = rowActivityWork_[iRow] + sumFixed[iRow];
rowActivityWork_[iRow] = value;
switch(getRowStatus(iRow)) {
case basic:
break;
case atUpperBound:
//rowActivityWork_[iRow]=rowUpperWork_[iRow];
break;
case ClpSimplex::isFixed:
case atLowerBound:
//rowActivityWork_[iRow]=rowLowerWork_[iRow];
break;
case isFree:
break;
// superbasic
case superBasic:
break;
}
}
delete [] sumFixed;
nonLinearCost_->checkInfeasibilities();
COIN_DETAIL_PRINT(printf("in original %d infeasibilities summing to %g\n",
nonLinearCost_->numberInfeasibilities(), nonLinearCost_->sumInfeasibilities()));
// Initialize weights
primalColumnPivot_ = new ClpPrimalColumnSteepest(10);
primalColumnPivot_->saveWeights(this, 2);
#ifndef NDEBUG
// Check status
ClpSimplex * xxxx = this;
int nBasic = 0;
for (iColumn = 0; iColumn < xxxx->numberRows_ + xxxx->numberColumns_; iColumn++)
if (xxxx->getStatus(iColumn) == basic)
nBasic++;
assert (nBasic == xxxx->numberRows_);
for (iRow = 0; iRow < xxxx->numberRows_; iRow++) {
int iPivot = xxxx->pivotVariable_[iRow];
assert (xxxx->getStatus(iPivot) == basic);
}
#endif
}
// Pass in Event handler (cloned and deleted at end)
void
ClpSimplex::passInEventHandler(const ClpEventHandler * eventHandler)
{
delete eventHandler_;
eventHandler_ = eventHandler->clone();
eventHandler_->setSimplex(this);
}
#ifndef NDEBUG
// For errors to make sure print to screen
// only called in debug mode
static void indexError(int index,
std::string methodName)
{
std::cerr << "Illegal index " << index << " in ClpSimplex::" << methodName << std::endl;
throw CoinError("Illegal index", methodName, "ClpSimplex");
}
#endif
// These are only to be used using startFinishOptions (ClpSimplexDual, ClpSimplexPrimal)
//Get a row of the tableau (slack part in slack if not NULL)
void
ClpSimplex::getBInvARow(int row, double* z, double * slack)
{
#ifndef NDEBUG
int n = numberRows();
if (row < 0 || row >= n) {
indexError(row, "getBInvARow");
}
#endif
if (!rowArray_[0]) {
printf("ClpSimplexPrimal or ClpSimplexDual must have been called with correct startFinishOption\n");
abort();
}
CoinIndexedVector * rowArray0 = rowArray(0);
CoinIndexedVector * rowArray1 = rowArray(1);
CoinIndexedVector * columnArray0 = columnArray(0);
CoinIndexedVector * columnArray1 = columnArray(1);
rowArray0->clear();
rowArray1->clear();
columnArray0->clear();
columnArray1->clear();
// put +1 in row
// But swap if pivot variable was slack as clp stores slack as -1.0
int pivot = pivotVariable_[row];
double value;
// And if scaled then adjust
if (!rowScale_) {
if (pivot < numberColumns_)
value = 1.0;
else
value = -1.0;
} else {
if (pivot < numberColumns_)
value = columnScale_[pivot];
else
value = -1.0 * inverseRowScale_[pivot-numberColumns_];
}
rowArray1->insert(row, value);
factorization_->updateColumnTranspose(rowArray0, rowArray1);
// put row of tableau in rowArray1 and columnArray0
clpMatrix()->transposeTimes(this, 1.0,
rowArray1, columnArray1, columnArray0);
if (!rowScale_) {
CoinMemcpyN(columnArray0->denseVector(), numberColumns_, z);
} else {
double * array = columnArray0->denseVector();
for (int i = 0; i < numberColumns_; i++)
z[i] = array[i] * inverseColumnScale_[i];
}
if (slack) {
if (!rowScale_) {
CoinMemcpyN(rowArray1->denseVector(), numberRows_, slack);
} else {
double * array = rowArray1->denseVector();
for (int i = 0; i < numberRows_; i++)
slack[i] = array[i] * rowScale_[i];
}
}
// don't need to clear everything always, but doesn't cost
rowArray0->clear();
rowArray1->clear();
columnArray0->clear();
columnArray1->clear();
}
//Get a row of the basis inverse
void
ClpSimplex::getBInvRow(int row, double* z)
{
#ifndef NDEBUG
int n = numberRows();
if (row < 0 || row >= n) {
indexError(row, "getBInvRow");
}
#endif
if (!rowArray_[0]) {
printf("ClpSimplexPrimal or ClpSimplexDual must have been called with correct startFinishOption\n");
abort();
}
ClpFactorization * factorization = factorization_;
CoinIndexedVector * rowArray0 = rowArray(0);
CoinIndexedVector * rowArray1 = rowArray(1);
rowArray0->clear();
rowArray1->clear();
// put +1 in row
// But swap if pivot variable was slack as clp stores slack as -1.0
double value = (pivotVariable_[row] < numberColumns_) ? 1.0 : -1.0;
// but scale
if (rowScale_) {
int pivot = pivotVariable_[row];
if (pivot < numberColumns_)
value *= columnScale_[pivot];
else
value /= rowScale_[pivot-numberColumns_];
}
rowArray1->insert(row, value);
factorization->updateColumnTranspose(rowArray0, rowArray1);
if (!rowScale_) {
CoinMemcpyN(rowArray1->denseVector(), numberRows_, z);
} else {
double * array = rowArray1->denseVector();
for (int i = 0; i < numberRows_; i++) {
z[i] = array[i] * rowScale_[i];
}
}
rowArray1->clear();
}
//Get a column of the tableau
void
ClpSimplex::getBInvACol(int col, double* vec)
{
if (!rowArray_[0]) {
printf("ClpSimplexPrimal or ClpSimplexDual should have been called with correct startFinishOption\n");
abort();
}
CoinIndexedVector * rowArray0 = rowArray(0);
CoinIndexedVector * rowArray1 = rowArray(1);
rowArray0->clear();
rowArray1->clear();
// get column of matrix
#ifndef NDEBUG
int n = numberColumns_ + numberRows_;
if (col < 0 || col >= n) {
indexError(col, "getBInvACol");
}
#endif
if (!rowScale_) {
if (col < numberColumns_) {
unpack(rowArray1, col);
} else {
rowArray1->insert(col - numberColumns_, 1.0);
}
} else {
if (col < numberColumns_) {
unpack(rowArray1, col);
double multiplier = 1.0 * inverseColumnScale_[col];
int number = rowArray1->getNumElements();
int * index = rowArray1->getIndices();
double * array = rowArray1->denseVector();
for (int i = 0; i < number; i++) {
int iRow = index[i];
// make sure not packed
assert (array[iRow]);
array[iRow] *= multiplier;
}
} else {
rowArray1->insert(col - numberColumns_, rowScale_[col-numberColumns_]);
}
}
factorization_->updateColumn(rowArray0, rowArray1, false);
// But swap if pivot variable was slack as clp stores slack as -1.0
double * array = rowArray1->denseVector();
if (!rowScale_) {
for (int i = 0; i < numberRows_; i++) {
double multiplier = (pivotVariable_[i] < numberColumns_) ? 1.0 : -1.0;
vec[i] = multiplier * array[i];
}
} else {
for (int i = 0; i < numberRows_; i++) {
int pivot = pivotVariable_[i];
if (pivot < numberColumns_)
vec[i] = array[i] * columnScale_[pivot];
else
vec[i] = - array[i] / rowScale_[pivot-numberColumns_];
}
}
rowArray1->clear();
}
//Get a column of the basis inverse
void
ClpSimplex::getBInvCol(int col, double* vec)
{
if (!rowArray_[0]) {
printf("ClpSimplexPrimal or ClpSimplexDual must have been called with correct startFinishOption\n");
abort();
}
CoinIndexedVector * rowArray0 = rowArray(0);
CoinIndexedVector * rowArray1 = rowArray(1);
rowArray0->clear();
rowArray1->clear();
#ifndef NDEBUG
int n = numberRows();
if (col < 0 || col >= n) {
indexError(col, "getBInvCol");
}
#endif
// put +1 in row
// but scale
double value;
if (!rowScale_) {
value = 1.0;
} else {
value = rowScale_[col];
}
rowArray1->insert(col, value);
factorization_->updateColumn(rowArray0, rowArray1, false);
// But swap if pivot variable was slack as clp stores slack as -1.0
double * array = rowArray1->denseVector();
if (!rowScale_) {
for (int i = 0; i < numberRows_; i++) {
double multiplier = (pivotVariable_[i] < numberColumns_) ? 1.0 : -1.0;
vec[i] = multiplier * array[i];
}
} else {
for (int i = 0; i < numberRows_; i++) {
int pivot = pivotVariable_[i];
double value = array[i];
if (pivot < numberColumns_)
vec[i] = value * columnScale_[pivot];
else
vec[i] = - value / rowScale_[pivot-numberColumns_];
}
}
rowArray1->clear();
}
/* Get basic indices (order of indices corresponds to the
order of elements in a vector retured by getBInvACol() and
getBInvCol()).
