/Users/kmartin/Documents/files/code/cpp/OScpp/COIN-OS/OS/src/OSCommonInterfaces/OSInstance.cpp

Go to the documentation of this file.
00001 
00018 #include "OSInstance.h"
00019 #include "OSMathUtil.h"
00020 #include "OSCommonUtil.h"
00021 #include "OSErrorClass.h"
00022 #include "OSParameters.h"
00023 #include<iostream>  
00024 #include<sstream>
00025 
00026 //#define DEBUG
00027  
00028 using namespace std;
00029 using CppAD::NearEqual;
00030 
00031 OSInstance::OSInstance(): 
00032         m_sInstanceName(""),
00033         m_sInstanceSource(""),  
00034         m_sInstanceDescription(""),
00035         m_bProcessVariables(false),
00036         m_iVariableNumber(-1),
00037         m_iNumberOfIntegerVariables( 0),
00038         m_iNumberOfBinaryVariables( 0),
00039         m_iNumberOfQuadraticRowIndexes( 0),
00040         m_bQuadraticRowIndexesProcessed( false),
00041         m_miQuadRowIndexes( NULL),
00042         m_iNumberOfNonlinearExpressionTreeIndexes( 0),
00043         m_bNonlinearExpressionTreeIndexesProcessed( false),
00044         m_miNonlinearExpressionTreeIndexes( NULL),
00045         m_iNumberOfNonlinearExpressionTreeModIndexes( 0),
00046         m_bNonlinearExpressionTreeModIndexesProcessed( false),
00047         m_miNonlinearExpressionTreeModIndexes( NULL),           
00048         m_msVariableNames(NULL),
00049         m_mdVariableInitialValues(NULL),
00050         m_msVariableInitialStringValues(NULL),
00051         m_mcVariableTypes(NULL),
00052         m_mdVariableLowerBounds(NULL),
00053         m_mdVariableUpperBounds(NULL),
00054         m_bProcessObjectives(false),
00055         m_iObjectiveNumber(-1),
00056         m_iObjectiveNumberNonlinear( 0),
00057         m_msObjectiveNames(NULL),
00058         m_msMaxOrMins(NULL),
00059         m_miNumberOfObjCoef(NULL),
00060         m_mdObjectiveConstants(NULL),
00061         m_mdObjectiveWeights(NULL),
00062         m_mObjectiveCoefficients(NULL),
00063         m_bGetDenseObjectives(false),
00064         m_mmdDenseObjectiveCoefficients(NULL),
00065         m_bProcessConstraints(false),
00066         m_iConstraintNumber(-1),
00067         m_iConstraintNumberNonlinear( 0),       
00068         m_msConstraintNames(NULL),
00069         m_mdConstraintLowerBounds(NULL),
00070         m_mdConstraintUpperBounds(NULL),
00071         m_mdConstraintConstants( NULL),
00072         m_mcConstraintTypes(NULL),
00073         m_bProcessLinearConstraintCoefficients(false),  
00074         m_iLinearConstraintCoefficientNumber(-1),
00075         m_bColumnMajor(true),
00076         m_binitForAlgDiff( false),      
00077         m_linearConstraintCoefficientsInColumnMajor(NULL),
00078         m_linearConstraintCoefficientsInRowMajor(NULL), 
00079         m_bProcessQuadraticTerms(false),
00080         m_iQuadraticTermNumber(-1),
00081         m_mdConstraintFunctionValues( NULL),
00082         m_mdObjectiveFunctionValues( NULL),
00083         m_iJacValueSize( 0),
00084         m_miJacStart( NULL),
00085         m_miJacIndex( NULL),
00086         m_mdJacValue( NULL),
00087         m_miJacNumConTerms( NULL),
00088         m_sparseJacMatrix( NULL),       
00089         m_iHighestTaylorCoeffOrder(-1), 
00090         m_quadraticTerms( NULL),        
00091         m_bQTermsAdded( false), 
00092         m_iNumberOfNonlinearVariables( 0),
00093         m_bProcessNonlinearExpressions( false),
00094         m_iNonlinearExpressionNumber( -1),              
00095         m_miNonlinearExpressionIndexes( NULL),
00096         m_bProcessExpressionTrees( false),
00097         m_bProcessExpressionTreesMod( false),
00098         m_LagrangianExpTree(NULL),
00099         m_bLagrangianExpTreeCreated( false),
00100         m_LagrangianSparseHessian( NULL),
00101         m_bLagrangianSparseHessianCreated( false),
00102         m_miNonLinearVarsReverseMap( NULL),
00103         m_bAllNonlinearVariablesIndex( false),
00104         m_bCppADFunIsCreated( false),
00105         m_bCppADTapesBuilt( false),
00106         m_bCppADMustReTape( false),
00107         m_bDuplicateExpressionTreesMap( false),
00108         m_bNonLinearStructuresInitialized( false),
00109         m_bSparseJacobianCalculated( false),
00110         m_iHighestOrderEvaluated( -1),
00111         m_mmdObjGradient( NULL),
00112         m_bProcessTimeDomain( false),
00113         m_bProcessTimeStages( false),
00114         m_bProcessTimeInterval( false),
00115         m_bFiniteTimeStages( false),
00116         m_iNumberOfTimeStages(-1),
00117         bUseExpTreeForFunEval( false)
00118 
00119 {    
00120         #ifdef DEBUG
00121         cout << "Inside OSInstance Constructor" << endl;
00122         #endif
00123         this->instanceHeader = new InstanceHeader();
00124         this->instanceData = new InstanceData();
00125 }  
00126 
00127 OSInstance::~OSInstance(){
00128         #ifdef DEBUG
00129         cout << "OSInstance Destructor Called" << endl;
00130         #endif
00131         std::map<int, OSExpressionTree*>::iterator posMapExpTree;
00132         // delete  the temporary arrays
00133         delete[] m_msVariableNames;
00134         m_msVariableNames = NULL;
00135         delete[] m_mdVariableInitialValues;
00136         m_mdVariableInitialValues = NULL ;
00137         delete[] m_msVariableInitialStringValues;
00138         m_msVariableInitialStringValues = NULL;
00139         delete[] m_mcVariableTypes;
00140         m_mcVariableTypes = NULL;
00153         delete[] m_miNonLinearVarsReverseMap;
00154         m_miNonLinearVarsReverseMap = NULL;
00155         int i;
00156         //if(instanceData->objectives->numberOfObjectives > 0 && m_mObjectiveCoefficients != NULL){
00157         if(m_bProcessObjectives == true ){
00158                 for(i = 0; i < instanceData->objectives->numberOfObjectives; i++){
00159                         #ifdef DEBUG
00160                         std::cout <<  "Delete m_mObjectiveCoefficients[i]" << std::endl;
00161                         #endif
00162                         delete m_mObjectiveCoefficients[i];
00163                         m_mObjectiveCoefficients[i] = NULL;
00164                 }
00165                 #ifdef DEBUG
00166                 std::cout <<  "Delete m_msObjectiveNames" << std::endl;
00167                 std::cout <<  "Delete m_msMaxOrMins" << std::endl;
00168                 std::cout <<  "Delete m_miNumberOfObjCoef" << std::endl;
00169                 std::cout <<  "Delete m_mdObjectiveConstants" << std::endl;
00170                 std::cout <<  "Delete m_mdObjectiveWeights" << std::endl;
00171                 #endif          
00172                 delete[] m_msObjectiveNames;
00173                 m_msObjectiveNames = NULL;
00174                 delete[] m_msMaxOrMins;
00175                 m_msMaxOrMins = NULL;
00176                 delete[] m_miNumberOfObjCoef;
00177                 m_miNumberOfObjCoef = NULL;
00178                 delete[] m_mdObjectiveConstants; 
00179                 m_mdObjectiveConstants = NULL;
00180                 delete[] m_mdObjectiveWeights;
00181                 m_mdObjectiveWeights = NULL;
00182                 delete[] m_mObjectiveCoefficients;
00183                 m_mObjectiveCoefficients = NULL;
00184         }
00185         if(instanceData->objectives->numberOfObjectives > 0 && m_mmdDenseObjectiveCoefficients != NULL){
00186                 for(i = 0; i < instanceData->objectives->numberOfObjectives; i++){
00187                         //delete m_mmdDenseObjectiveCoefficients[i];
00188                         #ifdef DEBUG
00189                         std::cout <<  "delete m_mmdDenseObjectiveCoefficients[i]" << std::endl;
00190                         #endif
00191                     delete[] m_mmdDenseObjectiveCoefficients[i];
00192                         m_mmdDenseObjectiveCoefficients[i] = NULL;
00193                 }
00194                 delete[] m_mmdDenseObjectiveCoefficients;
00195                 m_mmdDenseObjectiveCoefficients = NULL;
00196         }
00197 
00198         if( (m_binitForAlgDiff == true)  ){     
00199                 if(instanceData->objectives->numberOfObjectives > 0 && m_mmdObjGradient != NULL){
00200                         
00201                         #ifdef DEBUG
00202                         std::cout <<  "The number of objectives =  " << instanceData->objectives->numberOfObjectives << std::endl;
00203                         #endif
00204                         for(i = 0; i < instanceData->objectives->numberOfObjectives; i++){
00205                                 #ifdef DEBUG
00206                                 std::cout << "deleting Objective function gradient " << i << std::endl;
00207                                 #endif
00208                                 delete[] m_mmdObjGradient[i];
00209 
00210                                 m_mmdObjGradient[i] = NULL;
00211                         }
00212                         delete[] m_mmdObjGradient;
00213                         m_mmdObjGradient = NULL;
00214                 }
00215         }
00216 
00217         if(m_bProcessLinearConstraintCoefficients == true && m_bColumnMajor == true) delete m_linearConstraintCoefficientsInColumnMajor;
00218         if(m_bProcessLinearConstraintCoefficients == true && m_bColumnMajor == false) delete m_linearConstraintCoefficientsInRowMajor;
00219         //if(m_linearConstraintCoefficientsInRowMajor != NULL) delete m_linearConstraintCoefficientsInRowMajor;
00220         //if(m_linearConstraintCoefficientsInColumnMajor != NULL) delete m_linearConstraintCoefficientsInColumnMajor;
00221         delete[] m_msConstraintNames;
00222         m_msConstraintNames = NULL;
00223         delete[] m_mcConstraintTypes;
00224         m_mcConstraintTypes = NULL;
00225         delete[]  m_mdConstraintConstants;
00226          m_mdConstraintConstants = NULL;
00227         delete[] m_mdConstraintLowerBounds;
00228         m_mdConstraintLowerBounds = NULL;
00229         delete[] m_mdConstraintUpperBounds;
00230         m_mdConstraintUpperBounds = NULL;
00231         delete[] m_mdVariableLowerBounds;
00232         m_mdVariableLowerBounds = NULL;
00233         delete[] m_mdVariableUpperBounds;
00234         m_mdVariableUpperBounds = NULL;
00235         //std::cout << "Do garbage collection for the nonlinear API" << std::endl;
00236         // garbage collection for the gradient
00237         if(m_bNonLinearStructuresInitialized == true ){
00238                 delete[] m_mdObjectiveFunctionValues;
00239                 m_mdObjectiveFunctionValues = NULL;     
00240                 delete[] m_mdConstraintFunctionValues;
00241                 m_mdConstraintFunctionValues = NULL;
00242         }
00243         if(m_bSparseJacobianCalculated == true){
00244                 delete[] m_miJacStart;
00245                 m_miJacStart = NULL;
00246                 delete[] m_miJacIndex;
00247                 m_miJacIndex = NULL;
00248                 delete[] m_mdJacValue;
00249                 m_mdJacValue = NULL;
00250                 delete[] m_miJacNumConTerms;
00251                 m_miJacNumConTerms = NULL;
00252         }
00253         if( m_bLagrangianExpTreeCreated == true){
00254                 delete m_LagrangianExpTree;
00255                 m_LagrangianExpTree = NULL;
00256         }
00257         if( m_bLagrangianSparseHessianCreated == true){
00258                 delete m_LagrangianSparseHessian;
00259                 m_LagrangianSparseHessian = NULL;
00260         }
00261         if( m_bSparseJacobianCalculated == true){
00262                 delete m_sparseJacMatrix;
00263                 m_sparseJacMatrix = NULL;
00264         }
00265         if( (instanceData->quadraticCoefficients->qTerm != NULL) && (m_bProcessQuadraticTerms == true) ){
00266                 delete m_quadraticTerms;
00267                 m_quadraticTerms = NULL;
00268         }
00269         if( (instanceData->quadraticCoefficients->qTerm != NULL)  && (m_bQuadraticRowIndexesProcessed == true) ){
00270                 delete[] m_miQuadRowIndexes;
00271                 m_miQuadRowIndexes = NULL;
00272         }
00273         //
00274         // delete the new expression trees that got created
00275         //if( m_bLagrangianExpTreeCreated == false  ||  m_bLagrangianExpTreeCreated == true){
00276         if( (m_bProcessExpressionTrees == true) && (m_bDuplicateExpressionTreesMap == false)  ) {
00277                 for(posMapExpTree = m_mapExpressionTrees.begin(); posMapExpTree != m_mapExpressionTrees.end(); ++posMapExpTree){
00278                         std::cout << "Deleting an expression tree from the map for row  " << posMapExpTree->first  << std::endl;
00279                         delete m_mapExpressionTrees[ posMapExpTree->first ];
00280                 }
00281         }
00282         if( m_bDuplicateExpressionTreesMap == true)   {
00283                 for(posMapExpTree = m_mapExpressionTreesMod.begin(); posMapExpTree != m_mapExpressionTreesMod.end(); ++posMapExpTree){          
00284                         #ifdef DEBUG
00285                                 std::cout << "Deleting an expression tree from m_mapExpressionTreesMod" << std::endl;
00286                         #endif
00287                         delete m_mapExpressionTreesMod[ posMapExpTree->first ];
00288                 }
00289         }
00290         //}
00292         if( (m_bNonlinearExpressionTreeIndexesProcessed == true) && (m_mapExpressionTrees.size() > 0) ){
00293                 std::cout << "Deleting  m_miNonlinearExpressionTreeIndexes" << std::endl;
00294                 delete[] m_miNonlinearExpressionTreeIndexes;
00295                 std::cout << "Done Deleting  m_miNonlinearExpressionTreeIndexes" << std::endl;
00296                 m_miNonlinearExpressionTreeIndexes = NULL;
00297         }
00298         if( (m_bNonlinearExpressionTreeModIndexesProcessed == true) && (m_mapExpressionTreesMod.size() > 0) ){
00299                 std::cout << "Deleting  m_miNonlinearExpressionTreeModIndexes" << std::endl;
00300                 delete[] m_miNonlinearExpressionTreeModIndexes;
00301                 std::cout << "Done Deleting  m_miNonlinearExpressionTreeModIndexes" << std::endl;
00302                 m_miNonlinearExpressionTreeModIndexes = NULL;
00303         }
00304         if(m_bCppADFunIsCreated == true){
00305                 delete Fad;
00306                 Fad = NULL;
00307         }
00308 //      if( (instanceData->timeDomain->stages->stage != NULL) && (m_bProcessTimeStages == true) ){
00309 //              delete m_Stages;
00310 //              m_Stages = NULL;
00311 //      }
00312         
00313         // delete the two children of OSInstance
00314         //delete instanceHeader object
00315         delete instanceHeader;
00316         instanceHeader = NULL;
00317         //delete instanceData object
00318         delete instanceData;
00319         instanceData = NULL;
00320 }//OSInstance Destructor
00321 
00322 InstanceHeader::InstanceHeader():
00323         description(""),
00324         name(""),
00325         source("")
00326 
00327 { 
00328         #ifdef DEBUG
00329         cout << "Inside the InstanceHeader Constructor" << endl;
00330         #endif
00331 } 
00332 
00333 
00334 InstanceHeader::~InstanceHeader(){
00335         #ifdef DEBUG  
00336         cout << "Inside the InstanceHeader Destructor" << endl;
00337         #endif
00338 } 
00339 
00340 Variable::Variable():
00341         lb(0.0),
00342         ub(OSDBL_MAX),
00343         init(OSNAN), 
00344         type('C'), 
00345         name(""),
00346         initString("")
00347 {  
00348         #ifdef DEBUG
00349         cout << "Inside the Variable Constructor" << endl;
00350         #endif 
00351 } 
00352 
00353 Variable::~Variable(){  
00354         #ifdef DEBUG
00355         cout << "Inside the Variable Destructor" << endl; 
00356         #endif
00357 } 
00358 
00359 Variables::Variables(){  
00360         #ifdef DEBUG
00361         cout << "Inside the Variables Constructor" << endl; 
00362         #endif 
00363         numberOfVariables = 0;
00364         var = NULL; 
00365 }
00366 
00367 Variables::~Variables(){ 
00368         #ifdef DEBUG 
00369         cout << "Inside the Variables Destructor" << endl;
00370         #endif
00371         int i;
00372         if(numberOfVariables > 0 && var != NULL){
00373                 for(i = 0; i < numberOfVariables; i++){
00374                         #ifdef DEBUG 
00375                         cout << "Deleting var[ i]" << endl;
00376                         #endif
00377                         delete var[i];
00378                         var[i] = NULL;
00379                 }
00380         }
00381         delete[] var;
00382         var = NULL; 
00383 }  
00384 
00385 ObjCoef::ObjCoef():
00386         idx(-1),
00387         value(0.0)  
00388 {  
00389         #ifdef DEBUG
00390         cout << "Inside the Coef Constructor" << endl;
00391         #endif 
00392 }
00393 
00394 ObjCoef::~ObjCoef(){ 
00395         #ifdef DEBUG
00396         cout << "Inside the ObjCoef Destructor" << endl;  
00397         #endif
00398 }