*/
void
ClpSimplex::getBasics(int* index)
{
if (!rowArray_[0]) {
printf("ClpSimplexPrimal or ClpSimplexDual must have been called with correct startFinishOption\n");
abort();
}
CoinAssert (index);
CoinMemcpyN(pivotVariable(), numberRows(), index);
}
/* Set an objective function coefficient */
void
ClpSimplex::setObjectiveCoefficient( int elementIndex, double elementValue )
{
#ifndef NDEBUG
if (elementIndex < 0 || elementIndex >= numberColumns_) {
indexError(elementIndex, "setObjectiveCoefficient");
}
#endif
if (objective()[elementIndex] != elementValue) {
objective()[elementIndex] = elementValue;
if ((whatsChanged_ & 1) != 0) {
// work arrays exist - update as well
whatsChanged_ &= ~64;
double direction = optimizationDirection_ * objectiveScale_;
if (!rowScale_) {
objectiveWork_[elementIndex] = direction * elementValue;
} else {
objectiveWork_[elementIndex] = direction * elementValue
* columnScale_[elementIndex];
}
}
}
}
/* Set a single row lower bound<br>
Use -DBL_MAX for -infinity. */
void
ClpSimplex::setRowLower( int elementIndex, double elementValue )
{
#ifndef NDEBUG
int n = numberRows_;
if (elementIndex < 0 || elementIndex >= n) {
indexError(elementIndex, "setRowLower");
}
#endif
if (elementValue < -1.0e27)
elementValue = -COIN_DBL_MAX;
if (rowLower_[elementIndex] != elementValue) {
rowLower_[elementIndex] = elementValue;
if ((whatsChanged_ & 1) != 0) {
// work arrays exist - update as well
whatsChanged_ &= ~16;
if (rowLower_[elementIndex] == -COIN_DBL_MAX) {
rowLowerWork_[elementIndex] = -COIN_DBL_MAX;
} else if (!rowScale_) {
rowLowerWork_[elementIndex] = elementValue * rhsScale_;
} else {
rowLowerWork_[elementIndex] = elementValue * rhsScale_
* rowScale_[elementIndex];
}
}
}
}
/* Set a single row upper bound<br>
Use DBL_MAX for infinity. */
void
ClpSimplex::setRowUpper( int elementIndex, double elementValue )
{
#ifndef NDEBUG
int n = numberRows_;
if (elementIndex < 0 || elementIndex >= n) {
indexError(elementIndex, "setRowUpper");
}
#endif
if (elementValue > 1.0e27)
elementValue = COIN_DBL_MAX;
if (rowUpper_[elementIndex] != elementValue) {
rowUpper_[elementIndex] = elementValue;
if ((whatsChanged_ & 1) != 0) {
// work arrays exist - update as well
whatsChanged_ &= ~32;
if (rowUpper_[elementIndex] == COIN_DBL_MAX) {
rowUpperWork_[elementIndex] = COIN_DBL_MAX;
} else if (!rowScale_) {
rowUpperWork_[elementIndex] = elementValue * rhsScale_;
} else {
rowUpperWork_[elementIndex] = elementValue * rhsScale_
* rowScale_[elementIndex];
}
}
}
}
/* Set a single row lower and upper bound */
void
ClpSimplex::setRowBounds( int elementIndex,
double lowerValue, double upperValue )
{
#ifndef NDEBUG
int n = numberRows_;
if (elementIndex < 0 || elementIndex >= n) {
indexError(elementIndex, "setRowBounds");
}
#endif
if (lowerValue < -1.0e27)
lowerValue = -COIN_DBL_MAX;
if (upperValue > 1.0e27)
upperValue = COIN_DBL_MAX;
//CoinAssert (upperValue>=lowerValue);
if (rowLower_[elementIndex] != lowerValue) {
rowLower_[elementIndex] = lowerValue;
if ((whatsChanged_ & 1) != 0) {
// work arrays exist - update as well
whatsChanged_ &= ~16;
if (rowLower_[elementIndex] == -COIN_DBL_MAX) {
rowLowerWork_[elementIndex] = -COIN_DBL_MAX;
} else if (!rowScale_) {
rowLowerWork_[elementIndex] = lowerValue * rhsScale_;
} else {
rowLowerWork_[elementIndex] = lowerValue * rhsScale_
* rowScale_[elementIndex];
}
}
}
if (rowUpper_[elementIndex] != upperValue) {
rowUpper_[elementIndex] = upperValue;
if ((whatsChanged_ & 1) != 0) {
// work arrays exist - update as well
whatsChanged_ &= ~32;
if (rowUpper_[elementIndex] == COIN_DBL_MAX) {
rowUpperWork_[elementIndex] = COIN_DBL_MAX;
} else if (!rowScale_) {
rowUpperWork_[elementIndex] = upperValue * rhsScale_;
} else {
rowUpperWork_[elementIndex] = upperValue * rhsScale_
* rowScale_[elementIndex];
}
}
}
}
void ClpSimplex::setRowSetBounds(const int* indexFirst,
const int* indexLast,
const double* boundList)
{
#ifndef NDEBUG
int n = numberRows_;
#endif
int numberChanged = 0;
const int * saveFirst = indexFirst;
while (indexFirst != indexLast) {
const int iRow = *indexFirst++;
#ifndef NDEBUG
if (iRow < 0 || iRow >= n) {
indexError(iRow, "setRowSetBounds");
}
#endif
double lowerValue = *boundList++;
double upperValue = *boundList++;
if (lowerValue < -1.0e27)
lowerValue = -COIN_DBL_MAX;
if (upperValue > 1.0e27)
upperValue = COIN_DBL_MAX;
//CoinAssert (upperValue>=lowerValue);
if (rowLower_[iRow] != lowerValue) {
rowLower_[iRow] = lowerValue;
whatsChanged_ &= ~16;
numberChanged++;
}
if (rowUpper_[iRow] != upperValue) {
rowUpper_[iRow] = upperValue;
whatsChanged_ &= ~32;
numberChanged++;
}
}
if (numberChanged && (whatsChanged_ & 1) != 0) {
indexFirst = saveFirst;
while (indexFirst != indexLast) {
const int iRow = *indexFirst++;
if (rowLower_[iRow] == -COIN_DBL_MAX) {
rowLowerWork_[iRow] = -COIN_DBL_MAX;
} else if (!rowScale_) {
rowLowerWork_[iRow] = rowLower_[iRow] * rhsScale_;
} else {
rowLowerWork_[iRow] = rowLower_[iRow] * rhsScale_
* rowScale_[iRow];
}
if (rowUpper_[iRow] == COIN_DBL_MAX) {
rowUpperWork_[iRow] = COIN_DBL_MAX;
} else if (!rowScale_) {
rowUpperWork_[iRow] = rowUpper_[iRow] * rhsScale_;
} else {
rowUpperWork_[iRow] = rowUpper_[iRow] * rhsScale_
* rowScale_[iRow];
}
}
}
}
//-----------------------------------------------------------------------------
/* Set a single column lower bound<br>
Use -DBL_MAX for -infinity. */
void
ClpSimplex::setColumnLower( int elementIndex, double elementValue )
{
#ifndef NDEBUG
int n = numberColumns_;
if (elementIndex < 0 || elementIndex >= n) {
indexError(elementIndex, "setColumnLower");
}
#endif
if (elementValue < -1.0e27)
elementValue = -COIN_DBL_MAX;
if (columnLower_[elementIndex] != elementValue) {
columnLower_[elementIndex] = elementValue;
if ((whatsChanged_ & 1) != 0) {
// work arrays exist - update as well
whatsChanged_ &= ~128;
double value;
if (columnLower_[elementIndex] == -COIN_DBL_MAX) {
value = -COIN_DBL_MAX;
} else if (!columnScale_) {
value = elementValue * rhsScale_;
} else {
value = elementValue * rhsScale_
/ columnScale_[elementIndex];
}
lower_[elementIndex] = value;
if (maximumRows_ >= 0)
lower_[elementIndex+maximumRows_+maximumColumns_] = value;
}
}
}
/* Set a single column upper bound<br>
Use DBL_MAX for infinity. */
void
ClpSimplex::setColumnUpper( int elementIndex, double elementValue )
{
#ifndef NDEBUG
int n = numberColumns_;
if (elementIndex < 0 || elementIndex >= n) {
indexError(elementIndex, "setColumnUpper");
}
#endif
if (elementValue > 1.0e27)
elementValue = COIN_DBL_MAX;
if (columnUpper_[elementIndex] != elementValue) {
columnUpper_[elementIndex] = elementValue;
if ((whatsChanged_ & 1) != 0) {
// work arrays exist - update as well
whatsChanged_ &= ~256;
double value;
if (columnUpper_[elementIndex] == COIN_DBL_MAX) {
value = COIN_DBL_MAX;
} else if (!columnScale_) {
value = elementValue * rhsScale_;
} else {
value = elementValue * rhsScale_
/ columnScale_[elementIndex];
}
//assert (columnUpperWork_==upper_);
upper_[elementIndex] = value;
if (maximumRows_ >= 0)
upper_[elementIndex+maximumRows_+maximumColumns_] = value;
}
}
}
/* Set a single column lower and upper bound */
void
ClpSimplex::setColumnBounds( int elementIndex,
double lowerValue, double upperValue )
{
#ifndef NDEBUG
int n = numberColumns_;
if (elementIndex < 0 || elementIndex >= n) {
indexError(elementIndex, "setColumnBounds");
}
#endif
if (lowerValue < -1.0e27)
lowerValue = -COIN_DBL_MAX;
if (columnLower_[elementIndex] != lowerValue) {
columnLower_[elementIndex] = lowerValue;
if ((whatsChanged_ & 1) != 0) {
// work arrays exist - update as well
whatsChanged_ &= ~128;
if (columnLower_[elementIndex] == -COIN_DBL_MAX) {
lower_[elementIndex] = -COIN_DBL_MAX;
} else if (!columnScale_) {
lower_[elementIndex] = lowerValue * rhsScale_;
} else {
lower_[elementIndex] = lowerValue * rhsScale_
/ columnScale_[elementIndex];
}
}
}
if (upperValue > 1.0e27)
upperValue = COIN_DBL_MAX;
//CoinAssert (upperValue>=lowerValue);
if (columnUpper_[elementIndex] != upperValue) {
columnUpper_[elementIndex] = upperValue;
if ((whatsChanged_ & 1) != 0) {
// work arrays exist - update as well
whatsChanged_ &= ~256;
if (columnUpper_[elementIndex] == COIN_DBL_MAX) {
upper_[elementIndex] = COIN_DBL_MAX;
} else if (!columnScale_) {
upper_[elementIndex] = upperValue * rhsScale_;
} else {
upper_[elementIndex] = upperValue * rhsScale_
/ columnScale_[elementIndex];
}
}
}
}
void ClpSimplex::setColumnSetBounds(const int* indexFirst,
const int* indexLast,
const double* boundList)
{
#ifndef NDEBUG
int n = numberColumns_;
#endif
int numberChanged = 0;
const int * saveFirst = indexFirst;
while (indexFirst != indexLast) {
const int iColumn = *indexFirst++;
#ifndef NDEBUG
if (iColumn < 0 || iColumn >= n) {
indexError(iColumn, "setColumnSetBounds");
}
#endif
double lowerValue = *boundList++;
double upperValue = *boundList++;
if (lowerValue < -1.0e27)
lowerValue = -COIN_DBL_MAX;
if (upperValue > 1.0e27)
upperValue = COIN_DBL_MAX;
//CoinAssert (upperValue>=lowerValue);
if (columnLower_[iColumn] != lowerValue) {
columnLower_[iColumn] = lowerValue;
whatsChanged_ &= ~16;
numberChanged++;
}
if (columnUpper_[iColumn] != upperValue) {
columnUpper_[iColumn] = upperValue;
whatsChanged_ &= ~32;
numberChanged++;
}
}
if (numberChanged && (whatsChanged_ & 1) != 0) {
indexFirst = saveFirst;
while (indexFirst != indexLast) {
const int iColumn = *indexFirst++;
if (columnLower_[iColumn] == -COIN_DBL_MAX) {
lower_[iColumn] = -COIN_DBL_MAX;
} else if (!columnScale_) {
lower_[iColumn] = columnLower_[iColumn] * rhsScale_;
} else {
lower_[iColumn] = columnLower_[iColumn] * rhsScale_
/ columnScale_[iColumn];
}
if (columnUpper_[iColumn] == COIN_DBL_MAX) {
upper_[iColumn] = COIN_DBL_MAX;
} else if (!columnScale_) {
upper_[iColumn] = columnUpper_[iColumn] * rhsScale_;
} else {
upper_[iColumn] = columnUpper_[iColumn] * rhsScale_
/ columnScale_[iColumn];
}
}
}
}
/* Just check solution (for internal use) - sets sum of
infeasibilities etc. */
void
ClpSimplex::checkSolutionInternal()
{
double dualTolerance = dblParam_[ClpDualTolerance];
double primalTolerance = dblParam_[ClpPrimalTolerance];
double nonLinearOffset = 0.0;
const double * objective = objective_->gradient(this, columnActivity_,
nonLinearOffset, true);
int iRow, iColumn;
assert (!rowObjective_);
objectiveValue_ = -nonLinearOffset;
// now look at solution
sumPrimalInfeasibilities_ = 0.0;
numberPrimalInfeasibilities_ = 0;
sumDualInfeasibilities_ = 0.0;
numberDualInfeasibilities_ = 0;
double maxmin = optimizationDirection_;
for (iRow = 0; iRow < numberRows_; iRow++) {
double dualValue = dual_[iRow] * maxmin;
double primalValue = rowActivity_[iRow];
double lower = rowLower_[iRow];
double upper = rowUpper_[iRow];
ClpSimplex::Status status = getRowStatus(iRow);
if (status != basic) {
if (lower == upper) {
status = ClpSimplex::isFixed;
} else if (primalValue > upper - primalTolerance) {
status = ClpSimplex::atUpperBound;
} else if (primalValue < lower + primalTolerance) {
status = ClpSimplex::atLowerBound;
}
setRowStatus(iRow, status);
}
if (primalValue > upper + primalTolerance) {
sumPrimalInfeasibilities_ += primalValue - upper - primalTolerance;
numberPrimalInfeasibilities_ ++;
} else if (primalValue < lower - primalTolerance) {
sumPrimalInfeasibilities_ += lower - primalValue - primalTolerance;
numberPrimalInfeasibilities_ ++;
} else {
switch(status) {
case basic:
case ClpSimplex::isFixed:
break;
case atUpperBound:
// dual should not be positive
if (dualValue > dualTolerance) {
sumDualInfeasibilities_ += dualValue - dualTolerance_;
numberDualInfeasibilities_ ++;
}
break;
case atLowerBound:
// dual should not be negative
if (dualValue < -dualTolerance) {
sumDualInfeasibilities_ -= dualValue + dualTolerance_;
numberDualInfeasibilities_ ++;
}
break;
case superBasic:
case isFree:
if (primalValue < upper - primalTolerance) {
// dual should not be negative
if (dualValue < -dualTolerance) {
sumDualInfeasibilities_ -= dualValue + dualTolerance_;
numberDualInfeasibilities_ ++;
}
}
if (primalValue > lower + primalTolerance) {
// dual should not be positive
if (dualValue > dualTolerance) {
sumDualInfeasibilities_ += dualValue - dualTolerance_;
numberDualInfeasibilities_ ++;
}
}
break;
}
}
}
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
double dualValue = reducedCost_[iColumn] * maxmin;
double primalValue = columnActivity_[iColumn];
objectiveValue_ += objective[iColumn] * primalValue;
double lower = columnLower_[iColumn];
double upper = columnUpper_[iColumn];
ClpSimplex::Status status = getColumnStatus(iColumn);
if (status != basic && lower == upper) {
status = ClpSimplex::isFixed;
setColumnStatus(iColumn, ClpSimplex::isFixed);
}
if (primalValue > upper + primalTolerance) {
sumPrimalInfeasibilities_ += primalValue - upper - primalTolerance;
numberPrimalInfeasibilities_ ++;
} else if (primalValue < lower - primalTolerance) {
sumPrimalInfeasibilities_ += lower - primalValue - primalTolerance;
numberPrimalInfeasibilities_ ++;
} else {
switch(status) {
case basic:
// dual should be zero
if (fabs(dualValue) > 10.0 * dualTolerance) {
sumDualInfeasibilities_ += fabs(dualValue) - dualTolerance_;
numberDualInfeasibilities_ ++;
//if (fabs(dualValue) > 1000.0 * dualTolerance)
//setColumnStatus(iColumn,superBasic); Maybe on a switch
}
break;
case ClpSimplex::isFixed:
break;
case atUpperBound:
// dual should not be positive
if (dualValue > dualTolerance) {
sumDualInfeasibilities_ += dualValue - dualTolerance_;
numberDualInfeasibilities_ ++;
}
break;
case atLowerBound:
// dual should not be negative
if (dualValue < -dualTolerance) {
sumDualInfeasibilities_ -= dualValue + dualTolerance_;
numberDualInfeasibilities_ ++;
}
break;
case superBasic:
case isFree:
if (primalValue < upper - primalTolerance) {
// dual should not be negative
if (dualValue < -dualTolerance) {
sumDualInfeasibilities_ -= dualValue + dualTolerance_;
numberDualInfeasibilities_ ++;
}
}
if (primalValue > lower + primalTolerance) {
// dual should not be positive
if (dualValue > dualTolerance) {
sumDualInfeasibilities_ += dualValue - dualTolerance_;
numberDualInfeasibilities_ ++;
}
}
break;
}
}
}
objectiveValue_ += objective_->nonlinearOffset();
// But do direction
objectiveValue_ *= optimizationDirection_;
if (!numberDualInfeasibilities_ &&
!numberPrimalInfeasibilities_)
problemStatus_ = 0;
else
problemStatus_ = -1;
}
/*
When scaling is on it is possible that the scaled problem
is feasible but the unscaled is not. Clp returns a secondary
status code to that effect. This option allows for a cleanup.
If you use it I would suggest 1.
This only affects actions when scaled optimal
0 - no action
1 - clean up using dual if primal infeasibility
2 - clean up using dual if dual infeasibility
3 - clean up using dual if primal or dual infeasibility
11,12,13 - as 1,2,3 but use primal
*/
#ifdef COUNT_CLEANUPS
static int n1 = 0;
static int n2 = 0;
static int n3 = 0;
#endif
int
ClpSimplex::cleanup(int cleanupScaling)
{
#ifdef COUNT_CLEANUPS
n1++;
#endif
int returnCode = 0;
if (!problemStatus_ && cleanupScaling) {
int check = cleanupScaling % 10;
bool primal = (secondaryStatus_ == 2 || secondaryStatus_ == 4);
bool dual = (secondaryStatus_ == 3 || secondaryStatus_ == 4);
if (((check & 1) != 0 && primal) || (((check & 2) != 0) && dual)) {
// need cleanup
int saveScalingFlag = scalingFlag_;
// say matrix changed
whatsChanged_ |= 1;
scaling(0);
if (cleanupScaling < 10) {
// dual
returnCode = this->dual();
} else {
// primal
returnCode = this->primal();
}
#ifdef COUNT_CLEANUPS
n2++;
n3 += numberIterations_;
//printf("**cleanup took %d iterations\n",numberIterations_);
#endif
scaling(saveScalingFlag);
}
}
return returnCode;
}
#ifdef COUNT_CLEANUPS
void printHowMany()
{
printf("There were %d cleanups out of %d solves and %d iterations\n",
n2, n1, n3);
}
#endif
#ifndef SLIM_CLP
#include "CoinWarmStartBasis.hpp"
// Returns a basis (to be deleted by user)
CoinWarmStartBasis *
ClpSimplex::getBasis() const
{
int iRow, iColumn;
CoinWarmStartBasis * basis = new CoinWarmStartBasis();
basis->setSize(numberColumns_, numberRows_);
if (statusExists()) {
// Flip slacks
int lookupA[] = {0, 1, 3, 2, 0, 2};
for (iRow = 0; iRow < numberRows_; iRow++) {
int iStatus = getRowStatus(iRow);
iStatus = lookupA[iStatus];
basis->setArtifStatus(iRow, static_cast<CoinWarmStartBasis::Status> (iStatus));
}
int lookupS[] = {0, 1, 2, 3, 0, 3};
for (iColumn = 0; iColumn < numberColumns_; iColumn++) {
int iStatus = getColumnStatus(iColumn);
iStatus = lookupS[iStatus];
basis->setStructStatus(iColumn, static_cast<CoinWarmStartBasis::Status> (iStatus));
}
}
return basis;
}
#endif
// Compute objective value from solution
void
ClpSimplex::computeObjectiveValue(bool useInternalArrays)
{
int iSequence;
objectiveValue_ = 0.0;
const double * obj = objective();
if (!useInternalArrays) {
for (iSequence = 0; iSequence < numberColumns_; iSequence++) {
double value = columnActivity_[iSequence];
objectiveValue_ += value * obj[iSequence];
}
// But remember direction as we are using external objective
objectiveValue_ *= optimizationDirection_;
} else if (!columnScale_) {
for (iSequence = 0; iSequence < numberColumns_; iSequence++) {
double value = columnActivityWork_[iSequence];
objectiveValue_ += value * obj[iSequence];
}
// But remember direction as we are using external objective
objectiveValue_ *= optimizationDirection_;
objectiveValue_ += objective_->nonlinearOffset();
objectiveValue_ /= (objectiveScale_ * rhsScale_);
} else {
for (iSequence = 0; iSequence < numberColumns_; iSequence++) {
double scaleFactor = columnScale_[iSequence];
double valueScaled = columnActivityWork_[iSequence];
objectiveValue_ += valueScaled * scaleFactor * obj[iSequence];
}
// But remember direction as we are using external objective
objectiveValue_ *= optimizationDirection_;
objectiveValue_ += objective_->nonlinearOffset();
objectiveValue_ /= (objectiveScale_ * rhsScale_);
}
}
// Compute minimization objective value from internal solution
double
ClpSimplex::computeInternalObjectiveValue()
{
int iSequence;
//double oldObj = objectiveValue_;
double objectiveValue = 0.0;
const double * obj = objective();
if (!columnScale_) {
for (iSequence = 0; iSequence < numberColumns_; iSequence++) {
double value = solution_[iSequence];
objectiveValue += value * obj[iSequence];
}
} else {
for (iSequence = 0; iSequence < numberColumns_; iSequence++) {
double value = solution_[iSequence] * columnScale_[iSequence];
objectiveValue += value * obj[iSequence];
}
}
objectiveValue *= optimizationDirection_ / rhsScale_;
objectiveValue -= dblParam_[ClpObjOffset];
return objectiveValue;
}
// Infeasibility/unbounded ray (NULL returned if none/wrong)
double *
ClpSimplex::infeasibilityRay(bool fullRay) const
{
double * array = NULL;
if (problemStatus_ == 1 && ray_) {
if (!fullRay) {
array = ClpCopyOfArray(ray_, numberRows_);
} else {
array = new double [numberRows_+numberColumns_];
memcpy(array,ray_,numberRows_*sizeof(double));
memset(array+numberRows_,0,numberColumns_*sizeof(double));
transposeTimes(-1.0,array,array+numberRows_);
}
#ifdef PRINT_RAY_METHOD
printf("Infeasibility ray obtained by algorithm %s - direction out %d\n",algorithm_>0 ?
"primal" : "dual",directionOut_);
#endif
}
return array;
}
// If user left factorization frequency then compute
void
ClpSimplex::defaultFactorizationFrequency()
{
if (factorizationFrequency() == 200) {
// User did not touch preset
const int cutoff1 = 10000;
const int base = 75;
const int freq0 = 50;
int frequency;
#ifndef ABC_INHERIT
const int cutoff2 = 100000;
const int freq1 = 200;
const int freq2 = 400;
const int maximum = 1000;
if (numberRows_ < cutoff1)
frequency = base + numberRows_ / freq0;
else if (numberRows_ < cutoff2)
frequency = base + cutoff1 / freq0 + (numberRows_ - cutoff1) / freq1;
else
frequency = base + cutoff1 / freq0 + (cutoff2 - cutoff1) / freq1 + (numberRows_ - cutoff2) / freq2;
#else
const int freq1 = 150;
const int maximum = 10000;
if (numberRows_ < cutoff1)
frequency = base + numberRows_ / freq0;
else
frequency = base + cutoff1 / freq0 + (numberRows_ - cutoff1) / freq1;
#endif
setFactorizationFrequency(CoinMin(maximum, frequency));
}
}
// Gets clean and emptyish factorization
ClpFactorization *
ClpSimplex::getEmptyFactorization()
{
if ((specialOptions_ & 65536) == 0) {
assert (!factorization_);
factorization_ = new ClpFactorization();
} else if (!factorization_) {
factorization_ = new ClpFactorization();
factorization_->setPersistenceFlag(1);
}
return factorization_;
}
// May delete or may make clean and emptyish factorization
void
ClpSimplex::setEmptyFactorization()
{
if (factorization_) {
factorization_->cleanUp();
if ((specialOptions_ & 65536) == 0) {
delete factorization_;
factorization_ = NULL;
} else if (factorization_) {
factorization_->almostDestructor();
}
}
}
/* Array persistence flag
If 0 then as now (delete/new)
1 then only do arrays if bigger needed
2 as 1 but give a bit extra if bigger needed
*/
void
ClpSimplex::setPersistenceFlag(int value)
{
if (value) {
//specialOptions_|=65536;
startPermanentArrays();
} else {
specialOptions_ &= ~65536;
}
if (factorization_)
factorization_->setPersistenceFlag(value);
}
// Move status and solution across
void
ClpSimplex::moveInfo(const ClpSimplex & rhs, bool justStatus)
{
objectiveValue_ = rhs.objectiveValue_;
numberIterations_ = rhs. numberIterations_;
problemStatus_ = rhs. problemStatus_;
secondaryStatus_ = rhs. secondaryStatus_;
if (numberRows_ == rhs.numberRows_ && numberColumns_ == rhs.numberColumns_ && !justStatus) {
if (rhs.status_) {
if (status_)
CoinMemcpyN(rhs.status_, numberRows_ + numberColumns_, status_);
else
status_ = CoinCopyOfArray(rhs.status_, numberRows_ + numberColumns_);
} else {
delete [] status_;
status_ = NULL;
}
CoinMemcpyN(rhs.columnActivity_, numberColumns_, columnActivity_);
CoinMemcpyN(rhs.reducedCost_, numberColumns_, reducedCost_);
CoinMemcpyN(rhs.rowActivity_, numberRows_, rowActivity_);
CoinMemcpyN(rhs.dual_, numberRows_, dual_);
}
}
// Save a copy of model with certain state - normally without cuts
void
ClpSimplex::makeBaseModel()
{
delete baseModel_;
baseModel_ = new ClpSimplex(*this);
}
// Switch off base model
void
ClpSimplex::deleteBaseModel()
{
delete baseModel_;
baseModel_ = NULL;
}
// Reset to base model
void
ClpSimplex::setToBaseModel(ClpSimplex * model)
{
if (!model)
model = baseModel_;
assert (model);
int multiplier = ((model->specialOptions_ & 65536) != 0) ? 2 : 1;
assert (multiplier == 2);
if (multiplier == 2) {
assert (model->maximumRows_ >= 0);
if (maximumRows_ < 0) {
specialOptions_ |= 65536;
maximumRows_ = model->maximumRows_;
maximumColumns_ = model->maximumColumns_;
}
}
COIN_DETAIL_PRINT(printf("resetbase a %d rows, %d maximum rows\n",
numberRows_, maximumRows_));
// temporary - later use maximumRows_ for rowUpper_ etc
assert (numberRows_ >= model->numberRows_);
abort();
}
// Start or reset using maximumRows_ and Columns_
bool
ClpSimplex::startPermanentArrays()
{
int maximumRows = maximumRows_;
int maximumColumns = maximumColumns_;
ClpModel::startPermanentArrays();
if (maximumRows != maximumRows_ ||
maximumColumns != maximumColumns_) {
#if 0
maximumInternalRows_ = maximumRows_;
maximumInternalColumns_ = maximumColumns_;
int numberTotal2 = (maximumRows_ + maximumColumns_) * 2;
delete [] cost_;
cost_ = new double[numberTotal2];
delete [] lower_;
delete [] upper_;
lower_ = new double[numberTotal2];
upper_ = new double[numberTotal2];
delete [] dj_;
dj_ = new double[numberTotal2];
delete [] solution_;
solution_ = new double[numberTotal2];
assert (scalingFlag_ > 0);
if (rowScale_ && rowScale_ != savedRowScale_)
delete [] rowScale_;
rowScale_ = NULL;
// Do initial scaling
delete [] savedRowScale_;
savedRowScale_ = new double [4*maximumRows_];
delete [] savedColumnScale_;
savedColumnScale_ = new double [4*maximumColumns_];
if (scalingFlag_ > 0) {
rowScale_ = savedRowScale_;
columnScale_ = savedColumnScale_;
if (matrix_->scale(this)) {
scalingFlag_ = -scalingFlag_; // not scaled after all
assert (!rowScale_);
}
int numberRows2 = numberRows_ + numberExtraRows_;
if (rowScale_) {
CoinMemcpyN(rowScale_, 2 * numberRows2, savedRowScale_ + 2 * maximumRows_);
CoinMemcpyN(columnScale_, 2 * numberColumns_, savedColumnScale_ + 2 * maximumColumns_);
} else {
abort();
CoinFillN(savedRowScale_ + 2 * maximumRows_, 2 * numberRows2, 1.0);
CoinFillN(savedColumnScale_ + 2 * maximumColumns_, 2 * numberColumns_, 1.0);
}
}
#else
createRim(63);
#endif
return true;
} else {
return false;
}
}
#include "ClpNode.hpp"
//#define COIN_DEVELOP
// Fathom - 1 if solution
int
ClpSimplex::fathom(void * stuff)
{
assert (stuff);
ClpNodeStuff * info = reinterpret_cast<ClpNodeStuff *> (stuff);
info->nNodes_ = 0;
// say can declare optimal