00399 
00400 Objective::Objective():
00401         name("") ,
00402         maxOrMin("min"),
00403         constant(0.0),
00404         weight(1.0),
00405         numberOfObjCoef(0),
00406         coef(NULL)
00407 { 
00408  
00409         #ifdef DEBUG
00410         cout << "Inside the Objective Constructor" << endl;
00411         #endif
00412 }
00413 
00414 Objective::~Objective(){
00415         #ifdef DEBUG  
00416         cout << "Inside the Objective Destructor" << endl;
00417         #endif
00418         int i;
00419         if(numberOfObjCoef > 0 && coef != NULL){
00420                 for(i = 0; i < numberOfObjCoef; i++){
00421                         delete coef[i];
00422                         coef[i] = NULL;
00423                 }
00424         }
00425         delete[] coef;
00426         coef = NULL;
00427 }  
00428 
00429 Objectives::Objectives()
00430 {  
00431         #ifdef DEBUG
00432         cout << "Inside the Objectives Constructor" << endl; 
00433         #endif
00434         numberOfObjectives = 0;
00435         obj = NULL;
00436 } 
00437 
00438 Objectives::~Objectives(){ 
00439         #ifdef DEBUG 
00440         cout << "Inside the Objectives Destructor" << endl;
00441         #endif
00442         int i;
00443         if(numberOfObjectives > 0 && obj != NULL){
00444                 for(i = 0; i < numberOfObjectives; i++){
00445                         delete obj[i];
00446                         obj[i] = NULL;
00447                 }
00448         }
00449         delete[] obj;
00450         obj = NULL;
00451 }
00452 
00453 Constraint::Constraint():
00454         name(""),
00455         constant(0.0),
00456         lb(-OSDBL_MAX),
00457         ub(OSDBL_MAX)
00458 
00459 {
00460         #ifdef DEBUG  
00461         cout << "Inside the Constraint Constructor" << endl;
00462         #endif
00463 } 
00464 
00465 Constraint::~Constraint(){  
00466         #ifdef DEBUG
00467         cout << "Inside the Constraint Destructor" << endl;
00468         #endif
00469 } 
00470 
00471 Constraints::Constraints():
00472         numberOfConstraints(0),
00473         con(NULL)
00474 {
00475         #ifdef DEBUG
00476         cout << "Inside the Constraints Constructor" << endl;
00477         #endif
00478 } 
00479 
00480 Constraints::~Constraints(){  
00481         #ifdef DEBUG
00482         cout << "Inside the Constraints Destructor" << endl;
00483         #endif
00484         int i;
00485         if(numberOfConstraints > 0 && con != NULL){
00486                 for( i = 0; i < numberOfConstraints; i++){
00487                         delete con[i];
00488                         con[i] = NULL;
00489                 }
00490         }
00491         delete[] con;
00492         con = NULL;
00493 } 
00494 
00495 
00496 
00497 LinearConstraintCoefficients::LinearConstraintCoefficients():
00498         numberOfValues(0) ,
00499         iNumberOfStartElements( 0)
00500 { 
00501         #ifdef DEBUG 
00502         cout << "Inside the LinearConstraintCoefficients Constructor" << endl; 
00503         #endif
00504         start = new IntVector();
00505         rowIdx = new IntVector();
00506         colIdx = new IntVector();
00507         value = new DoubleVector();
00508 
00509 } 
00510 
00511 
00512 LinearConstraintCoefficients::~LinearConstraintCoefficients(){  
00513         #ifdef DEBUG
00514         cout << "Inside the LinearConstraintCoefficients Destructor" << endl; 
00515         #endif
00516         delete start;
00517         start = NULL;
00518         delete rowIdx;
00519         rowIdx = NULL;
00520         delete colIdx;
00521         colIdx = NULL;
00522         delete value;
00523         value = NULL;
00524 }
00525 
00526 QuadraticTerm::QuadraticTerm():
00527 
00528         idx(0),   
00529         idxOne(-1),
00530         idxTwo(-1),
00531         coef(0.0)
00532 
00533 {
00534         #ifdef DEBUG  
00535         cout << "Inside the QuadraticTerm Constructor" << endl;
00536         #endif
00537 } 
00538 
00539 
00540 QuadraticTerm::~QuadraticTerm(){  
00541         #ifdef DEBUG
00542         cout << "Inside the QuadraticTerm Destructor" << endl;
00543         #endif
00544 }
00545 
00546 
00547 
00548 QuadraticCoefficients::QuadraticCoefficients():
00549         numberOfQuadraticTerms(0),
00550         qTerm(NULL)
00551 { 
00552         #ifdef DEBUG 
00553         cout << "Inside the QuadraticCoefficients Constructor" << endl;
00554         #endif
00555 }//end QuadraticCoefficients() 
00556 
00557 
00558 QuadraticCoefficients::~QuadraticCoefficients(){
00559         #ifdef DEBUG  
00560         cout << "Inside the QuadraticCoefficients Destructor" << endl;
00561         #endif
00562         int i;
00563         if(numberOfQuadraticTerms > 0 && qTerm != NULL){
00564                 for( i = 0; i < numberOfQuadraticTerms; i++){
00565                         delete qTerm[i];
00566                         qTerm[i] = NULL;
00567                 }
00568         }
00569         delete[] qTerm;
00570         qTerm = NULL;  
00571 }//end ~QuadraticCoefficients()  
00572 
00573 
00574 Nl::Nl(){
00575         idx = 0;
00576         osExpressionTree = NULL;
00577         m_bDeleteExpressionTree = true;
00578 }//end Nl
00579  
00580  
00581 Nl::~Nl(){
00582         #ifdef DEBUG  
00583         cout << "Inside the Nl Destructor" << endl;
00584         #endif
00585         // don't delete the expression tree if we created a map of the expression
00586         // trees, otherwise we would destroy twice
00587         if( m_bDeleteExpressionTree == true){
00588                 delete osExpressionTree;
00589                 osExpressionTree = NULL;
00590         }
00591         
00592 }//end ~Nl
00593 
00594 
00595 
00596 NonlinearExpressions::NonlinearExpressions():
00597         numberOfNonlinearExpressions(0) ,
00598         nl(NULL)
00599 { 
00600         #ifdef DEBUG 
00601         cout << "Inside the NonlinearExpressions Constructor" << endl;
00602         #endif
00603 }//end NonlinearExpressions() 
00604 
00605 NonlinearExpressions::~NonlinearExpressions(){
00606         #ifdef DEBUG  
00607         cout << "Inside the NonlinearExpressions Destructor" << endl;
00608         cout << "NUMBER OF NONLINEAR EXPRESSIONS = " << numberOfNonlinearExpressions << endl;
00609         #endif
00610         int i;
00611         if(numberOfNonlinearExpressions > 0 && nl != NULL){
00612                 for( i = 0; i < numberOfNonlinearExpressions; i++){
00613                         #ifdef DEBUG  
00614                                 cout << "DESTROYING EXPRESSION " << nl[ i]->idx << endl;
00615                         #endif
00616                         delete nl[i];
00617                         nl[i] = NULL;
00618                 }
00619         }
00620         delete[] nl;
00621         nl = NULL;  
00622 }//end ~NonlinearExpressions()  
00623 
00624 
00625 TimeDomainStage::TimeDomainStage():
00626         name(""),
00627         nvar(0),
00628         ncon(0),
00629         nobj(0)
00630 { 
00631         #ifdef DEBUG 
00632         cout << "Inside the Stage Constructor" << endl;
00633         #endif
00634         variables   = NULL;
00635         constraints = NULL;
00636         objectives  = NULL;
00637 }//end TimeDomainStage() 
00638 
00639 
00640 TimeDomainStage::~TimeDomainStage(){
00641         #ifdef DEBUG  
00642         cout << "Inside the Stage Destructor" << endl;
00643         #endif
00644         if (variables != NULL)
00645         {       delete [] variables;
00646                 variables = NULL;
00647         }
00648         if (constraints != NULL)
00649         {       delete []  constraints;
00650                 constraints = NULL;
00651         }
00652         if (objectives != NULL)
00653         {       delete [] objectives;
00654                 objectives = NULL;
00655         }
00656 }//end ~TimeDomainStage()  
00657 
00658 
00659 TimeDomainStages::TimeDomainStages():
00660         numberOfStages(0),
00661         stage(NULL)
00662 {
00663         #ifdef DEBUG  
00664         cout << "Inside the Stages Constructor" << endl;
00665         #endif
00666 } 
00667 
00668 
00669 TimeDomainStages::~TimeDomainStages(){  
00670         #ifdef DEBUG
00671         cout << "Inside the Stages Destructor" << endl;
00672         #endif
00673         int i;
00674         if(numberOfStages > 0 && stage != NULL){
00675                 for( i = 0; i < numberOfStages; i++){
00676                         delete stage[i];
00677                         stage[i] = NULL;
00678                 }
00679         }
00680         delete[] stage;
00681         stage = NULL;  
00682 }
00683 
00684 TimeDomainInterval::TimeDomainInterval():
00685         intervalHorizon(0.0),
00686         intervalStart(0.0)
00687 {
00688         #ifdef DEBUG  
00689         cout << "Inside the Interval Constructor" << endl;
00690         #endif
00691 } 
00692 
00693 
00694 TimeDomainInterval::~TimeDomainInterval(){  
00695         #ifdef DEBUG
00696         cout << "Inside the Interval Destructor" << endl;
00697         #endif
00698 }
00699 
00700 TimeDomain::TimeDomain()
00701 {
00702         #ifdef DEBUG
00703         cout << "Inside the TimeDomain Constructor" << endl;
00704         #endif
00705         stages = NULL;
00706         interval = NULL;
00707 }
00708 
00709 TimeDomain::~TimeDomain()
00710 {  
00711         #ifdef DEBUG
00712         cout << "Inside the TimeDomain Destructor" << endl;
00713         #endif
00714         if (stages != NULL)
00715         {       delete stages;
00716                 stages = NULL;
00717         };
00718         if (interval != NULL)
00719         {       delete interval;
00720                 interval = NULL;
00721         };
00722 } 
00723 
00724 
00725 InstanceData::InstanceData(){ 
00726         #ifdef DEBUG 
00727         cout << "Inside the InstanceData Constructor" << endl;
00728         #endif 
00729         variables = new Variables();
00730         objectives = new Objectives();
00731         constraints = new Constraints();
00732         linearConstraintCoefficients = new LinearConstraintCoefficients();
00733         quadraticCoefficients = new QuadraticCoefficients();
00734         nonlinearExpressions = new NonlinearExpressions();
00735         timeDomain = NULL;
00736 } 
00737 
00738 InstanceData::~InstanceData(){  
00739         #ifdef DEBUG
00740         cout << "Inside the InstanceData Destructor" << endl; 
00741         #endif
00742         delete variables;
00743         variables = NULL;
00744         delete objectives;
00745         objectives = NULL;
00746         delete constraints;
00747         constraints = NULL;
00748         delete linearConstraintCoefficients;
00749         linearConstraintCoefficients = NULL;
00750         delete quadraticCoefficients;
00751         quadraticCoefficients = NULL;
00752         delete nonlinearExpressions;
00753         nonlinearExpressions = NULL;
00754         delete timeDomain;
00755         if (timeDomain != NULL)
00756         {   delete timeDomain;
00757                 timeDomain = NULL;
00758         };
00759 } 
00760 
00761 string OSInstance::getInstanceName(){
00762         if(  m_sInstanceName.length() <= 0){
00763                 m_sInstanceName = instanceHeader->name;
00764         }
00765         return m_sInstanceName;
00766 }//getInstanceName
00767 
00768 
00769 string OSInstance::getInstanceSource(){
00770         if( m_sInstanceSource.length() <= 0){
00771                 m_sInstanceSource = instanceHeader->source;
00772         }
00773         return m_sInstanceSource;
00774 }//getInstanceSource
00775 
00776 string OSInstance::getInstanceDescription(){
00777         if(m_sInstanceDescription.length() <= 0){
00778                 m_sInstanceDescription = instanceHeader->description;
00779         }
00780         return m_sInstanceDescription;
00781 }//getInstanceDescription
00782 
00783 int OSInstance::getVariableNumber(){
00784         if(m_iVariableNumber == -1){
00785                 m_iVariableNumber = instanceData->variables->numberOfVariables;
00786         }
00787         return m_iVariableNumber;
00788 }//getVariableNumber
00789 
00790 int OSInstance::getNumberOfNonlinearExpressions(){
00791         if(m_iNonlinearExpressionNumber == -1){
00792                 m_iNonlinearExpressionNumber = instanceData->nonlinearExpressions->numberOfNonlinearExpressions;
00793         }
00794         return m_iNonlinearExpressionNumber;
00795 }//getNumberOfNonlinearExpressions
00796 
00797 
00798 
00799 bool OSInstance::processVariables() {
00800         if(m_bProcessVariables) return true;
00801         m_bProcessVariables = true;
00802         string vartype ="CBIS";
00803         int i = 0;
00804         int n = instanceData->variables->numberOfVariables;
00805         try{
00806                 if(instanceData->variables->var == NULL) throw ErrorClass("no variables defined");
00807                 if(instanceData->variables->var[0]->name.length() > 0 || instanceData->variables->var[n-1]->name.length() > 0){
00808                         m_msVariableNames = new string[n];
00809                         for(i = 0; i < n; i++) m_msVariableNames[i] = instanceData->variables->var[i]->name;
00810                 } 
00811                         m_mdVariableInitialValues = new double[n];
00812                         for(i = 0; i < n; i++){
00813                                 if(CommonUtil::ISOSNAN(instanceData->variables->var[ 0]->init) == true ){                               
00814                                         m_mdVariableInitialValues[i] =  1.7171;
00815                                 }
00816                                 else{
00817                                         m_mdVariableInitialValues[i] = instanceData->variables->var[i]->init;
00818                                 }
00819                         }
00820                 //}
00821                 if((instanceData->variables->var[0]->initString.length() > 0)){
00822                         m_msVariableInitialStringValues = new string[n];
00823                         for(i = 0; i < n; i++) m_msVariableInitialStringValues[i] = instanceData->variables->var[i]->initString;
00824                 }
00825                 m_mcVariableTypes = new char[n];
00826                 m_mdVariableLowerBounds = new double[n];
00827                 m_mdVariableUpperBounds = new double[n];
00828                 for(i = 0; i < n; i++){
00829                         if(vartype.find(instanceData->variables->var[i]->type) == string::npos) throw ErrorClass("wrong variable type");
00830                         m_mcVariableTypes[i] = instanceData->variables->var[i]->type;
00831                         if(m_mcVariableTypes[i] == 'B') m_iNumberOfBinaryVariables++;
00832                         if(m_mcVariableTypes[i] == 'I') m_iNumberOfIntegerVariables++;
00833                         m_mdVariableLowerBounds[i] = instanceData->variables->var[i]->lb;
00834                         m_mdVariableUpperBounds[i] = instanceData->variables->var[i]->ub;
00835                 }
00836                 return true;
00837         } //end try
00838         catch(const ErrorClass& eclass){
00839                 throw ErrorClass( eclass.errormsg);
00840         } 
00841 }//processVariables
00842         
00843 string* OSInstance::getVariableNames() {
00844         processVariables();
00845         return m_msVariableNames;
00846 }//getVariableNames     
00847 
00848 double* OSInstance::getVariableInitialValues() {
00849         processVariables();
00850         return m_mdVariableInitialValues;
00851 }//getVariableInitialValues
00852 
00853 string* OSInstance::getVariableInitialStringValues() {
00854         processVariables();
00855         return m_msVariableInitialStringValues;
00856 }//getVariableInitialStringValues
00857 
00858 char* OSInstance::getVariableTypes() {
00859         processVariables();
00860         return m_mcVariableTypes;
00861 }//getVariableTypes
00862 
00863 int OSInstance::getNumberOfIntegerVariables() {
00864         processVariables();
00865         return m_iNumberOfIntegerVariables;
00866 }//getNumberOfIntegerVariables
00867 
00868 int OSInstance::getNumberOfBinaryVariables() {
00869         processVariables();
00870         return m_iNumberOfBinaryVariables;
00871 }//getNumberOfBinaryVariables
00872 
00873 double* OSInstance::getVariableLowerBounds() {
00874         processVariables();
00875         return m_mdVariableLowerBounds;
00876 }//getVariableLowerBounds
00877 
00878 double* OSInstance::getVariableUpperBounds() {
00879         processVariables();
00880         return m_mdVariableUpperBounds;
00881 }//getVariableUpperBounds
00882 
00883 int OSInstance::getObjectiveNumber(){
00884         if(m_iObjectiveNumber == -1){
00885                 m_iObjectiveNumber = instanceData->objectives->numberOfObjectives;
00886         }
00887         return m_iObjectiveNumber;
00888 }//getObjectiveNumber
00889 
00890 
00891 bool OSInstance::processObjectives() {
00892         if(m_bProcessObjectives) return true;
00893         m_bProcessObjectives = true;
00894         int i = 0;
00895         int j = 0;
00896         if(instanceData == NULL || instanceData->objectives == NULL || instanceData->objectives->obj == NULL || instanceData->objectives->numberOfObjectives == 0) return true;
00897         int n = instanceData->objectives->numberOfObjectives;
00898         try{
00899                 if(instanceData->objectives->obj[0]->name.length() > 0 || instanceData->objectives->obj[n-1]->name.length() > 0){
00900                         m_msObjectiveNames = new string[n];
00901                         for(i = 0; i < n; i++) m_msObjectiveNames[i] = instanceData->objectives->obj[i]->name;
00902                 }
00903                 m_msMaxOrMins = new string[n];
00904                 m_miNumberOfObjCoef = new int[n];
00905                 m_mdObjectiveConstants = new double[n];
00906                 m_mdObjectiveWeights = new double[n];
00907                 m_mObjectiveCoefficients = new SparseVector*[n];
00908                 for(i = 0; i < n; i++){
00909                         m_mObjectiveCoefficients[i] = new SparseVector(instanceData->objectives->obj[ j]->numberOfObjCoef);
00910                         //m_mObjectiveCoefficients[i]->bDeleteArrays=false;
00911                 }
00912                 
00913                 //for(i = 0; i < n; i++){
00914                 //      m_mObjectiveCoefficients[i] = new SparseVector();
00915                 //      m_mObjectiveCoefficients[i]->number = instanceData->objectives->obj[ j]->numberOfObjCoef;
00916                 //}
00917                 for(i = 0; i < n; i++){
00918                         if((instanceData->objectives->obj[i]->maxOrMin.compare("max") != 0) && (instanceData->objectives->obj[i]->maxOrMin.compare("min") != 0 )) throw ErrorClass("wrong objective maxOrMin");
00919                         m_msMaxOrMins[i] = instanceData->objectives->obj[i]->maxOrMin;
00920                         m_miNumberOfObjCoef[i] = instanceData->objectives->obj[i]->numberOfObjCoef;
00921                         m_mdObjectiveConstants[i] = instanceData->objectives->obj[i]->constant;
00922                         m_mdObjectiveWeights[i] = instanceData->objectives->obj[i]->weight;
00923                         if(instanceData->objectives->obj[i]->coef == NULL && m_miNumberOfObjCoef[i] != 0){
00924                                 throw ErrorClass("objective coefficient number inconsistent with objective coefficient array");
00925                         }
00926                         if(instanceData->objectives->obj[i]->coef != NULL){
00927                                 for(j = 0; j < m_mObjectiveCoefficients[i]->number; j++){
00928                                         m_mObjectiveCoefficients[i]->indexes[j] = instanceData->objectives->obj[i]->coef[j]->idx;
00929                                         m_mObjectiveCoefficients[i]->values[j] = instanceData->objectives->obj[i]->coef[j]->value;                      
00930                                 }
00931                         }
00932                 }               
00933                 return true;
00934         }
00935         catch(const ErrorClass& eclass){
00936                 throw ErrorClass( eclass.errormsg);
00937         }
00938 }//processObjectives
00939 
00940 string* OSInstance::getObjectiveNames() {
00941         processObjectives();
00942         return m_msObjectiveNames;
00943 }//getObjectiveNames
00944 
00945 string* OSInstance::getObjectiveMaxOrMins() {
00946         processObjectives();
00947         return m_msMaxOrMins;
00948 }//getObjectiveMaxOrMins
00949 
00950 int* OSInstance::getObjectiveCoefficientNumbers(){
00951                 processObjectives();
00952                 return m_miNumberOfObjCoef;
00953         }//getObjectiveCoefficientNumbers
00954 
00955 double* OSInstance::getObjectiveConstants() {
00956         processObjectives();
00957         return m_mdObjectiveConstants;
00958 }//getObjectiveConstants
00959 
00960 double* OSInstance::getObjectiveWeights() {
00961         processObjectives();
00962         return m_mdObjectiveWeights;
00963 }//getObjectiveWeights
00964 
00965 SparseVector** OSInstance::getObjectiveCoefficients() {
00966         processObjectives();
00967         return m_mObjectiveCoefficients;
00968 }//getObjectiveCoefficients
00969 
00970 
00971 double** OSInstance::getDenseObjectiveCoefficients() {
00972         if(m_bGetDenseObjectives) return m_mmdDenseObjectiveCoefficients;
00973         m_bGetDenseObjectives = true;
00974         if(instanceData->objectives->obj == NULL || instanceData->objectives->numberOfObjectives == 0) return m_mmdDenseObjectiveCoefficients;
00975         int m = instanceData->objectives->numberOfObjectives;
00976         int n = instanceData->variables->numberOfVariables;
00977         m_mmdDenseObjectiveCoefficients = new double*[m];
00978         int i, j, numobjcoef;
00979         SparseVector *sparsevec;
00980         for(i = 0; i < m; i++){
00981                 sparsevec = this->getObjectiveCoefficients()[i];
00982                 m_mmdDenseObjectiveCoefficients[ i] = new double[n];
00983                 for(j = 0; j < n; j++){
00984                         m_mmdDenseObjectiveCoefficients[ i][j] = 0.0;
00985                 }
00986                 sparsevec =  this->getObjectiveCoefficients()[i];
00987                 numobjcoef = sparsevec->number;
00988                 for(j = 0; j < numobjcoef; j++){
00989                         m_mmdDenseObjectiveCoefficients[i][ sparsevec->indexes[ j]]
00990                         = sparsevec->values[ j];
00991                 }
00992         }
00993         return m_mmdDenseObjectiveCoefficients;
00994 }//getDenseObjectiveCoefficients
00995 
00996 int OSInstance::getConstraintNumber(){
00997         if(m_iConstraintNumber == -1){
00998                 m_iConstraintNumber = instanceData->constraints->numberOfConstraints;
00999         }
01000         return m_iConstraintNumber;
01001 }//getConstraintNumber
01002 
01003 bool OSInstance::processConstraints() {
01004         if(m_bProcessConstraints) return true;
01005         m_bProcessConstraints = true;
01006         int i = 0;
01007         ostringstream outStr;
01008         if(instanceData == NULL || instanceData->constraints == NULL || instanceData->constraints->con == NULL || instanceData->constraints->numberOfConstraints == 0) return true;
01009         int n = instanceData->constraints->numberOfConstraints;
01010         try{
01011                 if(instanceData->constraints->con[0]->name.length() > 0 || instanceData->constraints->con[n-1]->name.length() > 0){
01012                         m_msConstraintNames = new string[n];
01013                         for(i = 0; i < n; i++) m_msConstraintNames[i] = instanceData->constraints->con[i]->name;
01014                 }
01015                 m_mdConstraintLowerBounds = new double[n];
01016                 m_mdConstraintUpperBounds = new double[n];
01017                 m_mdConstraintConstants = new double[n];
01018                 m_mcConstraintTypes = new char[n];
01019                 for(i = 0; i < n; i++){
01020                         m_mdConstraintLowerBounds[i] = instanceData->constraints->con[i]->lb;
01021                         m_mdConstraintUpperBounds[i] = instanceData->constraints->con[i]->ub;
01022                         m_mdConstraintConstants[i] = instanceData->constraints->con[i]->constant;
01023                         if(m_mdConstraintLowerBounds[i] == OSDBL_MAX || m_mdConstraintUpperBounds[i] == -OSDBL_MAX) {
01024                                 throw ErrorClass( outStr.str() );
01025                         }
01026                         else if(m_mdConstraintLowerBounds[i] > m_mdConstraintUpperBounds[i]) {
01027                                 outStr << "Constraint  " ;
01028                                 outStr << i;
01029                                 outStr << " is infeasible";
01030                                 throw ErrorClass( outStr.str());
01031                         }
01032                         else if(m_mdConstraintLowerBounds[i] == -OSDBL_MAX && m_mdConstraintUpperBounds[i] == OSDBL_MAX)
01033                                 m_mcConstraintTypes[i] = 'U';
01034                         else if(m_mdConstraintLowerBounds[i] == m_mdConstraintUpperBounds[i]) 
01035                                 m_mcConstraintTypes[i] = 'E';
01036                         else if(m_mdConstraintLowerBounds[i] == -OSDBL_MAX)
01037                                 m_mcConstraintTypes[i] = 'L';
01038                         else if(m_mdConstraintUpperBounds[i] == OSDBL_MAX)
01039                                 m_mcConstraintTypes[i] = 'G';
01040                         else m_mcConstraintTypes[i] = 'R';
01041                 }
01042                 return true;
01043         }
01044         catch(const ErrorClass& eclass){
01045                 throw ErrorClass( eclass.errormsg);
01046         }
01047 }//processConstraints
01048 
01049 
01050 string* OSInstance::getConstraintNames() {
01051         processConstraints();
01052         return m_msConstraintNames;
01053 }//getConstraintNames
01054 
01055 
01056 double* OSInstance::getConstraintLowerBounds() {
01057         processConstraints();
01058         return m_mdConstraintLowerBounds;
01059 }//getConstraintLowerBounds
01060 
01061 char* OSInstance::getConstraintTypes() {
01062         processConstraints();
01063         return m_mcConstraintTypes;
01064 }//getConstraintTypes
01065 
01066 double* OSInstance::getConstraintUpperBounds() {
01067         processConstraints();
01068         return m_mdConstraintUpperBounds;
01069 }//getConstraintUpperBounds
01070 
01071 int OSInstance::getLinearConstraintCoefficientNumber(){
01072         if(m_iLinearConstraintCoefficientNumber == -1){
01073                 m_iLinearConstraintCoefficientNumber = instanceData->linearConstraintCoefficients->numberOfValues;
01074         }
01075         return m_iLinearConstraintCoefficientNumber; 
01076 }//getLinearConstraintCoefficientNumber
01077 
01078 bool OSInstance::processLinearConstraintCoefficients() {
01079         if(m_bProcessLinearConstraintCoefficients) return true;
01080         m_bProcessLinearConstraintCoefficients = true;
01081         try{
01082                 int n = instanceData->linearConstraintCoefficients->numberOfValues;
01083                 //value array
01084                 if((instanceData->linearConstraintCoefficients->value == NULL ) || (n == 0) ) return true;
01085                 //index array
01086                 if((instanceData->linearConstraintCoefficients->colIdx != NULL && instanceData->linearConstraintCoefficients->colIdx->el != NULL) 
01087                 && (instanceData->linearConstraintCoefficients->rowIdx != NULL && instanceData->linearConstraintCoefficients->rowIdx->el != NULL))
01088                         throw ErrorClass("ambiguous linear constraint coefficient major");
01089                 else if(instanceData->linearConstraintCoefficients->value->el == NULL) return true;
01090                 else{
01091                         if(instanceData->linearConstraintCoefficients->rowIdx->el != NULL){
01092                                 m_bColumnMajor = true;
01093                                 m_linearConstraintCoefficientsInColumnMajor = new SparseMatrix();
01094                                 m_linearConstraintCoefficientsInColumnMajor->bDeleteArrays = false;
01095                                 m_linearConstraintCoefficientsInColumnMajor->isColumnMajor = true;
01096                                 m_linearConstraintCoefficientsInColumnMajor->valueSize = n;
01097                                 m_linearConstraintCoefficientsInColumnMajor->startSize = instanceData->variables->numberOfVariables + 1;
01098                         }
01099                         else{ 
01100                                 m_bColumnMajor = false; 
01101                                 m_linearConstraintCoefficientsInRowMajor = new SparseMatrix();
01102                                 m_linearConstraintCoefficientsInRowMajor->bDeleteArrays = false;
01103                                 m_linearConstraintCoefficientsInRowMajor->isColumnMajor = false;
01104                                 m_linearConstraintCoefficientsInRowMajor->valueSize = n;
01105                                 m_linearConstraintCoefficientsInRowMajor->startSize = instanceData->constraints->numberOfConstraints + 1;
01106                         }
01107                 }                       
01108                 if(m_bColumnMajor == true){
01109                         m_linearConstraintCoefficientsInColumnMajor->values = instanceData->linearConstraintCoefficients->value->el;
01110                         m_linearConstraintCoefficientsInColumnMajor->indexes = instanceData->linearConstraintCoefficients->rowIdx->el;
01111                         m_linearConstraintCoefficientsInColumnMajor->starts = instanceData->linearConstraintCoefficients->start->el;                    
01112                 }
01113                 else{
01114                         m_linearConstraintCoefficientsInRowMajor->values = instanceData->linearConstraintCoefficients->value->el;
01115                         m_linearConstraintCoefficientsInRowMajor->indexes = instanceData->linearConstraintCoefficients->colIdx->el;
01116                         m_linearConstraintCoefficientsInRowMajor->starts = instanceData->linearConstraintCoefficients->start->el;                                               
01117                 }
01118                 return true;
01119         }
01120         catch(const ErrorClass& eclass){
01121                 throw ErrorClass( eclass.errormsg);
01122         }
01123 }//processLinearConstraintCoefficients
01124 
01125 bool OSInstance::getLinearConstraintCoefficientMajor() {
01126         processLinearConstraintCoefficients();  
01127         return m_bColumnMajor;          
01128 }//getLinearConstraintCoefficientMajor
01129 
01130 SparseMatrix* OSInstance::getLinearConstraintCoefficientsInColumnMajor() {
01131         processLinearConstraintCoefficients();
01132         if(m_linearConstraintCoefficientsInColumnMajor != NULL) return m_linearConstraintCoefficientsInColumnMajor;
01133         if(!m_bColumnMajor){
01134                 if(m_linearConstraintCoefficientsInRowMajor == NULL) return NULL;
01135                 m_linearConstraintCoefficientsInColumnMajor = 
01136                         MathUtil::convertLinearConstraintCoefficientMatrixToTheOtherMajor(false,
01137                                           m_linearConstraintCoefficientsInRowMajor->startSize,
01138                                           m_linearConstraintCoefficientsInRowMajor->valueSize,
01139                                           m_linearConstraintCoefficientsInRowMajor->starts,
01140                                           m_linearConstraintCoefficientsInRowMajor->indexes,
01141                                           m_linearConstraintCoefficientsInRowMajor->values,
01142                                           getVariableNumber());
01143         }
01144         return m_linearConstraintCoefficientsInColumnMajor;             
01145 }//getLinearConstraintCoefficientsInColumnMajor
01146 
01147 SparseMatrix* OSInstance::getLinearConstraintCoefficientsInRowMajor() {
01148         processLinearConstraintCoefficients();
01149         if(m_linearConstraintCoefficientsInRowMajor != NULL) return m_linearConstraintCoefficientsInRowMajor;
01150         if(m_bColumnMajor){
01151                 if(m_linearConstraintCoefficientsInColumnMajor == NULL) return NULL;
01152                 m_linearConstraintCoefficientsInRowMajor = 
01153                 MathUtil::convertLinearConstraintCoefficientMatrixToTheOtherMajor(true,
01154                                   m_linearConstraintCoefficientsInColumnMajor->startSize,
01155                                   m_linearConstraintCoefficientsInColumnMajor->valueSize,
01156                                   m_linearConstraintCoefficientsInColumnMajor->starts,
01157                                   m_linearConstraintCoefficientsInColumnMajor->indexes,
01158                                   m_linearConstraintCoefficientsInColumnMajor->values,
01159                                   getConstraintNumber());
01160         }
01161         return m_linearConstraintCoefficientsInRowMajor; 
01162 }//getLinearConstraintCoefficientsInRowMajor
01163 
01164 
01165 int OSInstance::getNumberOfQuadraticTerms(){
01166         if(m_iQuadraticTermNumber == -1){
01167         // if m_iQuadraticTermNumber == -1 then the parser did not find any q terms so 
01168         // must new the object
01169                 if(instanceData->quadraticCoefficients == NULL)instanceData->quadraticCoefficients = new QuadraticCoefficients();
01170                 m_iQuadraticTermNumber = instanceData->quadraticCoefficients->numberOfQuadraticTerms;
01171         }
01172         return m_iQuadraticTermNumber;
01173 }//getNumberOfQuadraticTerms
01174 