moreSpecialOptions_ |= 8;
int saveMaxIterations = maximumIterations();
setMaximumIterations((((moreSpecialOptions_&2048)==0) ? 100 : 2000)
+ 5 * (numberRows_ + numberColumns_));
double saveObjLimit;
getDblParam(ClpDualObjectiveLimit, saveObjLimit);
if (perturbation_<100) {
double limit = saveObjLimit * optimizationDirection_;
setDblParam(ClpDualObjectiveLimit,
(limit+1.0e-2+1.0e-7*fabs(limit))*optimizationDirection_);
}
#if 0
bool onOptimal = (numberColumns_==100);
double optVal[133];
{
memset(optVal, 0, sizeof(optVal));
#if 0
int intIndicesV[] = {61, 62, 65, 66, 67, 68, 69, 70};
double intSolnV[] = {4., 21., 4., 4., 6., 1., 25., 8.};
int vecLen = sizeof(intIndicesV) / sizeof(int);
for (int i = 0; i < vecLen; i++) {
optVal[intIndicesV[i]] = intSolnV[i];
}
#else
int intIndicesAt1[] = { 0, 18, 25, 36, 44, 59, 61, 77, 82, 93 };
int vecLen = sizeof(intIndicesAt1) / sizeof(int);
for (int i = 0; i < vecLen; i++) {
optVal[intIndicesAt1[i]] = 1;
}
#endif
}
if (numberColumns_ == 100) {
const char * integerType = integerInformation();
for (int i = 0; i < 100; i++) {
if (integerType[i]) {
if (columnLower_[i] > optVal[i] || columnUpper_[i] < optVal[i]) {
onOptimal = false;
break;
}
}
}
if (onOptimal) {
printf("On optimal path fathom\n");
}
}
#endif
if (info->presolveType_) {
// crunch down
bool feasible = true;
// Use dual region
double * rhs = dual_;
int * whichRow = new int[3*numberRows_];
int * whichColumn = new int[2*numberColumns_];
int nBound;
bool tightenBounds = ((specialOptions_ & 64) == 0) ? false : true;
ClpSimplex * small =
static_cast<ClpSimplexOther *> (this)->crunch(rhs, whichRow, whichColumn,
nBound, false, tightenBounds);
if (small) {
//double limit = 0.0;
//getDblParam(ClpDualObjectiveLimit, limit);
//printf("objlimit a %g",limit);
//small->getDblParam(ClpDualObjectiveLimit, limit);
//printf(" %g\n",limit);
// pack down pseudocosts
small->moreSpecialOptions_ = moreSpecialOptions_;
if (info->upPseudo_) {
const char * integerType2 = small->integerInformation();
int n = small->numberColumns();
int k = 0;
int jColumn = 0;
int j = 0;
for (int i = 0; i < n; i++) {
if (integerType2[i]) {
int iColumn = whichColumn[i];
// find
while (jColumn != iColumn) {
if (integerType_[jColumn])
j++;
jColumn++;
}
info->priority_[k] = info->priority_[j];
info->upPseudo_[k] = info->upPseudo_[j];
info->numberUp_[k] = info->numberUp_[j];
info->numberUpInfeasible_[k] = info->numberUpInfeasible_[j];
info->downPseudo_[k] = info->downPseudo_[j];
info->numberDown_[k] = info->numberDown_[j];
info->numberDownInfeasible_[k] = info->numberDownInfeasible_[j];
assert (info->upPseudo_[k] > 1.0e-40 && info->upPseudo_[k] < 1.0e40);
assert (info->downPseudo_[k] > 1.0e-40 && info->downPseudo_[k] < 1.0e40);
k++;
}
}
}
#if 0
small->dual();
if (small->problemStatus() == 0) {
//problemStatus_ = 0;
} else if (small->problemStatus() != 3) {
feasible = false;
} else {
if (small->problemStatus_ == 3) {
// may be problems
printf("need coding from OsiClp for crunch\n");
abort();
}
}
#endif
} else {
feasible = false;
}
int returnCode = 0;
if (feasible) {
info->presolveType_ = 0;
// save and move pseudo costs
returnCode = small->fathom(stuff);
// restore pseudocosts
if (info->upPseudo_) {
int n = small->numberColumns();
int * back = new int [numberColumns_];
int numberIntegers = 0;
for (int i = 0; i < numberColumns_; i++) {
if (integerType_[i]) {
back[i] = -10 - numberIntegers;
numberIntegers++;
} else {
back[i] = -1;
}
}
const char * integerType2 = small->integerInformation();
int numberIntegers2 = 0;
for (int i = 0; i < n; i++) {
int iColumn = whichColumn[i];
if (integerType2[i]) {
int iBack = -back[iColumn];
assert (iBack >= 10);
iBack -= 10;
back[iColumn] = iBack;
numberIntegers2++;
}
}
int k = numberIntegers2;
for (int i = numberColumns_ - 1; i >= 0; i--) {
int iBack = back[i];
if (iBack <= -10) {
// fixed integer
numberIntegers--;
info->numberUp_[numberIntegers] = -1; // say not updated
} else if (iBack >= 0) {
// not fixed integer
numberIntegers--;
k--;
assert (info->upPseudo_[k] > 1.0e-40 && info->upPseudo_[k] < 1.0e40);
assert (info->downPseudo_[k] > 1.0e-40 && info->downPseudo_[k] < 1.0e40);
info->upPseudo_[numberIntegers] = info->upPseudo_[k];
info->numberUp_[numberIntegers] = info->numberUp_[k];
info->numberUpInfeasible_[numberIntegers] = info->numberUpInfeasible_[k];
info->downPseudo_[numberIntegers] = info->downPseudo_[k];
info->numberDown_[numberIntegers] = info->numberDown_[k];
info->numberDownInfeasible_[numberIntegers] = info->numberDownInfeasible_[k];
}
}
delete [] back;
}
if (returnCode) {
bool fixBounds = (info->nNodes_ >= 0) ? true : false;
//check this does everything
static_cast<ClpSimplexOther *> (this)->afterCrunch(*small,
whichRow, whichColumn, nBound);
bool badSolution = false;
for (int i = 0; i < numberColumns_; i++) {
if (integerType_[i]) {
double value = columnActivity_[i];
double value2 = floor(value + 0.5);
if (fabs(value - value2) >= 1.0e-4) {
// Very odd - can't use
badSolution = true;
}
columnActivity_[i] = value2;
if (fixBounds) {
columnLower_[i] = value2;
columnUpper_[i] = value2;
}
}
}
if (badSolution) {
info->nNodes_ = -1;
returnCode = 0;
}
//setLogLevel(63);
//double objectiveValue=doubleCheck();
//printf("Solution of %g\n",objectiveValue);
}
delete small;
}
delete [] whichRow;
delete [] whichColumn;
setMaximumIterations(saveMaxIterations);
setDblParam(ClpDualObjectiveLimit, saveObjLimit);
return returnCode;
}
int returnCode = startFastDual2(info);
if (returnCode) {
stopFastDual2(info);
setMaximumIterations(saveMaxIterations);
setDblParam(ClpDualObjectiveLimit, saveObjLimit);
return returnCode;
}
// Get fake bounds correctly
//(static_cast<ClpSimplexDual *>(this))->resetFakeBounds(1);
gutsOfSolution ( NULL, NULL);
double dummyChange;
(static_cast<ClpSimplexDual *>(this))->changeBounds(3, NULL, dummyChange);
int saveNumberFake = numberFake_;
int status = fastDual2(info);
#if 0
{
int iPivot;
double * array = rowArray_[3]->denseVector();
int i;
for (iPivot = 0; iPivot < numberRows_; iPivot++) {
int iSequence = pivotVariable_[iPivot];
unpack(rowArray_[3], iSequence);
factorization_->updateColumn(rowArray_[2], rowArray_[3]);
assert (fabs(array[iPivot] - 1.0) < 1.0e-4);
array[iPivot] = 0.0;
for (i = 0; i < numberRows_; i++)
assert (fabs(array[i]) < 1.0e-4);
rowArray_[3]->clear();
}
}
#endif
CoinAssert (problemStatus_ || objectiveValue_ < 1.0e50);
if (status && problemStatus_ != 3) {
// not finished - might be optimal
checkPrimalSolution(rowActivityWork_, columnActivityWork_);
double limit = 0.0;
getDblParam(ClpDualObjectiveLimit, limit);
//printf("objlimit b %g\n",limit);
if (!numberPrimalInfeasibilities_ && objectiveValue()*optimizationDirection_ < limit) {
problemStatus_ = 0;
}
status = problemStatus_;
}
if (problemStatus_ != 0 && problemStatus_ != 1) {
#ifdef COIN_DEVELOP
printf("bad status %d on initial fast dual %d its\n", problemStatus_,
numberIterations_);
#endif
info->nNodes_ = -1;
setMaximumIterations(saveMaxIterations);
setDblParam(ClpDualObjectiveLimit, saveObjLimit);
return 0;
}
int numberNodes = 1;
int numberIterations = numberIterations_;
#if defined(COIN_DEVELOP) || !defined(NO_FATHOM_PRINT)
int printFrequency = 2000;
#endif
if (problemStatus_ == 1) {
//printf("fathom infeasible on initial\n");
stopFastDual2(info);
info->numberNodesExplored_ = 1;
info->numberIterations_ = numberIterations;
setMaximumIterations(saveMaxIterations);
setDblParam(ClpDualObjectiveLimit, saveObjLimit);
return 0;
} else if (problemStatus_ != 0) {
stopFastDual2(info);
info->numberNodesExplored_ = 1;
info->numberIterations_ = numberIterations;
setMaximumIterations(saveMaxIterations);
setDblParam(ClpDualObjectiveLimit, saveObjLimit);
// say bad
info->nNodes_ = -1;
return 0;
}
if (!columnScale_) {
CoinMemcpyN(solution_, numberColumns_, columnActivity_);
} else {
assert(columnActivity_);
assert(columnScale_);
assert(solution_);
int j;
for (j = 0; j < numberColumns_; j++)
columnActivity_[j] = solution_[j] * columnScale_[j];
}
double increment = info->integerIncrement_;
int maxDepthSize = 10;
int maxDepth = 0;
int depth = 0;
// Get fake bounds correctly
(static_cast<ClpSimplexDual *>(this))->changeBounds(3, NULL, dummyChange);
saveNumberFake = numberFake_;
ClpNode ** nodes = new ClpNode * [maxDepthSize];
int numberTotal = numberRows_ + numberColumns_;
double * saveLower = CoinCopyOfArray(columnLower_, numberColumns_);
double * saveUpper = CoinCopyOfArray(columnUpper_, numberColumns_);
double * saveLowerInternal = CoinCopyOfArray(lower_, numberTotal);
double * saveUpperInternal = CoinCopyOfArray(upper_, numberTotal);
double * bestLower = NULL;
double * bestUpper = NULL;
int * back = new int [numberColumns_];
int numberIntegers = 0;
double sumChanges = 1.0e-5;
int numberChanges = 1;
for (int i = 0; i < numberColumns_; i++) {
if (integerType_[i])
back[i] = numberIntegers++;
else
back[i] = -1;
}
unsigned char * bestStatus = NULL;
double bestObjective;
getDblParam(ClpDualObjectiveLimit, bestObjective);
double saveBestObjective = bestObjective;
bool backtrack = false;
bool printing = handler_->logLevel() > 0;
while (depth >= 0) {
// If backtrack get to correct depth
if (backtrack) {
depth--;
while (depth >= 0) {
if (!nodes[depth]->fathomed()) {
nodes[depth]->changeState();
break;
}
//if (printing)
//printf("deleting node at depth %d\n",depth);
//delete nodes[depth];
//nodes[depth]=NULL;
depth--;
}
if (depth < 0)
break;
// apply
// First if backtracking we need to restore factorization, bounds and weights
CoinMemcpyN(saveLowerInternal, numberTotal, lower_);
CoinMemcpyN(saveUpperInternal, numberTotal, upper_);
CoinMemcpyN(saveLower, numberColumns_, columnLower_);
CoinMemcpyN(saveUpper, numberColumns_, columnUpper_);
for (int i = 0; i < depth; i++) {
nodes[i]->applyNode(this, 0);
}
nodes[depth]->applyNode(this, 1);
int iColumn = nodes[depth]->sequence();
if (printing)
printf("after backtracking - applying node at depth %d - variable %d (%g,%g)\n",
depth, iColumn,
columnLower_[iColumn], columnUpper_[iColumn]);
depth++;
} else {
// just bounds
if (depth > 0) {
nodes[depth-1]->applyNode(this, 0);
int iColumn = nodes[depth-1]->sequence();
if (printing)
printf("No backtracking - applying node at depth-m %d - variable %d (%g,%g)\n",
depth - 1, iColumn,
columnLower_[iColumn], columnUpper_[iColumn]);
}
}
// solve
#if 0
{
int iPivot;
double * array = rowArray_[3]->denseVector();
int i;
for (iPivot = 0; iPivot < numberRows_; iPivot++) {
int iSequence = pivotVariable_[iPivot];
unpack(rowArray_[3], iSequence);
factorization_->updateColumn(rowArray_[2], rowArray_[3]);
assert (fabs(array[iPivot] - 1.0) < 1.0e-4);
array[iPivot] = 0.0;
for (i = 0; i < numberRows_; i++)
assert (fabs(array[i]) < 1.0e-4);
rowArray_[3]->clear();
}
}
#endif
#ifdef COIN_DEVELOP
static int zzzzzz=0;
zzzzzz++;
if ((zzzzzz%100000)==0)
printf("%d fathom solves\n",zzzzzz);
if (zzzzzz==-1) {
printf("TROUBLE\n");
}
#endif
// Get fake bounds correctly
(static_cast<ClpSimplexDual *>(this))->changeBounds(3, NULL, dummyChange);
fastDual2(info);
#if 0
{
int iPivot;
double * array = rowArray_[3]->denseVector();
int i;
for (iPivot = 0; iPivot < numberRows_; iPivot++) {
int iSequence = pivotVariable_[iPivot];
unpack(rowArray_[3], iSequence);
factorization_->updateColumn(rowArray_[2], rowArray_[3]);
assert (fabs(array[iPivot] - 1.0) < 1.0e-4);
array[iPivot] = 0.0;
for (i = 0; i < numberRows_; i++)
assert (fabs(array[i]) < 1.0e-4);
rowArray_[3]->clear();
}
}
#endif
// give up if odd
if (problemStatus_ > 1) {
info->nNodes_ = -1;
#ifdef COIN_DEVELOP
printf("OUCH giving up on loop! %d %d %d %d - zzzzzz %d - max %d\n",
numberNodes, numberIterations, problemStatus_, numberIterations_,zzzzzz,intParam_[0]);
printf("xx %d\n", numberIterations*(numberRows_ + numberColumns_));
//abort();
#endif
break;
}
numberNodes++;
numberIterations += numberIterations_;
if ((numberNodes % 1000) == 0) {
#ifdef COIN_DEVELOP
if ((numberNodes % printFrequency) == 0) {
printf("Fathoming from node %d - %d nodes (%d iterations) - current depth %d\n",
info->nodeCalled_,numberNodes,
numberIterations, depth+info->startingDepth_);
printFrequency *= 2;
}
#elif !defined(NO_FATHOM_PRINT)
if ((numberNodes % printFrequency) == 0) {
if ((moreSpecialOptions_&2048)!=0)
info->handler_->message(CLP_FATHOM_STATUS, messages_)
<< info->nodeCalled_ << numberNodes
<< numberIterations << depth+info->startingDepth_
<< CoinMessageEol;
printFrequency *= 2;
}
#endif
if ((numberIterations*(numberRows_ + numberColumns_) > 5.0e10 ||
numberNodes > 2.0e4) &&
(moreSpecialOptions_&4096)==0) {
// give up
info->nNodes_ = -1;
#ifdef COIN_DEVELOP
printf("OUCH giving up on nodes %d %d\n", numberNodes, numberIterations);
printf("xx %d\n", numberIterations*(numberRows_ + numberColumns_));
//abort();
#endif
break;
}
}
if (problemStatus_ == 1 ||
(problemStatus_ == 0 && objectiveValue()*optimizationDirection_ > bestObjective)) {
backtrack = true;
if (printing)
printf("infeasible at depth %d\n", depth);
if (depth > 0) {
int way = nodes[depth-1]->way();
int sequence = nodes[depth-1]->sequence();
#ifndef NDEBUG
double branchingValue = nodes[depth-1]->branchingValue();
if (way > 0)
assert (columnLower_[sequence] == ceil(branchingValue));
else
assert (columnUpper_[sequence] == floor(branchingValue));
#endif
sequence = back[sequence];
double change = bestObjective - nodes[depth-1]->objectiveValue();
if (change > 1.0e10)
change = 10.0 * sumChanges / (1.0 + numberChanges);
info->update(way, sequence, change, false);
}
} else if (problemStatus_ != 0) {
abort();
} else {
// Create node
ClpNode * node;
computeDuals(NULL);
if (depth > 0) {
int way = nodes[depth-1]->way();
int sequence = nodes[depth-1]->sequence();
#ifndef NDEBUG
double branchingValue = nodes[depth-1]->branchingValue();
if (way > 0)
assert (columnLower_[sequence] == ceil(branchingValue));
else
assert (columnUpper_[sequence] == floor(branchingValue));
#endif
sequence = back[sequence];
info->update(way, sequence,
objectiveValue() - nodes[depth-1]->objectiveValue(),
true);
numberChanges++;
sumChanges += objectiveValue() - nodes[depth-1]->objectiveValue();
}
if (depth < maxDepth) {
node = nodes[depth];
node->gutsOfConstructor(this, info, 1, depth);
} else {
node = new ClpNode(this, info, depth);
if (depth == maxDepthSize) {
maxDepthSize = 2 * maxDepthSize + 10;
ClpNode ** temp = new ClpNode * [maxDepthSize];
for (int i = 0; i < depth; i++)
temp[i] = nodes[i];
delete [] nodes;
nodes = temp;
}
nodes[maxDepth++] = node;
}
#if 0
if (numberColumns_ == 100 && onOptimal) {
const char * integerType = integerInformation();
bool localOptimal=true;
for (int i = 0; i < 100; i++) {
if (integerType[i]) {
if (columnLower_[i] > optVal[i] || columnUpper_[i] < optVal[i]) {
localOptimal = false;
printf("bad %d %g %g %g\n", i, columnLower_[i], optVal[i],
columnUpper_[i]);
break;
}
}
}
if (localOptimal) {
printf("still on optimal\n");
}
assert (onOptimal);
}
#endif
double objectiveValue=0.0;
if (node->sequence() < 0) {
objectiveValue = doubleCheck();
node->gutsOfConstructor(this, info, 1, depth);
}
if (node->sequence() < 0) {
// solution
//double objectiveValue = doubleCheck();
if (objectiveValue < bestObjective) {
#ifdef COIN_DEVELOP
printf("Fathoming from node %d - solution of %g after %d nodes at depth %d\n",
info->nodeCalled_,objectiveValue,
numberNodes, depth+info->startingDepth_);
#elif !defined(NO_FATHOM_PRINT)
if ((moreSpecialOptions_&2048)!=0)
info->handler_->message(CLP_FATHOM_SOLUTION, messages_)
<< info->nodeCalled_ << objectiveValue
<< numberNodes << depth+info->startingDepth_
<< CoinMessageEol;
#endif
// later then lower_ not columnLower_ (and total?)
delete [] bestLower;
bestLower = CoinCopyOfArray(columnLower_, numberColumns_);
delete [] bestUpper;
bestUpper = CoinCopyOfArray(columnUpper_, numberColumns_);
delete [] bestStatus;
bestStatus = CoinCopyOfArray(status_, numberTotal);
bestObjective = objectiveValue - increment;
if (perturbation_<100)
bestObjective += 1.0e-2+1.0e-7*fabs(bestObjective);
setDblParam(ClpDualObjectiveLimit, bestObjective * optimizationDirection_);
} else {
//#define CLP_INVESTIGATE
#ifdef COIN_DEVELOP
printf("why bad solution feasible\n");
#endif
}
//delete node;
backtrack = true;
} else {
if (printing)
printf("depth %d variable %d\n", depth, node->sequence());
depth++;
backtrack = false;
//nodes[depth++] = new ClpNode (this,info);
}
}
}
if (!info->nNodes_)
assert (depth == -1);
for (int i = 0; i < maxDepth; i++)
delete nodes[i];
delete [] nodes;
delete [] back;
stopFastDual2(info);
#ifndef NO_FATHOM_PRINT
if ((moreSpecialOptions_&2048)!=0 && numberNodes >= 10000)
info->handler_->message(CLP_FATHOM_FINISH, messages_)
<< info->nodeCalled_ << info->startingDepth_
<< numberNodes << numberIterations << maxDepth+info->startingDepth_
<< CoinMessageEol;
#endif
//printf("fathom finished after %d nodes\n",numberNodes);
if (bestStatus) {
CoinMemcpyN(bestLower, numberColumns_, columnLower_);
CoinMemcpyN(bestUpper, numberColumns_, columnUpper_);
CoinMemcpyN(bestStatus, numberTotal, status_);
delete [] bestLower;
delete [] bestUpper;
delete [] bestStatus;
setDblParam(ClpDualObjectiveLimit, saveBestObjective);
saveObjLimit = saveBestObjective;
int saveOptions = specialOptions_;
specialOptions_ &= ~65536;
dual();
specialOptions_ = saveOptions;
info->numberNodesExplored_ = numberNodes;
info->numberIterations_ = numberIterations;
returnCode = 1;
} else {
info->numberNodesExplored_ = numberNodes;
info->numberIterations_ = numberIterations;
returnCode = 0;
}
if (info->nNodes_ < 0) {
if (lower_) {
CoinMemcpyN(saveLowerInternal, numberTotal, lower_);
CoinMemcpyN(saveUpperInternal, numberTotal, upper_);
numberFake_ = saveNumberFake;
}
CoinMemcpyN(saveLower, numberColumns_, columnLower_);
CoinMemcpyN(saveUpper, numberColumns_, columnUpper_);
}
delete [] saveLower;
delete [] saveUpper;
delete [] saveLowerInternal;
delete [] saveUpperInternal;
setMaximumIterations(saveMaxIterations);
setDblParam(ClpDualObjectiveLimit, saveObjLimit);
return returnCode;
}
//#define CHECK_PATH
#ifdef CHECK_PATH
const double * debuggerSolution_Z = NULL;
int numberColumns_Z = -1;
int gotGoodNode_Z = -1;
#endif
/* Do up to N deep - returns
-1 - no solution nNodes_ valid nodes
>= if solution and that node gives solution
ClpNode array is 2**N long. Values for N and
array are in stuff (nNodes_ also in stuff) */
int
ClpSimplex::fathomMany(void * stuff)
{
assert (stuff);
ClpNodeStuff * info = reinterpret_cast<ClpNodeStuff *> (stuff);
int nNodes = info->maximumNodes();
int putNode = info->maximumSpace();
int goodNodes = 0;
info->nNodes_ = 0;
ClpNode ** nodeInfo = info->nodeInfo_;
assert (nodeInfo);
// say can declare optimal
moreSpecialOptions_ |= 8;
double limit = 0.0;
getDblParam(ClpDualObjectiveLimit, limit);
for (int j = 0; j < putNode; j++) {
if (nodeInfo[j]) {
nodeInfo[j]->setObjectiveValue(limit);
if (info->large_)
nodeInfo[j]->cleanUpForCrunch();
}
}
#ifdef CHECK_PATH
// Note - if code working can get assert on startOptimal==2 (if finds)
int startOptimal = 0;
if (numberColumns_ == numberColumns_Z) {
assert (debuggerSolution_Z);
startOptimal = 1;
for (int i = 0; i < numberColumns_; i++) {
if (columnUpper_[i] < debuggerSolution_Z[i] || columnLower_[i] > debuggerSolution_Z[i]) {
startOptimal = 0;
break;
}
}
if (startOptimal) {
printf("starting on optimal\n");
}
} else if (info->large_ && info->large_->numberColumns_ == numberColumns_Z) {
assert (debuggerSolution_Z);
startOptimal = 1;
for (int i = 0; i < info->large_->numberColumns_; i++) {
if (info->large_->columnUpper_[i] < debuggerSolution_Z[i] || info->large_->columnLower_[i] > debuggerSolution_Z[i]) {
startOptimal = 0;
break;
}
}
if (startOptimal) {
printf("starting on optimal (presolved) %d\n", numberColumns_);
}
}
#endif
int whichSolution = -1;
if (info->presolveType_) {
// crunch down
bool feasible = true;
// Use dual region
double * rhs = dual_;
int * whichRow = new int[3*numberRows_];
int * whichColumn = new int[2*numberColumns_];
int nBound;
bool tightenBounds = ((specialOptions_ & 64) == 0) ? false : true;
ClpSimplex * small =
static_cast<ClpSimplexOther *> (this)->crunch(rhs, whichRow, whichColumn,
nBound, false, tightenBounds);
if (small) {
info->large_ = this;
info->whichRow_ = whichRow;
info->whichColumn_ = whichColumn;
info->nBound_ = nBound;
//double limit = 0.0;
//getDblParam(ClpDualObjectiveLimit, limit);