01175 QuadraticTerms* OSInstance::getQuadraticTerms() {
01176         if(m_bProcessQuadraticTerms) return m_quadraticTerms;
01177         m_bProcessQuadraticTerms = true;
01178         if(instanceData->quadraticCoefficients->qTerm == 0) return 0;
01179         try{
01180                 int i = 0;
01181                 QuadraticCoefficients* quadraticCoefs = instanceData->quadraticCoefficients;
01182                 int n = quadraticCoefs->numberOfQuadraticTerms;
01183                 if(!quadraticCoefs->qTerm  && n != 0) 
01184                         throw ErrorClass("quadratic term number inconsistent with quadratic term array");               
01185                 m_quadraticTerms = new QuadraticTerms();
01186                 m_quadraticTerms->rowIndexes = new int[n];
01187                 m_quadraticTerms->varOneIndexes = new int[n];
01188                 m_quadraticTerms->varTwoIndexes = new int[n];
01189                 m_quadraticTerms->coefficients = new double[n];
01190                 for(i = 0; i < n; i++){
01191                         m_quadraticTerms->rowIndexes[i] = quadraticCoefs->qTerm[i]->idx;
01192                         m_quadraticTerms->varOneIndexes[i] = quadraticCoefs->qTerm[i]->idxOne;
01193                         m_quadraticTerms->varTwoIndexes[i] = quadraticCoefs->qTerm[i]->idxTwo;
01194                         m_quadraticTerms->coefficients[i] = quadraticCoefs->qTerm[i]->coef;
01195                 } 
01196                 return m_quadraticTerms;
01197         }
01198         catch(const ErrorClass& eclass){
01199                 throw ErrorClass( eclass.errormsg);
01200         } 
01201 }//getQuadraticTerms
01202 
01203 
01204 int* OSInstance::getQuadraticRowIndexes() {
01205         if(m_bQuadraticRowIndexesProcessed == true) return m_miQuadRowIndexes;
01206         m_bQuadraticRowIndexesProcessed = true;
01207         int n = getNumberOfQuadraticTerms();    
01208         if(n <= 0) return NULL;
01209         QuadraticTerms *qTerms = NULL;
01210         qTerms = getQuadraticTerms();
01211         std::map<int, int> foundIdx;
01212         std::map<int, int>::iterator pos;
01213         int i;
01214         try{
01215                 for(i = 0; i < n; i++){
01216                         // add the terms
01217                         foundIdx[ qTerms->rowIndexes[ i] ];      
01218                 }
01219                 // now put the term into an array
01220                 m_iNumberOfQuadraticRowIndexes = foundIdx.size();
01221                 m_miQuadRowIndexes = new int[ m_iNumberOfQuadraticRowIndexes ]  ;
01222                 i = 0;
01223                 for(pos = foundIdx.begin(); pos != foundIdx.end(); ++pos){
01224                         m_miQuadRowIndexes[ i++] = pos->first;  
01225                 }
01226                 foundIdx.clear();       
01227                 return m_miQuadRowIndexes;
01228         }
01229         catch(const ErrorClass& eclass){
01230                 throw ErrorClass( eclass.errormsg);
01231         } 
01232 }//getQuadraticRowIndexes
01233 
01234 
01235 int OSInstance::getNumberOfQuadraticRowIndexes() {
01236         if(m_bQuadraticRowIndexesProcessed == false) getQuadraticRowIndexes();
01237         return m_iNumberOfQuadraticRowIndexes;
01238 }//getNumberOfQuadraticRowIndexes
01239 
01240 int* OSInstance::getNonlinearExpressionTreeIndexes(){
01241         if(m_bNonlinearExpressionTreeIndexesProcessed == true) return m_miNonlinearExpressionTreeIndexes;
01242         m_bNonlinearExpressionTreeIndexesProcessed = true;
01243         std::map<int, OSExpressionTree*> expTrees;
01244         expTrees = getAllNonlinearExpressionTrees();    
01245         std::map<int, OSExpressionTree*>::iterator pos;
01246         try{
01247                 // now put the term into an array
01248                 m_iNumberOfNonlinearExpressionTreeIndexes = expTrees.size();
01249                 m_miNonlinearExpressionTreeIndexes = new int[ m_iNumberOfNonlinearExpressionTreeIndexes ]       ;
01250                 int i = 0;
01251                 for(pos = expTrees.begin(); pos != expTrees.end(); ++pos){
01252                         m_miNonlinearExpressionTreeIndexes[ i++] = pos->first;  
01253                 }
01254                 expTrees.clear();       
01255                 return m_miNonlinearExpressionTreeIndexes;
01256         }
01257         catch(const ErrorClass& eclass){
01258                 throw ErrorClass( eclass.errormsg);
01259         } 
01260 }//getNonlinearExpressionTreeIndexes
01261 
01262 int OSInstance::getNumberOfNonlinearExpressionTreeIndexes() {
01263         if(m_bNonlinearExpressionTreeIndexesProcessed == false) getNonlinearExpressionTreeIndexes();
01264         return m_iNumberOfNonlinearExpressionTreeIndexes;
01265 }//getNumberOfNonlinearExpressionTreeIndexes
01266 
01267 
01268 
01269 int* OSInstance::getNonlinearExpressionTreeModIndexes(){
01270         if(m_bNonlinearExpressionTreeModIndexesProcessed == true) return m_miNonlinearExpressionTreeModIndexes;
01271         m_bNonlinearExpressionTreeModIndexesProcessed = true;
01272         std::map<int, OSExpressionTree*> expTrees;
01273         expTrees = getAllNonlinearExpressionTreesMod(); 
01274         std::map<int, OSExpressionTree*>::iterator pos;
01275         try{
01276                 // now put the term into an array
01277                 m_iNumberOfNonlinearExpressionTreeModIndexes = expTrees.size();
01278                 m_miNonlinearExpressionTreeModIndexes = new int[ m_iNumberOfNonlinearExpressionTreeModIndexes ] ;
01279                 int i = 0;
01280                 for(pos = expTrees.begin(); pos != expTrees.end(); ++pos){
01281                         m_miNonlinearExpressionTreeModIndexes[ i++] = pos->first;       
01282                 }
01283                 expTrees.clear();       
01284                 return m_miNonlinearExpressionTreeModIndexes;
01285         }
01286         catch(const ErrorClass& eclass){
01287                 throw ErrorClass( eclass.errormsg);
01288         } 
01289 }//getNonlinearExpressionTreeModIndexes
01290 
01291 int OSInstance::getNumberOfNonlinearExpressionTreeModIndexes() {
01292         if(m_bNonlinearExpressionTreeModIndexesProcessed == false) getNonlinearExpressionTreeModIndexes();
01293         return m_iNumberOfNonlinearExpressionTreeModIndexes;
01294 }//getNumberOfNonlinearExpressionTreeModIndexes
01295 
01296 
01297 int OSInstance::getNumberOfNonlinearConstraints(){
01298         if( m_bProcessExpressionTrees == false ){
01299                 getAllNonlinearExpressionTrees();
01300                 return m_iConstraintNumberNonlinear;
01301         }
01302         else return m_iConstraintNumberNonlinear;
01303 }//getNumberOfNonlinearConstraints
01304 
01305 int OSInstance::getNumberOfNonlinearObjectives(){
01306         if( m_bProcessExpressionTrees == false ){
01307                 getAllNonlinearExpressionTrees();
01308                 return m_iObjectiveNumberNonlinear;
01309         }
01310         else return m_iObjectiveNumberNonlinear;
01311 }//getNumberOfNonlinearObjectivess
01312 
01313 
01314 OSExpressionTree* OSInstance::getNonlinearExpressionTree(int rowIdx){
01315         if( m_bProcessExpressionTrees == false ){
01316                 getAllNonlinearExpressionTrees();
01317                 return m_mapExpressionTrees[ rowIdx];
01318         } 
01319         else{
01320                 //if( m_mapExpressionTrees.find( rowIdx) != m_mapExpressionTrees.end()) return NULL;
01321                 //else return m_mapExpressionTrees[ rowIdx];
01322                 if( m_mapExpressionTrees.find( rowIdx) != m_mapExpressionTrees.end()) return m_mapExpressionTrees[ rowIdx];
01323                 else return NULL ;
01324                 // check to make sure rowIdx has a nonlinear term and is in the map
01326                 //std::map<int, OSExpressionTree*>::iterator pos;
01328                 //      if(pos->first == rowIdx)return m_mapExpressionTrees[ rowIdx];
01329                 //}
01330                 // if we are rowIdx has no nonlinar terms so return a null
01331                 //return NULL;
01332         }     
01333 }// getNonlinearExpressionTree for a specific index   
01334 
01335 
01336 OSExpressionTree* OSInstance::getNonlinearExpressionTreeMod(int rowIdx){
01337         if( m_bProcessExpressionTreesMod == false ){
01338                 getAllNonlinearExpressionTreesMod();
01339                 return m_mapExpressionTreesMod[ rowIdx];
01340         } 
01341         else{
01342                 //if( m_mapExpressionTrees.find( rowIdx) != m_mapExpressionTrees.end()) return NULL;
01343                 //else return m_mapExpressionTrees[ rowIdx];
01344                 if( m_mapExpressionTreesMod.find( rowIdx) != m_mapExpressionTreesMod.end()) return m_mapExpressionTreesMod[ rowIdx];
01345                 else return NULL ;
01346                 // check to make sure rowIdx has a nonlinear term and is in the map
01348                 //std::map<int, OSExpressionTree*>::iterator pos;
01350                 //      if(pos->first == rowIdx)return m_mapExpressionTrees[ rowIdx];
01351                 //}
01352                 // if we are rowIdx has no nonlinar terms so return a null
01353                 //return NULL;
01354         }     
01355 }// getNonlinearExpressionTreeMod for a specific index 
01356 
01357 
01358 std::vector<OSnLNode*> OSInstance::getNonlinearExpressionTreeInPostfix( int rowIdx){
01359         if( m_bProcessExpressionTrees == false ) getAllNonlinearExpressionTrees();
01360         std::vector<OSnLNode*> postfixVec;
01361         try{
01362                 if( m_mapExpressionTrees.find( rowIdx) != m_mapExpressionTrees.end()){
01363                         OSExpressionTree* expTree = getNonlinearExpressionTree( rowIdx);
01364                         postfixVec = expTree->m_treeRoot->getPostfixFromExpressionTree();
01365                         
01366                 }  
01367                 else{
01368                         throw ErrorClass("Error in getNonlinearExpressionTreeInPostfix, rowIdx not valid");
01369                 }
01370                 return postfixVec;      
01371         }
01372         catch(const ErrorClass& eclass){
01373                 throw ErrorClass( eclass.errormsg);
01374         } 
01375 }//getNonlinearExpressionTreeInPostfix
01376 
01377 
01378 std::vector<OSnLNode*> OSInstance::getNonlinearExpressionTreeModInPostfix( int rowIdx){
01379         if( m_bProcessExpressionTreesMod == false ) getAllNonlinearExpressionTreesMod();
01380         std::vector<OSnLNode*> postfixVec;
01381         try{
01382                 if( m_mapExpressionTreesMod.find( rowIdx) != m_mapExpressionTreesMod.end()){
01383                         OSExpressionTree* expTree = getNonlinearExpressionTreeMod( rowIdx);
01384                         postfixVec = expTree->m_treeRoot->getPostfixFromExpressionTree();
01385                         
01386                 }  
01387                 else{
01388                         throw ErrorClass("Error in getNonlinearExpressionTreeModInPostfix, rowIdx not valid");
01389                 }
01390                 return postfixVec;      
01391         }
01392         catch(const ErrorClass& eclass){
01393                 throw ErrorClass( eclass.errormsg);
01394         } 
01395 }//getNonlinearExpressionTreeModInPostfix
01396 
01397 
01398 std::vector<OSnLNode*> OSInstance::getNonlinearExpressionTreeInPrefix( int rowIdx){
01399         if( m_bProcessExpressionTrees == false ) getAllNonlinearExpressionTrees();
01400         std::vector<OSnLNode*> prefixVec;
01401         try{
01402                 if( m_mapExpressionTrees.find( rowIdx) != m_mapExpressionTrees.end()){
01403                         OSExpressionTree* expTree = getNonlinearExpressionTree( rowIdx);
01404                         prefixVec = expTree->m_treeRoot->getPrefixFromExpressionTree();
01405                         
01406                 }  
01407                 else{
01408                         throw ErrorClass("Error in getNonlinearExpressionTreeInPrefix, rowIdx not valid");
01409                 }
01410                 return prefixVec;       
01411         }
01412         catch(const ErrorClass& eclass){
01413                 throw ErrorClass( eclass.errormsg);
01414         } 
01415 }//getNonlinearExpressionTreeInPrefix
01416 
01417 std::vector<OSnLNode*> OSInstance::getNonlinearExpressionTreeModInPrefix( int rowIdx){
01418         if( m_bProcessExpressionTreesMod == false ) getAllNonlinearExpressionTreesMod();
01419         std::vector<OSnLNode*> prefixVec;
01420         try{
01421                 if( m_mapExpressionTreesMod.find( rowIdx) != m_mapExpressionTreesMod.end()){
01422                         OSExpressionTree* expTree = getNonlinearExpressionTreeMod( rowIdx);
01423                         prefixVec = expTree->m_treeRoot->getPrefixFromExpressionTree();
01424                         
01425                 }  
01426                 else{
01427                         throw ErrorClass("Error in getNonlinearExpressionTreeInPrefix, rowIdx not valid");
01428                 }
01429                 return prefixVec;       
01430         }
01431         catch(const ErrorClass& eclass){
01432                 throw ErrorClass( eclass.errormsg);
01433         } 
01434 }//getNonlinearExpressionTreeInPrefix
01435 
01436 std::map<int, OSExpressionTree*> OSInstance::getAllNonlinearExpressionTrees(){
01437         if(m_bProcessExpressionTrees == true) return m_mapExpressionTrees;
01438         std::map<int, int> foundIdx;
01439         std::map<int, int>::iterator pos;
01440         OSnLNodePlus *nlNodePlus;
01441         OSExpressionTree *expTree;  
01442         m_iObjectiveNumberNonlinear = 0;   
01443         m_iConstraintNumberNonlinear = 0;    
01444         int i;   
01445         // important -- tell the nl nodes not to destroy the OSExpression Objects
01446         if( instanceData->nonlinearExpressions->numberOfNonlinearExpressions > 0 && instanceData->nonlinearExpressions->nl != NULL){
01447                 for( i = 0; i < instanceData->nonlinearExpressions->numberOfNonlinearExpressions; i++){
01448                         instanceData->nonlinearExpressions->nl[i]->m_bDeleteExpressionTree = false;
01449                 }
01450         }
01451         int index;  
01452         // kipp -- what should we return if instanceData->nonlinearExpressions->numberOfNonlinearExpressions is zero
01453         for(i = 0; i < instanceData->nonlinearExpressions->numberOfNonlinearExpressions; i++){
01454                 index = instanceData->nonlinearExpressions->nl[ i]->idx;
01455                 if(foundIdx.find( index) != foundIdx.end() ){  
01456                 //if(foundIdx[ index] > 0 ){ 
01457                         //std::cout << "OLD INDEX FOUND " << index << std::endl;
01458                         //std::cout << "foundIdx[ index] " << index << std::endl;
01459                         // found an existing index
01460                         // important -- at this time m_mapExpressionTrees[ index] points to 
01461                         // the last OSExpressionTree with this index, it does not point to the 
01462                         // the just found OSExpressionTree with this index
01463                         nlNodePlus = new OSnLNodePlus();
01464                         //expTree = new OSExpressionTree(); 
01465                         expTree =  instanceData->nonlinearExpressions->nl[ i]->osExpressionTree;
01466                         // set left child to old index and right child to new one 
01467                         nlNodePlus->m_mChildren[ 0] = m_mapExpressionTrees[ index]->m_treeRoot;
01468                         nlNodePlus->m_mChildren[ 1] = instanceData->nonlinearExpressions->nl[ i]->osExpressionTree->m_treeRoot;
01469                         // we must delete the Expression tree corresponding to the old index value but not the nl nodes
01470                         instanceData->nonlinearExpressions->nl[ foundIdx[ index]  ]->m_bDeleteExpressionTree = true;
01471                         instanceData->nonlinearExpressions->nl[ foundIdx[ index]  ]->osExpressionTree->bDestroyNlNodes = false;
01472                         //point to the new expression tree
01473                         m_mapExpressionTrees[ index] = expTree;
01474                         m_mapExpressionTrees[ index]->m_treeRoot = nlNodePlus;
01475                         foundIdx[ index] = i;
01476                 }
01477                 else{  
01478                         // we have a new index
01479                         m_mapExpressionTrees[ index] = instanceData->nonlinearExpressions->nl[ i]->osExpressionTree;
01480                         m_mapExpressionTrees[ index]->m_treeRoot = instanceData->nonlinearExpressions->nl[ i]->osExpressionTree->m_treeRoot;
01481                         foundIdx[ index] = i;
01482                 }
01483                 //foundIdx[ index]++;    
01484         }
01485         // count the number of constraints and objective functions with nonlinear terms.