//printf("objlimit a %g",limit);
//small->getDblParam(ClpDualObjectiveLimit, limit);
//printf(" %g\n",limit);
// pack down pseudocosts
if (info->upPseudo_) {
const char * integerType2 = small->integerInformation();
int n = small->numberColumns();
int k = 0;
int jColumn = 0;
int j = 0;
for (int i = 0; i < n; i++) {
if (integerType2[i]) {
int iColumn = whichColumn[i];
// find
while (jColumn != iColumn) {
if (integerType_[jColumn])
j++;
jColumn++;
}
info->upPseudo_[k] = info->upPseudo_[j];
info->numberUp_[k] = info->numberUp_[j];
info->numberUpInfeasible_[k] = info->numberUpInfeasible_[j];
info->downPseudo_[k] = info->downPseudo_[j];
info->numberDown_[k] = info->numberDown_[j];
info->numberDownInfeasible_[k] = info->numberDownInfeasible_[j];
assert (info->upPseudo_[k] > 1.0e-40 && info->upPseudo_[k] < 1.0e40);
assert (info->downPseudo_[k] > 1.0e-40 && info->downPseudo_[k] < 1.0e40);
k++;
}
}
}
} else {
feasible = false;
}
if (feasible) {
info->presolveType_ = 0;
// save and move pseudo costs
whichSolution = small->fathomMany(stuff);
// restore pseudocosts
if (info->upPseudo_) {
int n = small->numberColumns();
int * back = new int [numberColumns_];
int numberIntegers = 0;
for (int i = 0; i < numberColumns_; i++) {
if (integerType_[i]) {
back[i] = -10 - numberIntegers;
numberIntegers++;
} else {
back[i] = -1;
}
}
const char * integerType2 = small->integerInformation();
int numberIntegers2 = 0;
for (int i = 0; i < n; i++) {
int iColumn = whichColumn[i];
if (integerType2[i]) {
int iBack = -back[iColumn];
assert (iBack >= 10);
iBack -= 10;
back[iColumn] = iBack;
numberIntegers2++;
}
}
int k = numberIntegers2;
for (int i = numberColumns_ - 1; i >= 0; i--) {
int iBack = back[i];
if (iBack <= -10) {
// fixed integer
numberIntegers--;
info->numberUp_[numberIntegers] = -1; // say not updated
} else if (iBack >= 0) {
// not fixed integer
numberIntegers--;
k--;
assert (info->upPseudo_[k] > 1.0e-40 && info->upPseudo_[k] < 1.0e40);
assert (info->downPseudo_[k] > 1.0e-40 && info->downPseudo_[k] < 1.0e40);
info->upPseudo_[numberIntegers] = info->upPseudo_[k];
info->numberUp_[numberIntegers] = info->numberUp_[k];
info->numberUpInfeasible_[numberIntegers] = info->numberUpInfeasible_[k];
info->downPseudo_[numberIntegers] = info->downPseudo_[k];
info->numberDown_[numberIntegers] = info->numberDown_[k];
info->numberDownInfeasible_[numberIntegers] = info->numberDownInfeasible_[k];
}
}
delete [] back;
}
delete small;
}
info->large_ = NULL;
info->whichRow_ = NULL;
info->whichColumn_ = NULL;
delete [] whichRow;
delete [] whichColumn;
return whichSolution;
}
#ifndef DEBUG
{
int nBasic = 0;
int i;
for (i = 0; i < numberRows_ + numberColumns_; i++) {
if (getColumnStatus(i) == basic)
nBasic++;
}
assert (nBasic == numberRows_);
}
#endif
int returnCode = startFastDual2(info);
if (returnCode) {
stopFastDual2(info);
abort();
return -1;
}
gutsOfSolution ( NULL, NULL);
int status = fastDual2(info);
CoinAssert (problemStatus_ || objectiveValue_ < 1.0e50);
if (status && problemStatus_ != 3) {
// not finished - might be optimal
checkPrimalSolution(rowActivityWork_, columnActivityWork_);
double limit = 0.0;
getDblParam(ClpDualObjectiveLimit, limit);
//printf("objlimit b %g\n",limit);
if (!numberPrimalInfeasibilities_ && objectiveValue()*optimizationDirection_ < limit) {
problemStatus_ = 0;
}
status = problemStatus_;
}
assert (problemStatus_ == 0 || problemStatus_ == 1); //(static_cast<ClpSimplexDual *> this)->dual(0,0);
if (problemStatus_ == 10) {
printf("Cleaning up with primal - need coding without createRim!\n");
abort();
}
int numberNodes = 0;
int numberIterations = numberIterations_;
if (problemStatus_ == 1) {
//printf("fathom infeasible on initial\n");
stopFastDual2(info);
info->nNodes_ = 0;
info->numberNodesExplored_ = 0;
info->numberIterations_ = numberIterations;
return -1;
} else if (problemStatus_ != 0) {
abort();
}
if (!columnScale_) {
CoinMemcpyN(solution_, numberColumns_, columnActivity_);
} else {
assert(columnActivity_);
assert(columnScale_);
assert(solution_);
int j;
for (j = 0; j < numberColumns_; j++)
columnActivity_[j] = solution_[j] * columnScale_[j];
}
double increment = info->integerIncrement_;
int depth = 0;
int numberTotal = numberRows_ + numberColumns_;
double * saveLower = CoinCopyOfArray(columnLower_, numberColumns_);
double * saveUpper = CoinCopyOfArray(columnUpper_, numberColumns_);
double * saveLowerInternal = CoinCopyOfArray(lower_, numberTotal);
double * saveUpperInternal = CoinCopyOfArray(upper_, numberTotal);
//double * bestLower = NULL;
//double * bestUpper = NULL;
int * back = new int [numberColumns_];
int numberIntegers = 0;
double sumChanges = 1.0e-5;
int numberChanges = 1;
for (int i = 0; i < numberColumns_; i++) {
if (integerType_[i])
back[i] = numberIntegers++;
else
back[i] = -1;
}
//unsigned char * bestStatus = NULL;
double bestObjective;
getDblParam(ClpDualObjectiveLimit, bestObjective);
double saveBestObjective = bestObjective;
bool backtrack = false;
bool printing = handler_->logLevel() > 0;
#ifdef CHECK_PATH
if (startOptimal)
printing = true;
#endif
/* Use nodeInfo for storage
depth 0 will be putNode-1, 1 putNode-2 etc */
int useDepth = putNode - 1;
bool justDive = (info->solverOptions_ & 32) != 0;
//printf("putNode %d nDepth %d\n");
while (depth >= 0) {
bool stopAtOnce = false;
// If backtrack get to correct depth
if (backtrack) {
depth--;
useDepth++;
while (depth >= 0) {
if (!nodeInfo[useDepth]->fathomed()) {
nodeInfo[useDepth]->changeState();
break;
}
//if (printing)
//printf("deleting node at depth %d\n",depth);
//delete nodes[useDepth];
//nodes[useDepth]=NULL;
depth--;
useDepth++;
}
if (depth < 0)
break;
// apply
// First if backtracking we need to restore factorization, bounds and weights
CoinMemcpyN(saveLowerInternal, numberTotal, lower_);
CoinMemcpyN(saveUpperInternal, numberTotal, upper_);
CoinMemcpyN(saveLower, numberColumns_, columnLower_);
CoinMemcpyN(saveUpper, numberColumns_, columnUpper_);
for (int i = 0; i < depth; i++) {
nodeInfo[putNode-1-i]->applyNode(this, 0);
}
nodeInfo[useDepth]->applyNode(this, 1);
if (justDive)
stopAtOnce = true;
int iColumn = nodeInfo[useDepth]->sequence();
if (printing)
printf("after backtracking - applying node at depth %d - variable %d (%g,%g)\n",
depth, iColumn,
columnLower_[iColumn], columnUpper_[iColumn]);
depth++;
useDepth--;
} else {
// just bounds
if (depth > 0) {
// Choose variable here!!
nodeInfo[useDepth+1]->chooseVariable(this, info);
nodeInfo[useDepth+1]->applyNode(this, 0);
int iColumn = nodeInfo[useDepth+1]->sequence();
if (printing)
printf("No backtracking - applying node at depth-m %d - variable %d (%g,%g)\n",
depth - 1, iColumn,
columnLower_[iColumn], columnUpper_[iColumn]);
}
}
// solve
double dummyChange;
(static_cast<ClpSimplexDual *>(this))->changeBounds(3, NULL, dummyChange);
//int saveNumberFake = numberFake_;
fastDual2(info);
numberNodes++;
numberIterations += numberIterations_;
if ((numberNodes % 1000) == 0 && printing)
printf("After %d nodes (%d iterations) - best solution %g - current depth %d\n",
numberNodes, numberIterations, bestObjective, depth);
if (problemStatus_ == 1 ||
(problemStatus_ == 0 && objectiveValue()*optimizationDirection_ > bestObjective)) {
backtrack = true;
if (printing)
printf("infeasible at depth %d\n", depth);
#ifdef CHECK_PATH
if (startOptimal && numberColumns_ == numberColumns_Z) {
bool onOptimal = true;
for (int i = 0; i < numberColumns_; i++) {
if (columnUpper_[i] < debuggerSolution_Z[i] || columnLower_[i] > debuggerSolution_Z[i]) {
onOptimal = false;
break;
}
}
if (onOptimal) {
printf("INF on optimal fathom at depth %d\n", depth);
abort();
}
} else if (info->large_ && startOptimal && info->large_->numberColumns_ == numberColumns_Z) {
bool onOptimal = true;
for (int i = 0; i < info->large_->numberColumns_; i++) {
if (info->large_->columnUpper_[i] < debuggerSolution_Z[i] || info->large_->columnLower_[i] > debuggerSolution_Z[i]) {
onOptimal = false;
break;
}
}
if (onOptimal) {
printf("INF on optimal (pre) fathom at depth %d\n", depth);
writeMps("fathom_pre.mps");
abort();
}
}
#endif
if (depth > 0) {
int way = nodeInfo[useDepth+1]->way();
int sequence = nodeInfo[useDepth+1]->sequence();
#ifndef NDEBUG
double branchingValue = nodeInfo[useDepth+1]->branchingValue();
if (way > 0)
assert (columnLower_[sequence] == ceil(branchingValue));
else
assert (columnUpper_[sequence] == floor(branchingValue));
#endif
sequence = back[sequence];
double change = bestObjective - nodeInfo[useDepth+1]->objectiveValue();
if (change > 1.0e10)
change = 10.0 * sumChanges / (1.0 + numberChanges);
info->update(way, sequence, change, false);
}
} else if (problemStatus_ != 0) {
abort();
} else {
// Create node
ClpNode * node;
computeDuals(NULL);
if (depth > 0) {
int way = nodeInfo[useDepth+1]->way();
int sequence = nodeInfo[useDepth+1]->sequence();
#ifndef NDEBUG
double branchingValue = nodeInfo[useDepth+1]->branchingValue();
if (way > 0)
assert (columnLower_[sequence] == ceil(branchingValue));
else
assert (columnUpper_[sequence] == floor(branchingValue));
#endif
sequence = back[sequence];
info->update(way, sequence,
objectiveValue() - nodeInfo[useDepth+1]->objectiveValue(),
true);
numberChanges++;
sumChanges += objectiveValue() - nodeInfo[useDepth+1]->objectiveValue();
}
#ifdef CHECK_PATH
if (startOptimal && numberColumns_ == numberColumns_Z) {
bool onOptimal = true;
for (int i = 0; i < numberColumns_; i++) {
if (columnUpper_[i] < debuggerSolution_Z[i] || columnLower_[i] > debuggerSolution_Z[i]) {
onOptimal = false;
break;
}
}
if (onOptimal) {
if (depth >= info->nDepth_) {
printf("on optimal fathom at full depth %d %d %g\n",
depth, goodNodes, objectiveValue());
gotGoodNode_Z = goodNodes;
startOptimal = 2;
} else {
printf("on optimal fathom at depth %d\n", depth);
}
}
} else if (info->large_ && startOptimal && info->large_->numberColumns_ == numberColumns_Z) {
bool onOptimal = true;
// Fix bounds in large
for (int i = 0; i < numberColumns_; i++) {
if (integerType_[i]) {
int iColumn = info->whichColumn_[i];
info->large_->columnUpper_[iColumn] = columnUpper_[i];
info->large_->columnLower_[iColumn] = columnLower_[i];
COIN_DETAIL_PRINT(printf("%d dj %g dual %g scale %g\n",