01486         for(pos = foundIdx.begin(); pos != foundIdx.end(); ++pos){
01487                 if(pos->first == -1) {
01488                         m_iObjectiveNumberNonlinear++; 
01489                 }
01490                 else m_iConstraintNumberNonlinear++;
01491         }
01492         foundIdx.clear();
01493         m_bProcessExpressionTrees = true;
01494         return m_mapExpressionTrees;
01495 }// getAllNonlinearExpressionTrees
01496 
01497 std::map<int, OSExpressionTree*> OSInstance::getAllNonlinearExpressionTreesMod(){
01498         if( m_bProcessExpressionTreesMod == true ) return m_mapExpressionTreesMod;
01499         m_bProcessExpressionTreesMod = true;
01500         // make sure we have the modified map available
01501         if( m_bNonLinearStructuresInitialized == false) initializeNonLinearStructures( );
01502         return m_mapExpressionTreesMod;
01503 }// getAllNonlinearExpressionTreesMod
01504 
01505 //bool OSInstance::processTimeDomain() {
01506 //      if(m_bProcessTimeDomain) return true;
01507 //      m_bProcessTimeDomain = true;
01508 //}// processTimeDomain
01509 
01510 
01511 
01512 
01513 // the set() methods
01514 
01515 bool OSInstance::setInstanceSource(string source){
01516         instanceHeader->source = source;
01517         return true;
01518 }//setInstanceSource
01519 
01520 bool OSInstance::setInstanceDescription(string description){
01521         instanceHeader->description = description;
01522         return true;
01523 }//setInstanceDescription
01524 
01525 
01526 bool OSInstance::setInstanceName(string name){
01527          instanceHeader->name = name;
01528          return true;
01529 }//setInstanceName
01530 
01531 
01532 bool OSInstance::setVariableNumber(int number){
01533         // this method assume osinstance->instanceData->variables is not null
01534         if(number <= 0) return false;
01535         //if(instanceData->variables->numberOfVariables != -1  && instanceData->variables->numberOfVariables != number){
01536         //      delete[] instanceData->variables->var;
01537         //      instanceData->variables->var = NULL;
01538         //} 
01539         if(instanceData->variables == NULL) instanceData->variables = new Variables();
01540         instanceData->variables->numberOfVariables = number;
01541         if(instanceData->variables->var == NULL){
01542                 instanceData->variables->var = new Variable*[number];
01543         }
01544         return true;
01545 }//setVariableNumber
01546 
01547 
01548 bool OSInstance::addVariable(int index, string name, double lowerBound, double upperBound, char type, double init, string initString){
01549         instanceData->variables->var[index] = new Variable();
01550         if(index < 0 || instanceData->variables->numberOfVariables <= 0 || index >= instanceData->variables->numberOfVariables) return false;
01551         instanceData->variables->var[index]->name = name;
01552         //if(lowerBound != -OSDBL_MAX && lowerBound != -OSDBL_MAX)instanceData->variables->var[index]->lb = lowerBound;
01553         instanceData->variables->var[index]->lb = lowerBound;
01554         if(upperBound != OSDBL_MAX && upperBound != OSDBL_MAX)instanceData->variables->var[index]->ub = upperBound;
01555         instanceData->variables->var[index]->type = type;
01556         if(init != OSNAN) instanceData->variables->var[index]->init = init;
01557         instanceData->variables->var[index]->initString = initString;
01558         return true;
01559 }//addVariable
01560 
01561 
01562 bool OSInstance::setVariables(int number, string *names, double *lowerBounds, 
01563         double *upperBounds, char *types, double *inits, string *initsString){
01564         if(number <= 0) return false;
01565         try{
01566                 if(instanceData->variables == NULL){
01567                         throw ErrorClass("There is no variables object");
01568                 }
01569                 if(instanceData->variables->numberOfVariables != number){
01570                         throw ErrorClass("input number of variables not equal to number in class");
01571                 }
01572                 //instanceData->variables->var = new Variable*[number];
01573                 int i;
01574                 for(i = 0; i < number; i++){
01575                         instanceData->variables->var[ i] = new Variable();
01576                 }
01577                 if(names  != NULL){
01578                         for(i = 0; i < number; i++) instanceData->variables->var[i]->name = names[i];                           
01579                 }
01580                 if(lowerBounds != NULL){
01581                         for(i = 0; i < number; i++){
01582                                 if(lowerBounds[i] != -OSDBL_MAX && lowerBounds[i] != -OSDBL_MAX)instanceData->variables->var[i]->lb = lowerBounds[i];  
01583                         }
01584                 }
01585                 if(upperBounds != NULL){
01586                         for(i = 0; i < number; i++){
01587                                 if(upperBounds[i] != OSDBL_MAX && upperBounds[i] != OSDBL_MAX)instanceData->variables->var[i]->ub = upperBounds[i]; 
01588                         }
01589                 }
01590                 if(types != NULL){
01591                         for(i = 0; i < number; i++){
01592                                 instanceData->variables->var[i]->type = types[i];
01593                                 if(types[i] != 'C' && types[i] != 'B' && types[i] != 'I' && types[i] != 'S') types[i] = 'C';
01594                         } 
01595                 }
01596                 if(inits != NULL){
01597                         for(i = 0; i < number; i++) instanceData->variables->var[i]->init = inits[i];                           
01598                 }
01599                 if(initsString != NULL){
01600                         for(i = 0; i < number; i++) instanceData->variables->var[i]->initString = initsString[i];                       
01601                 }
01602                 return true;
01603         }
01604         catch(const ErrorClass& eclass){
01605                 throw ErrorClass(  eclass.errormsg); 
01606         }
01607 }//setVariables
01608 
01609 // begin checking again with Jun Ma
01610 
01611 bool OSInstance::setObjectiveNumber(int number){
01612         if(number < 0) return false;
01613         if(instanceData->objectives == NULL) instanceData->objectives = new Objectives();
01614         if(number == 0){
01615                 instanceData->objectives->numberOfObjectives = 0;
01616                 instanceData->objectives->obj = 0;
01617                 return true;
01618         }
01619         instanceData->objectives->numberOfObjectives = number;
01620         instanceData->objectives->obj = new Objective*[number];                         
01621         return true;
01622 }//setObjectiveNumber
01623 
01624 bool OSInstance::addObjective(int index, string name, string maxOrMin, double constant, double weight, SparseVector *objectiveCoefficients){
01625         if(index >= 0 || instanceData->objectives->numberOfObjectives <= 0 || abs(index) > instanceData->objectives->numberOfObjectives) return false;
01626         int arrayIndex = abs(index) -1;
01627         if(instanceData->objectives->obj == NULL) return false;
01628         instanceData->objectives->obj[arrayIndex] = new Objective();
01629         instanceData->objectives->obj[arrayIndex]->name = name;
01630         if( (maxOrMin != "max") && (maxOrMin != "min") ) return false;
01631         else instanceData->objectives->obj[arrayIndex]->maxOrMin = maxOrMin;
01632         instanceData->objectives->obj[arrayIndex]->constant = constant;
01633         instanceData->objectives->obj[arrayIndex]->weight = weight;
01634         int n = objectiveCoefficients->number;                  
01635         instanceData->objectives->obj[arrayIndex]->numberOfObjCoef = n;
01636         if(n == 0){
01637                 instanceData->objectives->obj[arrayIndex]->coef = 0;
01638         }
01639         else{
01640                 int i = 0;
01641                 instanceData->objectives->obj[arrayIndex]->coef = new ObjCoef*[n];
01642                 for(i = 0; i < n; i++) instanceData->objectives->obj[arrayIndex]->coef[i] = new ObjCoef();
01643                 for(i = 0; i < n; i++){
01644                         instanceData->objectives->obj[arrayIndex]->coef[i]->idx = objectiveCoefficients->indexes[i];
01645                         instanceData->objectives->obj[arrayIndex]->coef[i]->value = objectiveCoefficients->values[i];   
01646                 }  
01647         }
01648         return true;
01649 }//addObjective
01650 
01651 bool OSInstance::setObjectives(int number, string *names, string *maxOrMins, double *constants, double *weights, SparseVector **objectiveCoefficients){
01652         if(number < 0) return false;
01653         try{
01654                 if(instanceData->objectives == NULL){
01655                         throw ErrorClass("there is no objectives object");              
01656                 }               
01657                 if(instanceData->objectives->numberOfObjectives != number){
01658                         throw ErrorClass("input number of objective not equal to number in class");             
01659                 }
01660                 if(number == 0) return true;
01661                 int i = 0;
01662                 for(i = 0; i < number; i++)instanceData->objectives->obj[i] = new Objective();
01663                 int j = 0;
01664                 if(names != NULL){
01665                         for(i = 0; i < number; i++) instanceData->objectives->obj[i]->name = names[i];                          
01666                 }       
01667                 if(maxOrMins != NULL){
01668                         for(i = 0; i < number; i++){
01669                                 if(maxOrMins[i] == "" || (maxOrMins[i].compare("max") != 0 && maxOrMins[i].compare("min") !=0)) return false;
01670                                 instanceData->objectives->obj[i]->maxOrMin = maxOrMins[i];                      
01671                         }
01672                 }
01673                 if(constants != NULL){
01674                         for(i = 0; i < number; i++) instanceData->objectives->obj[i]->constant = constants[i];                          
01675                 }
01676                 if(weights != NULL){
01677                         for(i = 0; i < number; i++) instanceData->objectives->obj[i]->weight = weights[i];                      
01678                 }
01679                 if(objectiveCoefficients != NULL){
01680                         for(i = 0; i < number; i++){
01681                                 int n = (&objectiveCoefficients[i] == NULL || objectiveCoefficients[i]->indexes == NULL)?0:objectiveCoefficients[i]->number;            
01682                                 instanceData->objectives->obj[i]->numberOfObjCoef = n;
01683                                 if(n == 0){
01684                                         instanceData->objectives->obj[i]->coef = NULL;
01685                                 }
01686                                 else{
01687                                         instanceData->objectives->obj[i]->coef = new ObjCoef*[n];
01688                                         for(j = 0; j < n; j++){
01689                                                 instanceData->objectives->obj[i]->coef[j] = new ObjCoef();
01690                                                 instanceData->objectives->obj[i]->coef[j]->idx  = objectiveCoefficients[i]->indexes[j];
01691                                                 instanceData->objectives->obj[i]->coef[j]->value = objectiveCoefficients[i]->values[j];                         
01692                                         }                       
01693                                 }                                                       
01694                         }
01695                 }
01696                 return true;
01697         }
01698         catch(const ErrorClass& eclass){
01699                 throw ErrorClass(  eclass.errormsg); 
01700         }
01701 }//setObjectives
01702 
01703 
01704 bool OSInstance::setConstraintNumber(int number){
01705         if(number < 0) return false;
01706         if(instanceData->constraints == NULL) instanceData->constraints = new Constraints();
01707         if(number == 0){
01708                 instanceData->constraints->numberOfConstraints = 0;
01709                 instanceData->constraints->con = 0;
01710                 return true;
01711         }
01712         instanceData->constraints->numberOfConstraints = number;
01713         if(instanceData->constraints->con == 0 ){
01714                 instanceData->constraints->con = new Constraint*[number];
01715         }
01716         return true;
01717 }//setConstraintNumber
01718 
01719 bool OSInstance::addConstraint(int index, string name, double lowerBound, double upperBound, double constant) {
01720         instanceData->constraints->con[ index] = new Constraint();
01721         if(index < 0 || instanceData->constraints->numberOfConstraints <= 0 || index >= instanceData->constraints->numberOfConstraints) return false;
01722         instanceData->constraints->con[ index]->name = name;
01723         if(lowerBound != -OSDBL_MAX && lowerBound != -OSDBL_MAX) instanceData->constraints->con[ index]->lb = lowerBound;
01724         if(upperBound != OSDBL_MAX && upperBound != OSDBL_MAX)instanceData->constraints->con[ index]->ub = upperBound;
01725         instanceData->constraints->con[ index]->constant = constant;
01726         return true;
01727 }//addConstraint
01728 
01729 
01730 bool OSInstance::setConstraints(int number, string* names, double* lowerBounds, double* upperBounds, double* constants){
01731         if(number < 0) return false;
01732         if(number == 0){
01733                 // this is done in setConstraintNumber
01734                 //instanceData->constraints = new Constraints();
01735                 //instanceData->constraints->numberOfConstraints = 0;
01736                 //instanceData->constraints->con = NULL;
01737                 return true;
01738         }
01739         try{
01740                 
01741                 if(instanceData->constraints  == NULL){
01742                         throw ErrorClass("there is no constraints object");             
01743                 }               
01744                 if(instanceData->constraints->numberOfConstraints != number){
01745                         throw ErrorClass("input number of constrasints not equal to number in class");          
01746                 }
01747                 int i = 0;
01748                 for(i = 0; i < number; i++){
01749                         instanceData->constraints->con[i] = new Constraint();
01750                 }
01751                 if(names != NULL){
01752                         for(i = 0; i < number; i++) instanceData->constraints->con[i]->name = names[i];                         
01753                 }
01754                 if(lowerBounds != NULL){
01755                         for(i = 0; i < number; i++){
01756                                 if(lowerBounds[i] != -OSDBL_MAX && lowerBounds[i] != -OSDBL_MAX)instanceData->constraints->con[i]->lb = lowerBounds[i]; 
01757                         }
01758                 }
01759                 if(upperBounds != NULL){
01760                         for(i = 0; i < number; i++){
01761                                 if(upperBounds[i] != OSDBL_MAX && upperBounds[i] != OSDBL_MAX)instanceData->constraints->con[i]->ub = upperBounds[i]; 
01762                         }
01763                 }   
01764                 if(constants != NULL){
01765                         for(i = 0; i < number; i++) instanceData->constraints->con[i]->constant = constants[i];                         
01766                 }
01767                 return true;
01768         }
01769         catch(const ErrorClass& eclass){
01770                 throw ErrorClass(  eclass.errormsg); 
01771         }
01772 }//setConstraints
01773 
01774 bool OSInstance::setLinearConstraintCoefficients(int numberOfValues, bool isColumnMajor, 
01775                 double* values, int valuesBegin, int valuesEnd, 
01776                 int* indexes, int indexesBegin, int indexesEnd,                         
01777                 int* starts, int startsBegin, int startsEnd){
01778                 if(numberOfValues < 0) return false;
01779         if(instanceData->linearConstraintCoefficients == NULL) instanceData->linearConstraintCoefficients = new LinearConstraintCoefficients() ;
01780         if(numberOfValues == 0) return true;
01781         if((values == 0 ) ||
01782            (valuesBegin < 0 || (valuesEnd - valuesBegin + 1) != numberOfValues) ||
01783            (indexes == 0) ||
01784            (indexesBegin < 0 || (indexesEnd - indexesBegin + 1) != numberOfValues) ||
01785            (starts == 0 ) ||
01786            (startsBegin < 0  || startsBegin >= startsEnd)) return false;
01787         instanceData->linearConstraintCoefficients->numberOfValues = numberOfValues;
01788         int i = 0;
01789         //starts
01790         if(instanceData->linearConstraintCoefficients->start == NULL) instanceData->linearConstraintCoefficients->start = new IntVector();
01791         if(startsBegin == 0 ){
01792                 instanceData->linearConstraintCoefficients->start->el = starts;
01793         }
01794         else{
01795                 instanceData->linearConstraintCoefficients->start->el = new int[startsEnd - startsBegin + 1];
01796                 int k = 0;
01797                 for(i = startsBegin; i <= startsEnd; i++){
01798                         instanceData->linearConstraintCoefficients->start->el[k] = starts[i];
01799                         k++;
01800                 }
01801         }       
01802         //values
01803         if(instanceData->linearConstraintCoefficients->value == NULL) instanceData->linearConstraintCoefficients->value = new DoubleVector();
01804         if(valuesBegin == 0 ){
01805                 instanceData->linearConstraintCoefficients->value->el = values;
01806         }
01807         else{
01808                 instanceData->linearConstraintCoefficients->value->el = new double[numberOfValues];
01809                 int k = 0;
01810                 for(i = valuesBegin; i <= valuesEnd; i++){
01811                         instanceData->linearConstraintCoefficients->value->el[k] = values[i];
01812                         k++;
01813                 }
01814         }
01815         //indexes
01816         if(instanceData->linearConstraintCoefficients->rowIdx == NULL) instanceData->linearConstraintCoefficients->rowIdx = new IntVector();
01817         if(instanceData->linearConstraintCoefficients->colIdx == NULL) instanceData->linearConstraintCoefficients->colIdx = new IntVector();
01818         if(isColumnMajor){
01819                 if(indexesBegin == 0 ){
01820                         instanceData->linearConstraintCoefficients->rowIdx->el = indexes;
01821                 }
01822                 else{
01823                         instanceData->linearConstraintCoefficients->rowIdx->el = new int[numberOfValues];
01824                         int k = 0;
01825                         for(i = indexesBegin; i <= indexesEnd; i++){