iColumn, dj_[i], reducedCost_[i], columnScale_[i]));
}
}
for (int i = 0; i < info->large_->numberColumns_; i++) {
if (info->large_->columnUpper_[i] < debuggerSolution_Z[i] || info->large_->columnLower_[i] > debuggerSolution_Z[i]) {
onOptimal = false;
break;
}
}
if (onOptimal) {
if (depth >= info->nDepth_) {
printf("on (pre) tentative optimal fathom at full depth %d %d %g\n",
depth, goodNodes, objectiveValue());
for (int i = 0; i < info->large_->numberColumns_; i++)
printf("fathomA %d %g %g\n", i, info->large_->columnLower_[i],
info->large_->columnUpper_[i]);
} else {
printf("on (pre) optimal fathom at depth %d\n", depth);
}
}
}
#endif
if (depth < info->nDepth_ && !stopAtOnce) {
node = nodeInfo[useDepth];
if (node) {
node->gutsOfConstructor(this, info, 1, depth);
} else {
node = new ClpNode(this, info, depth);
nodeInfo[useDepth] = node;
}
} else {
// save
node = nodeInfo[goodNodes];
if (!node) {
node = new ClpNode(this, info, depth);
nodeInfo[goodNodes] = node;
}
if (!node->oddArraysExist())
node->createArrays(this);
node->gutsOfConstructor(this, info, 2, depth);
}
if (node->sequence() < 0) {
// solution
double objectiveValue = doubleCheck();
if (printing)
printf("Solution of %g after %d nodes at depth %d\n",
objectiveValue, numberNodes, depth);
if (objectiveValue < bestObjective && !problemStatus_) {
// make sure node exists
node = nodeInfo[goodNodes];
if (!node) {
node = new ClpNode(this, info, depth);
nodeInfo[goodNodes] = node;
}
if (info->large_) {
//check this does everything
// Fix bounds in large
for (int i = 0; i < numberColumns_; i++) {
if (integerType_[i]) {
int iColumn = info->whichColumn_[i];
info->large_->columnUpper_[iColumn] = columnUpper_[i];
info->large_->columnLower_[iColumn] = columnLower_[i];
}
}
static_cast<ClpSimplexOther *> (info->large_)->afterCrunch(*this,
info->whichRow_, info->whichColumn_, info->nBound_);
// do as for large
if (!node->oddArraysExist())
node->createArrays(info->large_);
node->gutsOfConstructor(info->large_, info, 2, depth);
} else {
if (!node->oddArraysExist())
node->createArrays(this);
node->gutsOfConstructor(this, info, 2, depth);
}
whichSolution = goodNodes;
goodNodes++;
if (goodNodes >= nNodes)
justDive = true; // clean up phase
assert (node->sequence() < 0);
bestObjective = objectiveValue - increment;
setDblParam(ClpDualObjectiveLimit, bestObjective * optimizationDirection_);
} else {
#ifdef CLP_INVESTIGATE
printf("why bad solution feasible\n");
abort();
#endif
}
backtrack = true;
} else {
//if (printing)
//printf("depth %d variable %d\n",depth,node->sequence());
if (depth == info->nDepth_ || stopAtOnce) {
if (info->large_) {
//check this does everything
// Fix bounds in large
for (int i = 0; i < numberColumns_; i++) {
if (integerType_[i]) {
int iColumn = info->whichColumn_[i];
info->large_->columnUpper_[iColumn] = columnUpper_[i];
info->large_->columnLower_[iColumn] = columnLower_[i];
}
}
#ifdef CHECK_PATH
if (startOptimal)
for (int i = 0; i < info->large_->numberColumns_; i++)
printf("fathomB %d %g %g %g\n", i, info->large_->columnLower_[i],
info->large_->columnUpper_[i],
node->dualSolution()[i]);
#endif
static_cast<ClpSimplexOther *> (info->large_)->afterCrunch(*this,
info->whichRow_, info->whichColumn_, info->nBound_);
#ifdef CHECK_PATH
if (startOptimal) {
bool onOptimal = true;
for (int i = 0; i < info->large_->numberColumns_; i++)
printf("fathomC %d %g %g\n", i, info->large_->columnLower_[i],
info->large_->columnUpper_[i]);
for (int i = 0; i < info->large_->numberColumns_; i++) {
if (info->large_->columnUpper_[i] < debuggerSolution_Z[i] || info->large_->columnLower_[i] > debuggerSolution_Z[i]) {
onOptimal = false;
break;
}
}
if (onOptimal) {
printf("on (pre) optimal fathom at full depth %d %d %g\n",
depth, goodNodes, info->large_->objectiveValue());
startOptimal = 2;
gotGoodNode_Z = goodNodes;
for (int i = 0; i < info->large_->numberColumns_; i++)
printf("fathom %d %g %g\n", i, info->large_->columnLower_[i],
info->large_->columnUpper_[i]);
}
}
#endif
// do as for large
node->gutsOfConstructor(info->large_, info, 2, depth);
}
goodNodes++;
if (goodNodes >= nNodes)
justDive = true; // clean up phase
backtrack = true;
} else {
depth++;
useDepth--;
backtrack = false;
}
}
}
}
//printf("nNodes %d nDepth %d, useDepth %d goodNodes %d\n",
// nNodes,info->nDepth_,useDepth,goodNodes);
#ifdef CHECK_PATH
if (startOptimal) {
assert(startOptimal == 2);
printf("got fathomed optimal at end %d\n", startOptimal);
if (startOptimal != 2)
abort();
}
#endif
assert (depth == -1);
delete [] saveLower;
delete [] saveUpper;
delete [] saveLowerInternal;
delete [] saveUpperInternal;
delete [] back;
//printf("fathom finished after %d nodes\n",numberNodes);
if (whichSolution >= 0) {
setDblParam(ClpDualObjectiveLimit, saveBestObjective);
}
stopFastDual2(info);
info->nNodes_ = goodNodes;
info->numberNodesExplored_ = numberNodes;
info->numberIterations_ = numberIterations;
return whichSolution;
}
// Double checks OK
double
ClpSimplex::doubleCheck()
{
#if 0
double * solution = CoinCopyOfArray(solution_, numberColumns_ + numberRows_);
gutsOfSolution ( NULL, NULL);
for (int i = 0; i < numberColumns_; i++) {
if (fabs(solution[i] - solution_[i]) > 1.0e-7)
printf("bada %d bad %g good %g\n",
i, solution[i], solution_[i]);
}
//abort();
#endif
// make sure clean
whatsChanged_=0;
dual(0, 7);
#if 0
for (int i = 0; i < numberColumns_; i++) {
if (fabs(solution[i] - solution_[i]) > 1.0e-7)
printf("badb %d bad %g good %g\n",
i, solution[i], solution_[i]);
}
dual(0, 1);
for (int i = 0; i < numberColumns_; i++) {
if (fabs(solution[i] - solution_[i]) > 1.0e-7)
printf("badc %d bad %g good %g\n",
i, solution[i], solution_[i]);
}
delete [] solution;
#endif
return objectiveValue() * optimizationDirection_;
}
// Start Fast dual
int
ClpSimplex::startFastDual2(ClpNodeStuff * info)
{
info->saveOptions_ = specialOptions_;
assert ((info->solverOptions_ & 65536) == 0);
info->solverOptions_ |= 65536;
if ((specialOptions_ & 65536) == 0) {
factorization_->setPersistenceFlag(2);
} else {
factorization_->setPersistenceFlag(2);
startPermanentArrays();
}
//assert (!lower_);
// create modifiable copies of model rim and do optional scaling
createRim(7 + 8 + 16 + 32, true, 0);
#ifndef NDEBUG
ClpPackedMatrix* clpMatrix =
dynamic_cast< ClpPackedMatrix*>(matrix_);
assert (clpMatrix && (clpMatrix->flags() & 1) == 0);
#endif
// mark all as current
whatsChanged_ = 0x3ffffff;
// change newLower and newUpper if scaled
// Do initial factorization
// and set certain stuff
// We can either set increasing rows so ...IsBasic gives pivot row
// or we can just increment iBasic one by one
// for now let ...iBasic give pivot row
int factorizationStatus = internalFactorize(0);
if (factorizationStatus < 0 ||
(factorizationStatus && factorizationStatus <= numberRows_)) {
// some error
#if 0
// we should either debug or ignore
#ifdef CLP_INVESTIGATE
//#ifndef NDEBUG
printf("***** ClpDual strong branching factorization error - debug\n");
abort();
//#endif
#endif
return -2;
#else
dual(0, 7);
createRim(7 + 8 + 16 + 32, true, 0);
int factorizationStatus = internalFactorize(0);
assert (factorizationStatus == 0);
if (factorizationStatus)
abort();
#endif
}
// Start of fast iterations
factorization_->sparseThreshold(0);
factorization_->goSparse();
assert (!info->saveCosts_);
int numberTotal = numberRows_ + numberColumns_;
double * save = new double [4*numberTotal];
CoinMemcpyN(cost_, numberTotal, save+3*numberTotal);
if (perturbation_<100) {
int saveIterations = numberIterations_;
//int saveOptions = moreSpecialOptions_;
int savePerturbation = perturbation_;
numberIterations_ = 0;
//moreSpecialOptions_ |= 128;
bool allZero = true;
for (int i=0;i<numberColumns_;i++) {
if (cost_[i]) {
if (upper_[i]>lower_[i]) {
allZero=false;
break;
}
}
}
if (allZero)
perturbation_ = 58;
static_cast< ClpSimplexDual *>(this)->perturb();
numberIterations_ = saveIterations;
//moreSpecialOptions_ = saveOptions;
perturbation_ = savePerturbation;
}
info->saveCosts_ = save;
CoinMemcpyN(cost_, numberTotal, save);
return 0;
}
// Like Fast dual
int
ClpSimplex::fastDual2(ClpNodeStuff * info)
{
assert ((info->solverOptions_ & 65536) != 0);
int numberTotal = numberRows_ + numberColumns_;
assert (info->saveCosts_);
double * save = info->saveCosts_;
CoinMemcpyN(save, numberTotal, cost_);
save += numberTotal;
CoinMemcpyN(lower_, numberTotal, save);
save += numberTotal;
CoinMemcpyN(upper_, numberTotal, save);
double dummyChange;
(static_cast<ClpSimplexDual *>(this))->changeBounds(3, NULL, dummyChange);
numberPrimalInfeasibilities_ = 1;
sumPrimalInfeasibilities_ = 0.5;
sumOfRelaxedDualInfeasibilities_ = 0.0;
sumOfRelaxedPrimalInfeasibilities_ = 0.5;
checkDualSolution();
//if (xxxxxx)
//checkPrimalSolution(rowActivityWork_,columnActivityWork_);
assert((specialOptions_ & 16384) == 0);
specialOptions_ |= 524288; // say use solution
ClpObjective * saveObjective = objective_;
#ifndef NDEBUG
//(static_cast<ClpSimplexDual *>(this))->resetFakeBounds(-1);
#endif
//int saveNumberFake = numberFake_;
int status = static_cast<ClpSimplexDual *> (this)->fastDual(true);
//numberFake_ = saveNumberFake;
specialOptions_ &= ~524288; // say dont use solution
CoinAssert (problemStatus_ || objectiveValue_ < 1.0e50);
if (status && problemStatus_ != 3) {
// not finished - might be optimal
checkPrimalSolution(rowActivityWork_, columnActivityWork_);
double limit = 0.0;
getDblParam(ClpDualObjectiveLimit, limit);
if (!numberPrimalInfeasibilities_ && objectiveValue()*optimizationDirection_ < limit) {
problemStatus_ = 0;
}
}
if (problemStatus_ == 10) {
// Say second call
moreSpecialOptions_ |= 256;
//printf("Cleaning up with primal\n");
//lastAlgorithm=1;
int savePerturbation = perturbation_;
int saveLog = handler_->logLevel();
//handler_->setLogLevel(63);
perturbation_ = 100;
bool denseFactorization = initialDenseFactorization();
// It will be safe to allow dense
setInitialDenseFactorization(true);
// Allow for catastrophe
int saveMax = intParam_[ClpMaxNumIteration];
if (intParam_[ClpMaxNumIteration] > 100000 + numberIterations_)
intParam_[ClpMaxNumIteration] = numberIterations_ + 1000 + 2 * numberRows_ + numberColumns_;
// check which algorithms allowed
baseIteration_ = numberIterations_;
status = static_cast<ClpSimplexPrimal *> (this)->primal(1, 7);
baseIteration_ = 0;
if (saveObjective != objective_) {
// We changed objective to see if infeasible
delete objective_;
objective_ = saveObjective;
if (!problemStatus_) {
// carry on
status = static_cast<ClpSimplexPrimal *> (this)->primal(1, 7);
}
}
if (problemStatus_ == 3 && numberIterations_ < saveMax) {
#ifdef COIN_DEVELOP
if (handler_->logLevel() > 0)
printf("looks like trouble - too many iterations in clean up - trying again\n");
#endif
// flatten solution and try again
int iColumn;
for (iColumn = 0; iColumn < numberTotal; iColumn++) {
if (getStatus(iColumn) != basic) {
setStatus(iColumn, superBasic);
// but put to bound if close
if (fabs(solution_[iColumn] - lower_[iColumn])
<= primalTolerance_) {
solution_[iColumn] = lower_[iColumn];
setStatus(iColumn, atLowerBound);
} else if (fabs(solution_[iColumn]
- upper_[iColumn])
<= primalTolerance_) {
solution_[iColumn] = upper_[iColumn];
setStatus(iColumn, atUpperBound);
}
}
}
problemStatus_ = -1;
intParam_[ClpMaxNumIteration] = CoinMin(numberIterations_ + 1000 +
2 * numberRows_ + numberColumns_, saveMax);
perturbation_ = savePerturbation;
baseIteration_ = numberIterations_;
status = static_cast<ClpSimplexPrimal *> (this)->primal(0);
baseIteration_ = 0;
computeObjectiveValue();
// can't rely on djs either
memset(reducedCost_, 0, numberColumns_ * sizeof(double));
#ifdef COIN_DEVELOP
if (problemStatus_ == 3 && numberIterations_ < saveMax && handler_->logLevel() > 0)
printf("looks like real trouble - too many iterations in second clean up - giving up\n");
#endif
}
// Say not second call
moreSpecialOptions_ &= ~256;
intParam_[ClpMaxNumIteration] = saveMax;
setInitialDenseFactorization(denseFactorization);
perturbation_ = savePerturbation;
if (problemStatus_ == 10) {
if (!numberPrimalInfeasibilities_)
problemStatus_ = 0;
else
problemStatus_ = 4;
}
handler_->setLogLevel(saveLog);
// if done primal arrays may be rubbish
save = info->saveCosts_ + numberTotal;
CoinMemcpyN(save, numberTotal, lower_);
save += numberTotal;
CoinMemcpyN(save, numberTotal, upper_);
}
status = problemStatus_;
if (!problemStatus_) {
int j;
// Move solution to external array
if (!columnScale_) {
CoinMemcpyN(solution_, numberColumns_, columnActivity_);
} else {
for (j = 0; j < numberColumns_; j++)
columnActivity_[j] = solution_[j] * columnScale_[j];
}
if ((info->solverOptions_ & 1) != 0) {
// reduced costs
if (!columnScale_) {
CoinMemcpyN(dj_, numberColumns_, reducedCost_);
} else {
for (j = 0; j < numberColumns_; j++)
reducedCost_[j] = dj_[j] * columnScale_[j+numberColumns_];
}
}
if ((info->solverOptions_ & 2) != 0) {
// dual
if (!rowScale_) {
//CoinMemcpyN(dual_,numberRows_,dj_+numberColumns_);
} else {
for (j = 0; j < numberRows_; j++)
dual_[j] = dj_[j+numberColumns_] * rowScale_[j];
}
}
if ((info->solverOptions_ & 4) != 0) {
// row activity
if (!rowScale_) {
CoinMemcpyN(solution_ + numberColumns_, numberRows_, rowActivity_);
} else {
for (j = 0; j < numberRows_; j++)
rowActivity_[j] = solution_[j+numberColumns_] * rowScale_[j+numberRows_];
}
}
}
save = info->saveCosts_;
CoinMemcpyN(save, numberTotal, cost_);
#if 0
save += numberTotal;
CoinMemcpyN(save, numberTotal, lower_);
save += numberTotal;
CoinMemcpyN(save, numberTotal, upper_);
#endif
return status;
}
// Stop Fast dual
void
ClpSimplex::stopFastDual2(ClpNodeStuff * info)
{
delete [] info->saveCosts_;
info->saveCosts_ = NULL;
specialOptions_ = info->saveOptions_;
// try just factorization
if ((specialOptions_ & 65536) == 0)
factorization_->setPersistenceFlag(0);
deleteRim(1);
whatsChanged_ &= ~0xffff;
assert ((info->solverOptions_ & 65536) != 0);
info->solverOptions_ &= ~65536;
}
// Deals with crunch aspects
ClpSimplex *
ClpSimplex::fastCrunch(ClpNodeStuff * info, int mode)
{
ClpSimplex * small = NULL;
if (mode == 0) {
// before crunch
// crunch down
// Use dual region
double * rhs = dual_;
int * whichRow = new int[3*numberRows_];
int * whichColumn = new int[2*numberColumns_];
int nBound;
bool tightenBounds = ((specialOptions_ & 64) == 0) ? false : true;
small =
static_cast<ClpSimplexOther *> (this)->crunch(rhs, whichRow, whichColumn,
nBound, false, tightenBounds);
if (small) {
info->large_ = this;
info->whichRow_ = whichRow;
info->whichColumn_ = whichColumn;
info->nBound_ = nBound;
if (info->upPseudo_) {
const char * integerType2 = small->integerInformation();
int n = small->numberColumns();
int k = 0;
int jColumn = 0;
int j = 0;
for (int i = 0; i < n; i++) {
if (integerType2[i]) {
int iColumn = whichColumn[i];
// find
while (jColumn != iColumn) {
if (integerType_[jColumn])
j++;
jColumn++;
}
info->upPseudo_[k] = info->upPseudo_[j];
info->numberUp_[k] = info->numberUp_[j];
info->numberUpInfeasible_[k] = info->numberUpInfeasible_[j];
info->downPseudo_[k] = info->downPseudo_[j];
info->numberDown_[k] = info->numberDown_[j];
info->numberDownInfeasible_[k] = info->numberDownInfeasible_[j];
assert (info->upPseudo_[k] > 1.0e-40 && info->upPseudo_[k] < 1.0e40);
assert (info->downPseudo_[k] > 1.0e-40 && info->downPseudo_[k] < 1.0e40);
k++;
}
}
}
} else {
delete [] whichRow;
delete [] whichColumn;
}
} else {
// after crunch
if (mode == 1) {
// has solution
ClpSimplex * other = info->large_;
assert (other != this);
static_cast<ClpSimplexOther *> (other)->afterCrunch(*this,
info->whichRow_,
info->whichColumn_, info->nBound_);
for (int i = 0; i < other->numberColumns_; i++) {
if (other->integerType_[i]) {
double value = other->columnActivity_[i];
double value2 = floor(value + 0.5);
assert (fabs(value - value2) < 1.0e-4);
other->columnActivity_[i] = value2;
other->columnLower_[i] = value2;
other->columnUpper_[i] = value2;
}
}
}
delete [] info->whichRow_;
delete [] info->whichColumn_;
}
return small;
}
// Resizes rim part of model
void
ClpSimplex::resize (int newNumberRows, int newNumberColumns)
{
ClpModel::resize(newNumberRows, newNumberColumns);
delete [] perturbationArray_;
perturbationArray_ = NULL;
maximumPerturbationSize_=0;
if (saveStatus_) {
// delete arrays
int saveOptions = specialOptions_;
specialOptions_ = 0;
gutsOfDelete(2);
specialOptions_ = saveOptions;
}
}
// Return true if the objective limit test can be relied upon
bool
ClpSimplex::isObjectiveLimitTestValid() const
{
if (problemStatus_ == 0) {
return true;
} else if (problemStatus_ == 1) {
// ok if dual
return (algorithm_ < 0);
} else if (problemStatus_ == 2) {
// ok if primal
return (algorithm_ > 0);
} else {
return false;
}
}
// Create C++ lines to get to current state
void
ClpSimplex::generateCpp( FILE * fp, bool defaultFactor)
{
ClpModel::generateCpp(fp);
ClpSimplex defaultModel;
ClpSimplex * other = &defaultModel;
int iValue1, iValue2;
double dValue1, dValue2;
// Stuff that can't be done easily
if (factorizationFrequency() == other->factorizationFrequency()) {
if (defaultFactor) {
fprintf(fp, "3 // For branchAndBound this may help\n");
fprintf(fp, "3 clpModel->defaultFactorizationFrequency();\n");
} else {
// tell user about default
fprintf(fp, "3 // For initialSolve you don't need below but ...\n");
fprintf(fp, "3 // clpModel->defaultFactorizationFrequency();\n");
}
}
iValue1 = this->factorizationFrequency();
iValue2 = other->factorizationFrequency();
fprintf(fp, "%d int save_factorizationFrequency = clpModel->factorizationFrequency();\n", iValue1 == iValue2 ? 2 : 1);
fprintf(fp, "%d clpModel->setFactorizationFrequency(%d);\n", iValue1 == iValue2 ? 4 : 3, iValue1);
fprintf(fp, "%d clpModel->setFactorizationFrequency(save_factorizationFrequency);\n", iValue1 == iValue2 ? 7 : 6);
dValue1 = this->dualBound();
dValue2 = other->dualBound();
fprintf(fp, "%d double save_dualBound = clpModel->dualBound();\n", dValue1 == dValue2 ? 2 : 1);
fprintf(fp, "%d clpModel->setDualBound(%g);\n", dValue1 == dValue2 ? 4 : 3, dValue1);
fprintf(fp, "%d clpModel->setDualBound(save_dualBound);\n", dValue1 == dValue2 ? 7 : 6);
dValue1 = this->infeasibilityCost();
dValue2 = other->infeasibilityCost();
fprintf(fp, "%d double save_infeasibilityCost = clpModel->infeasibilityCost();\n", dValue1 == dValue2 ? 2 : 1);
fprintf(fp, "%d clpModel->setInfeasibilityCost(%g);\n", dValue1 == dValue2 ? 4 : 3, dValue1);
fprintf(fp, "%d clpModel->setInfeasibilityCost(save_infeasibilityCost);\n", dValue1 == dValue2 ? 7 : 6);
iValue1 = this->perturbation();
iValue2 = other->perturbation();
fprintf(fp, "%d int save_perturbation = clpModel->perturbation();\n", iValue1 == iValue2 ? 2 : 1);
fprintf(fp, "%d clpModel->setPerturbation(%d);\n", iValue1 == iValue2 ? 4 : 3, iValue1);
fprintf(fp, "%d clpModel->setPerturbation(save_perturbation);\n", iValue1 == iValue2 ? 7 : 6);
}
// Copy across enabled stuff from one solver to another
void
ClpSimplex::copyEnabledStuff(const ClpSimplex * rhs)
{
solveType_=rhs->solveType_;
if (rhs->solution_) {
int numberTotal = numberRows_+numberColumns_;
assert (!solution_);
solution_ = CoinCopyOfArray(rhs->solution_,numberTotal);
lower_ = CoinCopyOfArray(rhs->lower_,numberTotal);
upper_ = CoinCopyOfArray(rhs->upper_,numberTotal);
dj_ = CoinCopyOfArray(rhs->dj_,numberTotal);
cost_ = CoinCopyOfArray(rhs->cost_,2*numberTotal);
reducedCostWork_ = dj_;
rowReducedCost_ = dj_ + numberColumns_;
columnActivityWork_ = solution_;
rowActivityWork_ = solution_ + numberColumns_;
objectiveWork_ = cost_;
rowObjectiveWork_ = cost_ + numberColumns_;
rowLowerWork_ = lower_ + numberColumns_;
columnLowerWork_ = lower_;
rowUpperWork_ = upper_ + numberColumns_;
columnUpperWork_ = upper_;
}
if (rhs->factorization_) {
delete factorization_;
factorization_ = new ClpFactorization(*rhs->factorization_);
delete [] pivotVariable_;
pivotVariable_ = CoinCopyOfArray(rhs->pivotVariable_,numberRows_);
}
for (int i = 0; i < 6; i++) {
if (rhs->rowArray_[i])
rowArray_[i] = new CoinIndexedVector(*rhs->rowArray_[i]);
if (rhs->columnArray_[i])
columnArray_[i] = new CoinIndexedVector(*rhs->columnArray_[i]);
}
if (rhs->nonLinearCost_)
nonLinearCost_=new ClpNonLinearCost(*rhs->nonLinearCost_);
if (rhs->dualRowPivot_)
dualRowPivot_ = rhs->dualRowPivot_->clone();
if (rhs->primalColumnPivot_)
primalColumnPivot_ = rhs->primalColumnPivot_->clone();
}