01826                                 instanceData->linearConstraintCoefficients->rowIdx->el[k] = indexes[i];
01827                                 k++;
01828                         }
01829                 }
01830         } 
01831         else{
01832                 if(indexesBegin == 0 ){
01833                         instanceData->linearConstraintCoefficients->colIdx->el = indexes;
01834                 }
01835                 else{
01836                         instanceData->linearConstraintCoefficients->colIdx->el = new int[numberOfValues];
01837                         int k = 0;
01838                         for(i = indexesBegin; i <= indexesEnd; i++){
01839                                 instanceData->linearConstraintCoefficients->colIdx->el[k] = indexes[i];
01840                                 k++;
01841                         }
01842                 }
01843         }
01844         return true;
01845 }//setLinearConstraintCoefficients
01846 
01847 bool OSInstance::setQuadraticTerms(int number, 
01848                 int* rowIndexes, int* varOneIndexes, int* varTwoIndexes, double* coefficients,
01849                 int begin, int end){
01850         if(number < 0) return false;
01851         if(number != (end - begin) + 1) return false;
01852         if(number == 0){
01853                 instanceData->quadraticCoefficients = 0;
01854                 return true;
01855         }
01856         if( ((end - begin + 1) != number) ||
01857            (rowIndexes == 0) ||                    
01858            (varOneIndexes == 0) ||
01859            (varTwoIndexes == 0) ||
01860            (coefficients == 0) ) return false;
01861         instanceData->quadraticCoefficients = new QuadraticCoefficients();
01862         instanceData->quadraticCoefficients->numberOfQuadraticTerms = number;
01863         int i = 0;
01864         instanceData->quadraticCoefficients->qTerm = new QuadraticTerm*[number];
01865         for(i = 0; i < number; i++) instanceData->quadraticCoefficients->qTerm[i] = new QuadraticTerm();
01866         int k = 0;
01867         for(i = begin; i <= end; i++){
01868                         instanceData->quadraticCoefficients->qTerm[k]->idx = rowIndexes[i];
01869                         instanceData->quadraticCoefficients->qTerm[k]->idxOne = varOneIndexes[i];
01870                         instanceData->quadraticCoefficients->qTerm[k]->idxTwo = varTwoIndexes[i];
01871                         instanceData->quadraticCoefficients->qTerm[k]->coef = coefficients[i];
01872                         k++;
01873         }
01874         return true;
01875 }//setQuadraticTerms
01876 
01877 bool OSInstance::setQuadraticTermsInNonlinearExpressions(int numQPTerms, int* rowIndexes, int* varOneIndexes, int* varTwoIndexes, double* coefficients){
01878                 instanceData->nonlinearExpressions->numberOfNonlinearExpressions = numQPTerms;
01879                 instanceData->nonlinearExpressions->nl = new Nl*[ numQPTerms ];
01880                 // define the vectors
01881                 OSnLNode *nlNodePoint;
01882                 OSnLNodeVariable *nlNodeVariablePoint;
01883                 std::vector<OSnLNode*> nlNodeVec;
01884                 //
01885                 //
01886                 int i;
01887                 for(i = 0; i < numQPTerms; i++){
01888                         instanceData->nonlinearExpressions->nl[ i] = new Nl();
01889                         instanceData->nonlinearExpressions->nl[ i]->idx = rowIndexes[ i];
01890                         instanceData->nonlinearExpressions->nl[ i]->osExpressionTree = new OSExpressionTree();
01891                 // create a variable nl node for x0
01892                 nlNodeVariablePoint = new OSnLNodeVariable();
01893                 nlNodeVariablePoint->idx = varOneIndexes[ i];
01894                 // give this variable the coefficient
01895                 nlNodeVariablePoint->coef = coefficients[ i];
01896                 nlNodeVec.push_back( nlNodeVariablePoint);
01897                 // create the nl node for x1
01898                 nlNodeVariablePoint = new OSnLNodeVariable();
01899                 nlNodeVariablePoint->idx = varTwoIndexes[ i];
01900                 nlNodeVec.push_back( nlNodeVariablePoint);
01901                 // create the nl node for *
01902                 nlNodePoint = new OSnLNodeTimes();
01903                 nlNodeVec.push_back( nlNodePoint);
01904                 // the vectors are in postfix format
01905                 // now the expression tree
01906                 instanceData->nonlinearExpressions->nl[ i]->osExpressionTree->m_treeRoot =
01907                         nlNodeVec[ 0]->createExpressionTreeFromPostfix( nlNodeVec);
01908                 nlNodeVec.clear();
01909                 }
01910         return true;
01911 }//setQuadraticTermsInNonlinearExpressions
01912 
01913 bool OSInstance::initializeNonLinearStructures( ){
01914         std::map<int, OSExpressionTree*>::iterator posMapExpTree;
01915         if( m_bNonLinearStructuresInitialized == true) return true;
01916         if( m_bProcessVariables == false) processVariables();
01917         if( m_bProcessObjectives == false) processObjectives();
01918         if( m_bProcessConstraints == false) processConstraints();
01919         m_iVariableNumber = instanceData->variables->numberOfVariables;
01920         m_iConstraintNumber = instanceData->constraints->numberOfConstraints;
01921         m_iObjectiveNumber = instanceData->objectives->numberOfObjectives;
01922         // get all of the expression trees
01923         if( m_bProcessExpressionTrees == false) getAllNonlinearExpressionTrees();
01924         // before proceeding get a copy of the map of the Expression Trees
01925         if( m_bDuplicateExpressionTreesMap == false) duplicateExpressionTreesMap();
01926         // now create all of the variable maps for each expression tree
01927         for(posMapExpTree = m_mapExpressionTreesMod.begin(); posMapExpTree != m_mapExpressionTreesMod.end(); ++posMapExpTree){  
01928                 (posMapExpTree->second)->getVariableIndiciesMap() ;
01929         }
01930         // add the quadratic terms if necessary
01931         if(getNumberOfQuadraticTerms() > 0) addQTermsToExressionTree();
01932         // now get the map of all nonlinear variables
01933         getAllNonlinearVariablesIndexMap( );
01934         getDenseObjectiveCoefficients();
01935         m_mdConstraintFunctionValues = new double[ this->instanceData->constraints->numberOfConstraints];
01936         m_mdObjectiveFunctionValues = new double[ this->instanceData->objectives->numberOfObjectives];
01937         //m_mdObjGradient = new double[ this->instanceData->variables->numberOfVariables];
01938         m_bNonLinearStructuresInitialized = true;
01939         return true;
01940 }
01941 
01942 SparseJacobianMatrix *OSInstance::getJacobianSparsityPattern( ){
01943         // if already called return the sparse Jacobian
01944         // it is important that this method NOT get called twice -- if
01945         // there are linear terms in <linearConstraintCoefficients> that
01946         // also appear in <nonlinearExpressions> then they will keep getting added
01947         // to the modified expession tree with each call to this method
01948         if( m_bSparseJacobianCalculated == true) return m_sparseJacMatrix;
01949         //std::cout << "INSIDE GET JACOBIAN SPARSITY PATTERN" << std::endl;
01950         // determine if we are in column or row major
01951         getLinearConstraintCoefficientMajor();
01952         // make sure the data structures have been inialized
01953         if( m_bNonLinearStructuresInitialized == false) initializeNonLinearStructures( );
01954         try{
01955                 if( m_bColumnMajor == true){
01956                          if( getSparseJacobianFromColumnMajor( ) == false) throw ErrorClass("An error occurred in getSpareJacobianFromColumnMajor");
01957                 }
01958                 else {
01959                         if( getSparseJacobianFromRowMajor( ) == false) throw ErrorClass("An error occurred in getSpareJacobianFromRowMajor");
01960                 }
01961         }
01962         catch(const ErrorClass& eclass){
01963                 throw ErrorClass(  eclass.errormsg); 
01964         }
01965         // now fill in the arrays of the sparseJacMatrix
01966         m_sparseJacMatrix = new SparseJacobianMatrix();
01967         // we point to memory already created so don't 
01968         // destroy during garbage collection
01969         m_sparseJacMatrix->bDeleteArrays = false;
01970         m_sparseJacMatrix->valueSize =  m_iJacValueSize;
01971         m_sparseJacMatrix->starts = m_miJacStart;
01972         m_sparseJacMatrix->conVals = m_miJacNumConTerms;
01973         m_sparseJacMatrix->indexes = m_miJacIndex;
01974         m_sparseJacMatrix->values = m_mdJacValue;
01975         m_bSparseJacobianCalculated = true;
01976         return m_sparseJacMatrix;
01977 }//getJacobianSparsityPatter
01978 
01979 bool OSInstance::addQTermsToExressionTree(){
01980         int i, k, idx;
01981         // get the number of qTerms
01982         int numQTerms = instanceData->quadraticCoefficients->numberOfQuadraticTerms;    
01983         if(numQTerms <= 0 || m_bQTermsAdded == true) return true;
01984         OSnLNodeVariable* nlNodeVariableOne;
01985         OSnLNodeVariable* nlNodeVariableTwo;
01986         OSnLNodeTimes* nlNodeTimes;
01987         OSnLNodePlus* nlNodePlus;
01988         OSExpressionTree* expTree;
01989         getQuadraticTerms();    
01990         std::cout << "PROCESSING QUADRATIC TERMS" << std::endl;
01991         for(i = 0; i < numQTerms; i++){
01992                 idx = m_quadraticTerms->rowIndexes[ i];
01993                 std::cout << "PROCESSING QTERM  = "  << i <<std::endl;
01994                 // see if row idx is in the expression tree
01995                 if( m_mapExpressionTreesMod.find( idx) != m_mapExpressionTreesMod.end() ) {
01996                         // row idx is in the expression tree
01997                         // add the qTerm in row idx  to the expression tree     
01998                         // define two new OSnLVariable nodes, an OSnLnodeTimes, and OSnLnodePlus 
01999                         nlNodeVariableOne = new OSnLNodeVariable();
02000                         nlNodeVariableOne->idx = m_quadraticTerms->varOneIndexes[ i];
02001                         // see if the variable indexed by nlNodeVariableOne->idx is in the expression tree for row idx
02002                         // if not, add to mapVarIdx
02003                         expTree = m_mapExpressionTreesMod[ idx];
02004                         if(  expTree->m_bIndexMapGenerated == false) expTree->getVariableIndiciesMap();
02005                         if( (*expTree->mapVarIdx).find( nlNodeVariableOne->idx) == (*expTree->mapVarIdx).end()  ){
02006                                 // add to map
02007                                 k = (*expTree->mapVarIdx).size();
02008                                 (*expTree->mapVarIdx)[ nlNodeVariableOne->idx] =  k + 1;
02009                                 std::cout << "ADDED THE FOLLOWING VAIRABLE TO THE MAP" <<  nlNodeVariableOne->idx << std::endl;
02010                         }
02011                         nlNodeVariableOne->coef = m_quadraticTerms->coefficients[ i];
02012                         nlNodeVariableTwo = new OSnLNodeVariable();
02013                         nlNodeVariableTwo->idx = m_quadraticTerms->varTwoIndexes[ i];
02014                         // see if the variable indexed by nlNodeVariableTwo->idx is in the expression tree for row idx
02015                         // if not, add to mapVarIdx
02016                         if( (*expTree->mapVarIdx).find( nlNodeVariableTwo->idx) == (*expTree->mapVarIdx).end()  ){
02017                                 // add to map
02018                                 k = (*expTree->mapVarIdx).size();
02019                                 (*expTree->mapVarIdx)[ nlNodeVariableTwo->idx] =  k + 1;
02020                                 std::cout << "ADDED THE FOLLOWING VAIRABLE TO THE MAP" <<  nlNodeVariableTwo->idx << std::endl;
02021                         }
02022                         nlNodeVariableTwo->coef = 1.;
02023                         // now multiply the two new variable nodes together
02024                         nlNodeTimes = new OSnLNodeTimes();
02025                         nlNodeTimes->m_mChildren[ 0] = nlNodeVariableOne;
02026                         nlNodeTimes->m_mChildren[ 1] = nlNodeVariableTwo;               
02027                         // now add the result to the expression tree
02028                         nlNodePlus = new OSnLNodePlus();
02029                         nlNodePlus->m_mChildren[ 0] = expTree->m_treeRoot;
02030                         nlNodePlus->m_mChildren[ 1] = nlNodeTimes;
02031                         //expTree = new OSExpressionTree();
02032                         expTree->m_treeRoot = nlNodePlus ;
02033                         // get rid of old variable map
02034                         if(expTree->m_bIndexMapGenerated == true){
02035                                 delete expTree->mapVarIdx;
02036                                 expTree->mapVarIdx = NULL;
02037                                 expTree->m_bIndexMapGenerated = false;
02038                         }       
02039                         //expTree->m_bIndexMapGenerated = false;
02040                         //m_mapExpressionTreesMod[ idx ]  = expTree;    
02041                         //expTree->mapVarIdx = m_mapExpressionTreesMod[ idx]->mapVarIdx;
02042                 }
02043                 else{ 
02044                         // create the quadratic expression to add to the expression tree
02045                         nlNodeVariableOne = new OSnLNodeVariable();
02046                         nlNodeVariableOne->idx = m_quadraticTerms->varOneIndexes[ i];
02047                         nlNodeVariableOne->coef = m_quadraticTerms->coefficients[ i];
02048                         nlNodeVariableTwo = new OSnLNodeVariable();
02049                         nlNodeVariableTwo->idx = m_quadraticTerms->varTwoIndexes[ i];
02050                         nlNodeVariableTwo->coef = 1.;
02051                         // now multiply the two new variable nodes together
02052                         nlNodeTimes = new OSnLNodeTimes();
02053                         nlNodeTimes->m_mChildren[ 0] = nlNodeVariableOne;
02054                         nlNodeTimes->m_mChildren[ 1] = nlNodeVariableTwo;
02055                         // create a new expression tree corresponding to row idx.
02056                         expTree = new OSExpressionTree();                                               
02057                         expTree->m_treeRoot = nlNodeTimes ;
02058                         expTree->mapVarIdx = expTree->getVariableIndiciesMap();         
02059                         m_mapExpressionTreesMod[ idx ]  = expTree;
02060                         if(idx < 0){
02061                                 m_iObjectiveNumberNonlinear++;
02062                                 m_bProcessExpressionTrees = true;
02063                         }
02064                         else{
02065                                 m_iConstraintNumberNonlinear++;
02066                                 m_bProcessExpressionTrees = true;
02067                         }
02068                         std::cout << "NUMBER OF EXPRESSION TREES = "  << m_mapExpressionTreesMod.size() <<std::endl;
02069                         std::cout << "NUMBER OF NONLINEAR OBJECTIVES = "  << getNumberOfNonlinearObjectives() <<std::endl;
02070                 } 
02071                 // if there were no nonlinear terms make this the expression tree
02072                 if(m_iNonlinearExpressionNumber <= 0) m_mapExpressionTrees = m_mapExpressionTreesMod;
02073                 m_bQTermsAdded =true;
02074         }
02075         return true;
02076 }
02077 
02078 double OSInstance::calculateFunctionValue(int idx, double *x, bool new_x){
02079         try{
02080 
02081                 int i, j;
02082                 double dvalue = 0;
02083                 if( m_binitForAlgDiff == false) initForAlgDiff();
02084                 if( m_bSparseJacobianCalculated == false) getJacobianSparsityPattern();
02085                 if(idx >= 0){ // we have a constraint
02086                         // make sure the index idx is valid
02087                         if( getConstraintNumber() <= idx  ) throw 
02088                         ErrorClass("constraint index not valid in OSInstance::calculateFunctionValue");
02089                         if( new_x == false) return *(m_mdConstraintFunctionValues + idx);
02090                         // get the nonlinear part
02091                         if( m_mapExpressionTreesMod.find( idx) != m_mapExpressionTreesMod.end() ){
02092                                 dvalue = m_mapExpressionTreesMod[ idx]->calculateFunction( x,  new_x);
02093                                 //dvalue = vdFunVals[ idx + 1];
02094                         }
02095                         // now the linear part
02096                         // be careful, loop over only the constant terms in sparseJacMatrix
02097                         i = m_sparseJacMatrix->starts[ idx];
02098                         j = m_sparseJacMatrix->starts[ idx + 1 ];
02099                         //while ( i <  j &&  (i - m_sparseJacMatrix->starts[ idx])  < m_sparseJacMatrix->conVals[ idx] ){
02100                         while ( (i - m_sparseJacMatrix->starts[ idx])  < m_sparseJacMatrix->conVals[ idx] ){
02101                                 //std::cout << "m_sparseJacMatrix->values[ i] " << m_sparseJacMatrix->values[ i] << std::endl;
02102                                 //std::cout << "m_sparseJacMatrix->indexes[ i] " << m_sparseJacMatrix->indexes[ i] << std::endl;
02103                                 dvalue += m_sparseJacMatrix->values[ i]*x[ m_sparseJacMatrix->indexes[ i] ];
02104                                 i++;
02105                         }       
02106                         // add in the constraint function constant
02107                         dvalue += m_mdConstraintConstants[ idx ];
02108                         return dvalue;
02109                 }
02110                 else{ // we have an objective function
02111                         // make sure the index idx is valid
02112                         if( getObjectiveNumber() <= ( abs( idx) - 1) ) throw 
02113                         ErrorClass("objective function index not valid in OSInstance::calculateFunctionValue");
02114                         if( new_x == false) return *(m_mdObjectiveFunctionValues + ( abs( idx) - 1));
02115                         // get the nonlinear part
02116                         if( m_mapExpressionTreesMod.find( idx) != m_mapExpressionTreesMod.end() ){
02117                                 dvalue = m_mapExpressionTreesMod[ idx]->calculateFunction( x,  new_x);
02118                         }
02119                         // get linear part
02120                         SparseVector **objCoef = getObjectiveCoefficients();
02121                         SparseVector *obj = objCoef[ abs( idx) - 1];
02122                         for(i = 0; i < obj->number; i++){
02123                                 dvalue += x[ obj->indexes[i]]*(obj->values[ i]);
02124                         }
02125                         // add in the objective function constant
02126                         dvalue += m_mdObjectiveConstants[ abs( idx) - 1 ];
02127                         // get the coefficients for objective function idx
02128                         *(m_mdObjectiveFunctionValues + ( abs( idx) - 1)) = dvalue;
02129                         return *(m_mdObjectiveFunctionValues + ( abs( idx) - 1));
02130                 }
02131         }
02132         catch(const ErrorClass& eclass){
02133                 throw ErrorClass( eclass.errormsg);
02134         } 
02135 }//calculateFunctionValue
02136 
02137 
02138 double *OSInstance::calculateAllConstraintFunctionValues( double* x, double *objLambda, double *conLambda,
02139         bool new_x, int highestOrder){  
02140         try{
02141                 if( new_x == true || (highestOrder > m_iHighestOrderEvaluated)  ) 
02142                         getIterateResults(x, objLambda, conLambda, new_x,  highestOrder);
02143         }
02144         catch(const ErrorClass& eclass){
02145                 throw ErrorClass( eclass.errormsg);
02146         } 
02147         return m_mdConstraintFunctionValues;
02148 }//calculateAllConstraintFunctionValues
02149 
02150 
02151 double *OSInstance::calculateAllConstraintFunctionValues(double* x, bool new_x){
02152         try{
02153                 m_iHighestOrderEvaluated = -1;
02154                 if( new_x == false) return m_mdConstraintFunctionValues;
02155                 int idx, numConstraints;
02156                 numConstraints = getConstraintNumber();
02157                 // loop over all constraints
02158                 for(idx = 0; idx < numConstraints; idx++){
02159                         m_mdConstraintFunctionValues[ idx]  = calculateFunctionValue(idx, x, new_x);    
02160                 }
02161                 
02162         }
02163         catch(const ErrorClass& eclass){
02164                 throw ErrorClass( eclass.errormsg);
02165         } 
02166         return m_mdConstraintFunctionValues;    
02167 }//end calculateAllConstraintFunctionValues
02168 
02169 
02170 double *OSInstance::calculateAllObjectiveFunctionValues( double* x, double *objLambda, double *conLambda,
02171         bool new_x, int highestOrder){  
02172         try{
02173                 if( new_x == true || (highestOrder > m_iHighestOrderEvaluated)  ) 
02174                         getIterateResults(x, objLambda, conLambda, new_x,  highestOrder);
02175         }
02176         catch(const ErrorClass& eclass){
02177                 throw ErrorClass( eclass.errormsg);
02178         } 
02179         return m_mdObjectiveFunctionValues;
02180 }//calculateAllConstraintFunctionValues
02181 
02182 
02183 double *OSInstance::calculateAllObjectiveFunctionValues( double* x, bool new_x){        
02184         try{
02185                 m_iHighestOrderEvaluated = -1;
02186                 if( new_x == false) return m_mdObjectiveFunctionValues;
02187                 int idx, numObjectives;
02188                 numObjectives = getObjectiveNumber();
02189                 // loop over all constraints
02190                 for(idx = 0; idx < numObjectives; idx++){
02191                         m_mdObjectiveFunctionValues[ idx]  = calculateFunctionValue(-idx -1, x, new_x); 
02192                 }
02193         }
02194         catch(const ErrorClass& eclass){
02195                 throw ErrorClass( eclass.errormsg);
02196         } 
02197         return m_mdObjectiveFunctionValues;
02198 }//calculateAllObjectiveFunctionValues
02199 
02200 
02201 SparseJacobianMatrix *OSInstance::calculateAllConstraintFunctionGradients(double* x, double *objLambda, double *conLambda,
02202                 bool new_x, int highestOrder){
02203         try{
02204                 if(highestOrder < 1 ) throw ErrorClass("When calling calculateAllConstraintFunctionGradients highestOrder should be 1 or 2");
02205                 if( new_x == true || (highestOrder > m_iHighestOrderEvaluated)  ) 
02206                         getIterateResults(x, objLambda, conLambda,  new_x,  highestOrder);
02207         }//end try
02208         catch(const ErrorClass& eclass){
02209                 throw ErrorClass( eclass.errormsg);
02210         } 
02211         return m_sparseJacMatrix;
02212 }//calculateAllConstraintFunctionGradients      
02213 
02214 
02215 
02216 SparseVector *OSInstance::calculateConstraintFunctionGradient(double* x, double *objLambda, double *conLambda,
02217                 int idx, bool new_x, int highestOrder){
02218         try{
02219                 if(highestOrder < 1 ) throw ErrorClass("When calling calculateConstraintFunctionGradient highestOrder should be 1 or 2");
02220                 if(idx > instanceData->variables->numberOfVariables ) 
02221                         throw ErrorClass("invalid index passed to calculateConstraintFunctionGrad");
02222                 SparseVector *sp;
02223                 sp = new SparseVector();
02224                 sp->bDeleteArrays = true;
02225                 int i;
02226                 if( new_x == true || (highestOrder > m_iHighestOrderEvaluated)  ) 
02227                         getIterateResults(x, objLambda, conLambda,  new_x,  highestOrder);
02228                 sp->number = m_miJacStart[ idx + 1] - m_miJacStart[ idx];
02229                 sp->values = new double[ sp->number];
02230                 sp->indexes = new int[ sp->number];
02231                 for(i = 0; i < sp->number; i++){
02232                         sp->values[ i] = m_mdJacValue[ m_miJacStart[ idx] +  i];
02233                         sp->indexes[ i] = m_miJacIndex[ m_miJacStart[ idx] +  i];
02234                 }
02235                 return sp;
02236         }
02237         catch(const ErrorClass& eclass){
02238                 throw ErrorClass( eclass.errormsg);
02239         } 
02240 }//calculateConstraintFunctionGradient
02241 
02242 
02243 SparseVector *OSInstance::calculateConstraintFunctionGradient(double* x, int idx, bool new_x){
02244         try{
02245                 if(idx > instanceData->variables->numberOfVariables ) 
02246                         throw ErrorClass("invalid index passed to calculateConstraintFunctionGrad");
02247                 SparseVector *sp;
02248                 sp = new SparseVector();
02249                 sp->bDeleteArrays = true;
02250                 int i;
02251                 if( new_x == true || (1 > m_iHighestOrderEvaluated)  ) 
02252                         getIterateResults(x, NULL, NULL,  new_x,  1);
02253                 sp->number = m_miJacStart[ idx + 1] - m_miJacStart[ idx];
02254                 sp->values = new double[ sp->number];
02255                 sp->indexes = new int[ sp->number];
02256                 for(i = 0; i < sp->number; i++){
02257                         sp->values[ i] = m_mdJacValue[ m_miJacStart[ idx] +  i];
02258                         sp->indexes[ i] = m_miJacIndex[ m_miJacStart[ idx] + i];
02259                 }
02260                 return sp;
02261         }
02262         catch(const ErrorClass& eclass){
02263                 throw ErrorClass( eclass.errormsg);
02264         } 
02265 }//calculateConstraintFunctionGradient
02266 
02267 
02268 double **OSInstance::calculateAllObjectiveFunctionGradients(double* x, double *objLambda, double *conLambda,
02269                 bool new_x, int highestOrder){
02270         try{
02271                 if(highestOrder < 1 ) throw ErrorClass("When calling calculateAllObjectiveFunctionGradients highestOrder should be 1 or 2");
02272                 if( new_x == true || (highestOrder > m_iHighestOrderEvaluated)  ) 
02273                         getIterateResults(x, objLambda, conLambda,  new_x,  highestOrder);
02274         }
02275         catch(const ErrorClass& eclass){
02276                 throw ErrorClass( eclass.errormsg);
02277         } 
02278         return m_mmdObjGradient;
02279 }// calculateAllObjectiveFunctionGradients
02280 
02281 double *OSInstance::calculateObjectiveFunctionGradient(double* x, double *objLambda, double *conLambda,
02282                 int objIdx, bool new_x, int highestOrder){
02283         try{
02284                 if(highestOrder < 1 ) throw ErrorClass("When calling calculateObjectiveFunctionGradient highestOrder should be 1 or 2");
02285                 if( new_x == true || (highestOrder > m_iHighestOrderEvaluated)  ) 
02286                         getIterateResults(x, objLambda, conLambda,  new_x,  highestOrder);
02287         }
02288         catch(const ErrorClass& eclass){
02289                 throw ErrorClass( eclass.errormsg);
02290         } 
02291         return m_mmdObjGradient[abs( objIdx) - 1];
02292 }//calculateObjectiveFunctionGradient
02293 
02294 
02295 double *OSInstance::calculateObjectiveFunctionGradient(double* x, int objIdx, bool new_x){
02296         try{
02297                 if( new_x == true && ( m_iHighestOrderEvaluated < 1)  ) 
02298                         getIterateResults(x, NULL, NULL, new_x,  1);
02299         }
02300         catch(const ErrorClass& eclass){
02301                 throw ErrorClass( eclass.errormsg);
02302         } 
02303         return m_mmdObjGradient[abs( objIdx) - 1];
02304 }//calculateObjectiveFunctionGradient   
02305 
02306 SparseHessianMatrix *OSInstance::calculateLagrangianHessian( double* x, double *objLambda, double *conLambda,
02307         bool new_x, int highestOrder){
02308         try{
02309                 if(highestOrder != 2 ) throw ErrorClass("When calling calculateLagrangianHessian highestOrder should be 2");
02310                 if( new_x == true || (highestOrder > m_iHighestOrderEvaluated)  ) {
02311                         //std::cout  << "CALL getIterateResults() FROM calculateLagrangianHessain" << std::endl;
02312                         getIterateResults(x, objLambda, conLambda,  new_x,  highestOrder);
02313                 }
02314         }
02315         catch(const ErrorClass& eclass){
02316                 throw ErrorClass( eclass.errormsg);
02317         } 
02318         return m_LagrangianSparseHessian;
02319 }//calculateLagrangianHessian
02320 
02321 SparseHessianMatrix *OSInstance::calculateHessian(double* x, int idx, bool new_x){
02322         try{
02323                 if(idx > instanceData->variables->numberOfVariables ) 
02324                         throw ErrorClass("invalid index passed to calculateHessian");
02325                 double *objLambda = new double[ getObjectiveNumber() ];
02326                 double *conLambda = new double[ getConstraintNumber() ];
02327                 //std::cout << "NUMBER OF OBJECTIVES = " << getObjectiveNumber() << std::endl;
02328                 //std::cout << "NUMBER OF CONSTRAINTS = " << getConstraintNumber() << std::endl;
02329                 int i;
02330                 // initialize all to zero
02331                 for(i = 0; i < getObjectiveNumber(); i++){
02332                         objLambda[ i] = 0.0;
02333                 }
02334                 for(i = 0; i < getConstraintNumber(); i++){
02335                         conLambda[ i] = 0.0;
02336                 }
02337                 // see if we have the index of an objective function or a constraint
02338                 // and more corresponding component 1.0
02339                 if(idx < 0){
02340                         objLambda[ abs(idx) - 1] = 1.0;
02341                 }
02342                 else{
02343                         conLambda[ idx] = 1.0;
02344                 }
02345                 if( new_x == true || (2 > m_iHighestOrderEvaluated)  ) {
02346                         getIterateResults(x, objLambda, conLambda, new_x,  2);
02347                         std::cout  << "CALL getIterateResults() FROM calculateHessian" << std::endl;
02348                 }
02349                 delete[] objLambda;
02350                 delete[] conLambda;
02351         }
02352         catch(const ErrorClass& eclass){
02353                 throw ErrorClass( eclass.errormsg);
02354         } 
02355         return m_LagrangianSparseHessian;
02356 }//calculateHessian                     
02357 
02358 
02359 
02360 bool OSInstance::getSparseJacobianFromColumnMajor( ){
02361         // we assume column major matrix
02362         if( m_bColumnMajor == false) return false;
02363         int iNumRowStarts = getConstraintNumber() + 1;  
02364         int i,j, iTemp;
02365         int iNumVariableStarts = getVariableNumber() ;
02366         int *start = NULL;
02367         int *index = NULL;
02368         double *value = NULL;
02369         if(this->instanceData->linearConstraintCoefficients->numberOfValues > 0){
02370                 start = this->instanceData->linearConstraintCoefficients->start->el;
02371                 index = this->instanceData->linearConstraintCoefficients->rowIdx->el;
02372                 value = this->instanceData->linearConstraintCoefficients->value->el;
02373         }
02374         m_miJacStart = new int[ iNumRowStarts];
02375         m_miJacNumConTerms = new int[ getConstraintNumber()];
02376         OSnLNodePlus *nlNodePlus;
02377         OSnLNodeVariable *nlNodeVariable;
02378         OSExpressionTree *expTree = NULL;
02379         // now initialize starts and variable index maps 
02380         for ( i = 0; i < iNumRowStarts; i++){                   
02381                 m_miJacStart [ i ] = 0;
02382                 // map the variables  in the nonlinear rows
02383                 if( m_mapExpressionTreesMod.find( i) != m_mapExpressionTreesMod.end() ){
02384                         // the following is equivalent to  m_treeRoot->getVariableIndexMap( i);
02385                         m_mapExpressionTreesMod[ i]->getVariableIndiciesMap();
02386                         
02387                 }
02388         }
02389         // only execute the following code if there are linear constraint coefficients
02390         if(this->instanceData->linearConstraintCoefficients->numberOfValues > 0){
02391                 // i is indexing columns (variables) and j is indexing row numbers 
02392                 for (i = 0; i < iNumVariableStarts; i++){       
02393                         for (j = start[i]; j < start[ i + 1 ]; j++){
02394                                 // index[ j] is a row index, we have just found an occurance of row index[j]
02395                                 // therefore we increase by 1 (or push back) the start of the row indexed by index[j] + 1, 
02396                                 // i.e. the start of the next row
02397                                 // check to see if variable i is linear/constant in the row index[ j] 
02398                                 // if so, increment m_miJacStart[ index[j] + 1]
02399                                 //
02400                                 if( (m_mapExpressionTreesMod.find( index[ j]) != m_mapExpressionTreesMod.end() ) &&
02401                                         ( (*m_mapExpressionTreesMod[ index[ j]]->mapVarIdx).find( i) != (*m_mapExpressionTreesMod[ index[ j]]->mapVarIdx).end()) ){
02402                                         // variable i is appears in the expression tree for row index[ j]
02403                                         // add the coefficient corresponding to variable i in row index[ j] to the expression tree      
02404                                         // define a new OSnLVariable and OSnLnodePlus 
02405                                         expTree = m_mapExpressionTreesMod[ index[j]  ];
02406                                         nlNodeVariable = new OSnLNodeVariable();
02407                                         nlNodeVariable->coef = value[ j];
02408                                         nlNodeVariable->idx = i;
02409                                         nlNodePlus = new OSnLNodePlus();
02410                                         nlNodePlus->m_mChildren[ 0] = m_mapExpressionTreesMod[ index[ j] ]->m_treeRoot;
02411                                         nlNodePlus->m_mChildren[ 1] = nlNodeVariable;
02412                                         //expTree = new OSExpressionTree();
02413                                         expTree->m_treeRoot = nlNodePlus ;      
02414                                         //expTree->mapVarIdx = m_mapExpressionTreesMod[ index[ j]]->mapVarIdx;
02415                                         //m_mapExpressionTreesMod[ index[ j] ]  = expTree;
02416                                         //std::cout << m_mapExpressionTreesMod[ index[ j] ]->m_treeRoot->getNonlinearExpressionInXML() << std::endl;    
02417                                         //std::cout << m_mapExpressionTrees[ index[ j] ]->m_treeRoot->getNonlinearExpressionInXML() << std::endl;
02418                                 }
02419                                 else{ 
02420                                         m_miJacStart[ index[j] + 1] ++;
02421                                 }                               
02422                         }
02423                 }
02424         }
02425         // at this point, m_miJacStart[ i] holds the number of columns with a linear/constant nonzero in row i - 1
02426         // we are not done with the start indicies, if we are here, and we
02427         // knew the correct starting point of row i -1, the correct starting point
02428         // for row i is m_miJacStart[i] + m_miJacStart [i - 1]
02429         m_miJacStart[0] = 0;
02430         for (i = 1; i < iNumRowStarts; i++ ){
02431                 m_miJacNumConTerms[ i - 1] = m_miJacStart[i];
02432                 if( m_mapExpressionTreesMod.find( i - 1) != m_mapExpressionTreesMod.end() ){
02433                         m_miJacStart[i] += (m_miJacStart[i - 1] + (*m_mapExpressionTreesMod[ i - 1]->mapVarIdx).size() );
02434                 }
02435                 else{
02436                         m_miJacStart[i] += m_miJacStart[i - 1];
02437                 }       
02438         }
02439         // dimension miIndex and mdValue here
02440         m_iJacValueSize =       m_miJacStart[ iNumRowStarts - 1];
02441         m_miJacIndex = new int[  m_iJacValueSize];
02442         m_mdJacValue = new double[ m_iJacValueSize ];
02443         // now get the values of the constant terms if there are any
02444         if(this->instanceData->linearConstraintCoefficients->numberOfValues > 0){
02445                 // loop over variables  
02446                 for (i = 0; i < iNumVariableStarts; i++){
02447                         // get row indices and values of the A matrix
02448                         // kipp -- should we have a check to see if start[i+1] > start[i]
02449                         for (j = start[i]; j < start[ i + 1 ]; j++){
02450                                 // store this variable index in every row where the variable appears
02451                                 // however, don't store this as constant term if it appears in mapVarIdx
02452                                 if( (m_mapExpressionTreesMod.find( index[ j]) == m_mapExpressionTreesMod.end() ) ||
02453                                         ( (*m_mapExpressionTreesMod[ index[ j]]->mapVarIdx).find( i) == (*m_mapExpressionTreesMod[ index[ j]]->mapVarIdx).end()) ){
02454                                         iTemp = m_miJacStart[ index[j]];
02455                                         m_miJacIndex[ iTemp] = i;
02456                                         m_mdJacValue[ iTemp] = value[j];
02457                                         m_miJacStart[ index[j]]++;      
02458                                 }               
02459                         }                       
02460                 } 
02461         }
02462         //
02463         std::map<int, int>::iterator posVarIdx;
02464         // m_miJacStart[ i] is now equal to the correct m_miJacStart[ i] + m_miJacNumConTerms[ i], so readjust
02465         for (i = 0; i < iNumRowStarts - 1; i++ ){
02466                 m_miJacStart[ i] = m_miJacStart [ i] - m_miJacNumConTerms[ i] ; 
02467                 iTemp = m_miJacStart[ i] + m_miJacNumConTerms[ i];
02468                 // if the row is in the list of expression trees read in idicies and values
02469                 if( m_mapExpressionTreesMod.find( i) != m_mapExpressionTreesMod.end() ){
02470                         for(posVarIdx = (*m_mapExpressionTreesMod[ i]->mapVarIdx).begin(); posVarIdx 
02471                         != (*m_mapExpressionTreesMod[ i]->mapVarIdx).end(); ++posVarIdx){
02472                                 m_miJacIndex[ iTemp] = posVarIdx->first;
02473                                 m_mdJacValue[ iTemp] = 0;
02474                                 iTemp++;
02475                         }
02476                 }
02477         }
02478         #ifdef DEBUG
02479         std::cout << "HERE ARE ROW STARTS:" << std::endl;
02480         for (i = 0; i < iNumRowStarts; i++ ){
02481                 std::cout <<  m_miJacStart[ i] << "  "; 
02482         }
02483         std::cout << std::endl << std::endl;
02484         std::cout << "HERE ARE VARIABLE INDICIES:" << std::endl;
02485         for (i = 0; i < m_miJacStart[ iNumRowStarts - 1]; i++ ){
02486                 std::cout <<  m_miJacIndex[ i] << "  "; 
02487         }
02488         std::cout << std::endl << std::endl;
02489         std::cout << "HERE ARE VALUES:" << std::endl;
02490         for (i = 0; i < m_miJacStart[ iNumRowStarts - 1]; i++ ){
02491                 std::cout <<  m_mdJacValue[ i] << "  "; 
02492         }
02493         std::cout << std::endl << std::endl;
02494 
02495         std::cout << "HERE ARE NUMBER OF CONSTANT TERMS:" << std::endl;
02496         for (i = 0; i < iNumRowStarts - 1; i++ ){
02497                 std::cout <<  m_miJacNumConTerms[ i ] << "  ";  
02498         }
02499         std::cout << std::endl << std::endl;
02500         #endif
02501         return true;
02502 }//getSparseJacobianFromColumnMajor
02503 
02504 
02505 
02506 bool OSInstance::getSparseJacobianFromRowMajor( ){
02507         // we assume row major matrix
02508         if( m_bColumnMajor == true) return false;
02509         int iNumJacRowStarts = getConstraintNumber() + 1;
02510         std::map<int, int>::iterator posVarIdx; 
02511         int i,j, k;
02512         int *start = NULL;
02513         int *index = NULL;
02514         double *value = NULL;
02515         if(this->instanceData->linearConstraintCoefficients->numberOfValues > 0){
02516                 start = this->instanceData->linearConstraintCoefficients->start->el;
02517                 index = this->instanceData->linearConstraintCoefficients->colIdx->el;
02518                 value = this->instanceData->linearConstraintCoefficients->value->el;
02519         }
02520         m_miJacStart = new int[ iNumJacRowStarts];
02521         m_miJacNumConTerms = new int[ getConstraintNumber()];
02522         OSnLNodePlus *nlNodePlus;
02523         OSnLNodeVariable *nlNodeVariable;
02524         //OSExpressionTree *expTree = NULL;
02525         // now initialize starts and variable index maps 
02526         for ( i = 0; i < iNumJacRowStarts; i++){                        
02527                 m_miJacStart [ i ] = 0;
02528                 // map the variables  in the nonlinear rows
02529                 if( m_mapExpressionTreesMod.find( i) != m_mapExpressionTreesMod.end() ){
02530                         // the following is equivalent to  m_treeRoot->getVariableIndexMap( i);
02531                         m_mapExpressionTreesMod[ i]->getVariableIndiciesMap();
02532                         
02533                 }
02534         }
02535         int loopLimit =  getConstraintNumber();
02536         // only execute the following code if there are linear constraint coefficients
02537         // determine the number of terms in constraint with constant partial derivative
02538         if(this->instanceData->linearConstraintCoefficients->numberOfValues > 0){
02539                 // i is indexing rows (constrains) and j is indexing column numbers 
02540                 for (i = 0; i < loopLimit; i++){
02541                         m_miJacNumConTerms[ i] = 0;
02542                         for (j = start[i]; j < start[ i + 1 ]; j++){
02543                                 // determine if variable index[j] appears in the Expression Tree for row i
02544                                 // if we pass if test below then variable i is in the expresssion tree and we add
02545                                 // the linear term to the expession tree
02546                                 if( (m_mapExpressionTreesMod.find( i) != m_mapExpressionTreesMod.end() ) &&
02547                                         ( (*m_mapExpressionTreesMod[ i]->mapVarIdx).find( index[ j]) != (*m_mapExpressionTreesMod[ i]->mapVarIdx).end()) ){
02548                                         // variable index[ j] appears in the expression tree for row i
02549                                         // add the coefficient corresponding to variable index[j] in row i to the expression tree       
02550                                         // define a new OSnLVariable and OSnLnodePlus 
02551                                         nlNodeVariable = new OSnLNodeVariable();
02552                                         nlNodeVariable->coef = value[ j];
02553                                         nlNodeVariable->idx = index[ j];
02554                                         nlNodePlus = new OSnLNodePlus();
02555                                         nlNodePlus->m_mChildren[ 0] = m_mapExpressionTreesMod[ i ]->m_treeRoot;
02556                                         nlNodePlus->m_mChildren[ 1] = nlNodeVariable;
02557                                         //expTree = new OSExpressionTree();
02558                                         //expTree->m_treeRoot = nlNodePlus ;
02559                                         //expTree->mapVarIdx = m_mapExpressionTreesMod[ i]->mapVarIdx;
02560                                         //m_mapExpressionTreesMod[ i ]  = expTree;      
02561                                         m_mapExpressionTreesMod[ i ]->m_treeRoot = nlNodePlus;
02562                                 }
02563                                 else{ 
02564                                         //the partial derivative of variable j in row i is constant
02565                                         m_miJacNumConTerms[ i]++;
02566                                 }                               
02567                         }
02568                 }
02569         }
02570         //
02571         m_miJacStart[0] = 0;
02572         for (i = 1; i < iNumJacRowStarts; i++ ){
02573                 if( m_mapExpressionTreesMod.find( i - 1) != m_mapExpressionTreesMod.end() ){
02574                         m_miJacStart[i] = m_miJacStart[i - 1] + (m_miJacNumConTerms[ i - 1] + (*m_mapExpressionTreesMod[ i - 1]->mapVarIdx).size() );
02575                 }
02576                 else{
02577                         m_miJacStart[i] = m_miJacStart[i - 1] + m_miJacNumConTerms[ i - 1];
02578                 }       
02579         }
02580         // we know how many constant terms and size of arrays
02581         // dimension miIndex and mdValue here
02582         m_iJacValueSize =       m_miJacStart[ iNumJacRowStarts - 1];
02583         m_miJacIndex = new int[  m_iJacValueSize];
02584         m_mdJacValue = new double[ m_iJacValueSize ];
02585         // now loop again and put in values and indicies
02586         // first put in the constant terms
02587         if(this->instanceData->linearConstraintCoefficients->numberOfValues > 0){
02588                 for (i = 0; i < loopLimit; i++){
02589                         k = 0;
02590                         for (j = start[i]; j < start[ i + 1 ]; j++){
02591                                 if( (m_mapExpressionTreesMod.find( i) == m_mapExpressionTreesMod.end() ) ||
02592                                         ( (*m_mapExpressionTreesMod[ i]->mapVarIdx).find( index[ j]) == (*m_mapExpressionTreesMod[ i]->mapVarIdx).end()) ){
02593                                                 m_miJacIndex[ m_miJacStart[i] + k ] = index[ j];
02594                                                 m_mdJacValue[ m_miJacStart[i] + k ] = value[ j];
02595                                                 k++;
02596                                         }                       
02597                         }
02598                 }
02599         }
02600         // put in terms from the modified nonlinear expression tree
02601         for (i = 0; i < loopLimit; i++ ){
02602                 k = m_miJacStart[i] + m_miJacNumConTerms[ i ];
02603                 // if the row is in the list of expression trees read in idicies and values
02604                 if( m_mapExpressionTreesMod.find( i) != m_mapExpressionTreesMod.end() ){
02605                         for(posVarIdx = (*m_mapExpressionTreesMod[ i]->mapVarIdx).begin(); posVarIdx 
02606                         != (*m_mapExpressionTreesMod[ i]->mapVarIdx).end(); ++posVarIdx){
02607                                 m_miJacIndex[ k] = posVarIdx->first;
02608                                 m_mdJacValue[ k] = 0;
02609                                 k++;
02610                         }
02611                 }
02612         }
02613         #ifdef DEBUG
02614         std::cout << "HERE ARE ROW STARTS:" << std::endl;
02615         for (i = 0; i < iNumJacRowStarts; i++ ){
02616                 std::cout <<  m_miJacStart[ i] << "  "; 
02617         }
02618         std::cout << std::endl << std::endl;
02619         std::cout << "HERE ARE VARIABLE INDICIES:" << std::endl;
02620         for (i = 0; i < m_miJacStart[ iNumJacRowStarts - 1]; i++ ){
02621                 std::cout <<  m_miJacIndex[ i] << "  "; 
02622         }
02623         std::cout << std::endl << std::endl;
02624         std::cout << "HERE ARE VALUES:" << std::endl;
02625         for (i = 0; i < m_miJacStart[ iNumJacRowStarts - 1]; i++ ){
02626                 std::cout <<  m_mdJacValue[ i] << "  "; 
02627         }
02628         std::cout << std::endl << std::endl;
02629 
02630         std::cout << "HERE ARE NUMBER OF CONSTANT TERMS:" << std::endl;
02631         for (i = 0; i < iNumJacRowStarts - 1; i++ ){
02632                 std::cout <<  m_miJacNumConTerms[ i ] << "  ";  
02633         }
02634         std::cout << std::endl << std::endl;
02635         #endif
02636         return true;
02637 }//getSparseJacobianFromRowMajor
02638 
02639 OSExpressionTree* OSInstance::getLagrangianExpTree( ){
02640         if( m_bLagrangianExpTreeCreated == true) return m_LagrangianExpTree;
02641         // we calculate the Lagrangian for all the objectives and constraints
02642         // with nonlinear terms
02643         // first initialize everything for nonlinear work
02644         if(m_bSparseJacobianCalculated == false) getJacobianSparsityPattern( ); 
02645         std::map<int, OSExpressionTree*>::iterator posMapExpTree;
02646         OSnLNodeTimes* nlNodeTimes = NULL;
02647         OSnLNodeVariable* nlNodeVariable = NULL;
02648         OSnLNodeSum* nlNodeSum = NULL;
02649         int numChildren = 0;
02650         int rowIdx;
02651         // create the sum node
02652         nlNodeSum = new OSnLNodeSum();
02653         nlNodeSum->inumberOfChildren = m_mapExpressionTreesMod.size();
02654         nlNodeSum->m_mChildren = new OSnLNode*[ nlNodeSum->inumberOfChildren];
02655         // create and expression tree for the sum node
02656         m_LagrangianExpTree = new OSExpressionTree();
02657         m_LagrangianExpTree->m_treeRoot = nlNodeSum;
02658         // now create the children of the sum node
02659         for(posMapExpTree = m_mapExpressionTreesMod.begin(); posMapExpTree != m_mapExpressionTreesMod.end(); ++posMapExpTree){
02660                 // this variable is the Lagrange multiplier
02661                 nlNodeVariable = new OSnLNodeVariable();
02662                 nlNodeVariable->coef = 1.;
02663                 // get the correct index --
02664                 // for rowIdx = 0, ..., m - 1 set idx = numVar + rowIdx
02665                 rowIdx = posMapExpTree->first;
02666                 if(rowIdx >= 0){
02667                         nlNodeVariable->idx = instanceData->variables->numberOfVariables + rowIdx;
02668                 }
02669                 else{
02670                         nlNodeVariable->idx = instanceData->variables->numberOfVariables + 
02671                         instanceData->constraints->numberOfConstraints + (abs(rowIdx) - 1);
02672                 }
02673                 // now create a times multiply the new variable times the root of the expression tree
02674                 nlNodeTimes = new OSnLNodeTimes();
02675                 nlNodeTimes->m_mChildren[ 0] = nlNodeVariable;
02676                 nlNodeTimes->m_mChildren[ 1] = m_mapExpressionTreesMod[ posMapExpTree->first ]->m_treeRoot;     
02677                 // the times node is the new child
02678                 nlNodeSum->m_mChildren[ numChildren] = nlNodeTimes;
02679                 numChildren++;
02680         }       
02681         // get a variable index map for the expression tree
02682         m_LagrangianExpTree->getVariableIndiciesMap();
02683         // print out the XML for this puppy
02684         //std::cout << "Here comes the Lagrangian Tree" << std::endl;
02685         //std::cout << m_LagrangianExpTree->m_treeRoot->getNonlinearExpressionInXML() << std::endl;
02686         //
02687         m_bLagrangianExpTreeCreated = true;
02688         return m_LagrangianExpTree;
02689 }//getLagrangianExpTree
02690 
02691 std::map<int, int> OSInstance::getAllNonlinearVariablesIndexMap( ){
02692         if(m_bAllNonlinearVariablesIndex == true) return m_mapAllNonlinearVariablesIndex;
02693         //loop over the map of expression tree and get a unique listing of all variables
02694         // put these in the map m_mapAllNonlinearVariablesIndex
02695         std::map<int, OSExpressionTree*>::iterator posMapExpTree;
02696         std::map<int, int>::iterator posVarIdx;
02697         OSExpressionTree *expTree;
02698         for(posMapExpTree = m_mapExpressionTreesMod.begin(); posMapExpTree != m_mapExpressionTreesMod.end(); ++posMapExpTree){
02699                 // get the index map for the expression tree
02700                 expTree = posMapExpTree->second;
02701                 if(expTree->m_bIndexMapGenerated == false)expTree->getVariableIndiciesMap();
02702                 for(posVarIdx = (*expTree->mapVarIdx).begin(); posVarIdx != (*expTree->mapVarIdx).end(); ++posVarIdx){
02703                         if( m_mapAllNonlinearVariablesIndex.find( posVarIdx->first) == m_mapAllNonlinearVariablesIndex.end() ){
02704                         // add the variable to the Lagragian map
02705                         m_mapAllNonlinearVariablesIndex[ posVarIdx->first] = 1;
02706                         }
02707                 }
02708         }
02709         m_miNonLinearVarsReverseMap = new int[m_mapAllNonlinearVariablesIndex.size()];
02710         // now order appropriately
02711         int kount = 0;
02712         //std::cout << "HERE IS THE LAGRANGIANN VARIABLE MAPPING" << std::endl;
02713         for(posVarIdx = m_mapAllNonlinearVariablesIndex.begin(); posVarIdx !=m_mapAllNonlinearVariablesIndex.end(); ++posVarIdx){
02714                 posVarIdx->second = kount;
02715                 m_miNonLinearVarsReverseMap[ kount] = posVarIdx->first;
02716                 kount++;
02717                 #ifdef DEBUG
02718                 std::cout <<  "POSITION FIRST =  "  << posVarIdx->first ;
02719                 std::cout <<  "   POSITION SECOND = "  << posVarIdx->second << std::endl;
02720                 #endif
02721         }
02722         m_iNumberOfNonlinearVariables = kount;
02723         //std::cout <<  "NUMBER OF NONLINEAR VARIABLES =  "  << kount ;
02724         m_bAllNonlinearVariablesIndex = true;
02725         return m_mapAllNonlinearVariablesIndex;
02726 }//getAllNonlinearVariablesIndexMap     
02727 
02728 SparseHessianMatrix* OSInstance::getLagrangianHessianSparsityPattern( ){
02729         // fill in the nonzeros in the sparse Hessian
02730         if( m_bLagrangianSparseHessianCreated == true) return m_LagrangianSparseHessian;
02731         if( m_iNumberOfNonlinearVariables == 0) return NULL;
02732         if( m_binitForAlgDiff == false ) initForAlgDiff();
02733         unsigned int i = 0;
02734         int numNonz = 0;
02735         // Create the CppAD function if necessary
02736         //
02737         std::vector<double> vx;
02738         std::map<int, int>::iterator posMap1, posMap2;  
02739         if( (m_bCppADFunIsCreated == false || m_bCppADMustReTape == true )  && (m_mapExpressionTreesMod.size() > 0) ) {
02740                 for(posMap1 = m_mapAllNonlinearVariablesIndex.begin(); posMap1 != m_mapAllNonlinearVariablesIndex.end(); ++posMap1){
02741                         vx.push_back( 1.0) ;
02742                 }
02743                 createCppADFun( vx);
02744         }
02745         //
02746         // Use CppAD to do a forward sparsity calculation
02747         std::vector<bool> r(m_iNumberOfNonlinearVariables * m_iNumberOfNonlinearVariables);
02748         unsigned int j;
02749         for(i = 0; i < m_iNumberOfNonlinearVariables; i++) { 
02750                 for(j = 0; j < m_iNumberOfNonlinearVariables; j++)
02751                         r[ i * m_iNumberOfNonlinearVariables + j ] = false;
02752                         r[ i * m_iNumberOfNonlinearVariables + i] = true;
02753         }
02754         // compute sparsity pattern for J(x) = F^{(1)} (x)
02755         (*Fad).ForSparseJac(m_iNumberOfNonlinearVariables, r);
02756         //
02757         //now the second derivative
02758         unsigned int m = m_mapExpressionTreesMod.size();
02759         std::vector<bool> e( m);
02760         //Vector s(m);
02761         for(i = 0; i < m; i++) e[i] = true;
02762         std::cout << "Computing Sparse Hessian" << std::endl;
02763         //m_vbLagHessNonz holds the sparsity pattern Lagrangian of the Hessian
02764         m_vbLagHessNonz = (*Fad).RevSparseHes(m_iNumberOfNonlinearVariables, e);
02765         for(i = 0; i < m_iNumberOfNonlinearVariables; i++){
02766                 //std::cout << "Row " << i << "  of Hessian " << std::endl;
02767                 for(j = i; j < m_iNumberOfNonlinearVariables; j++){
02768                         if(m_vbLagHessNonz[ i*m_iNumberOfNonlinearVariables + j]  == true) numNonz++;
02769                         //std::cout << m_vbLagHessNonz[ i*m_iNumberOfNonlinearVariables + j] <<  "  " ;
02770                 }
02771                 //std::cout << std::endl;
02772         }
02773         //std::cout << "Lagrangian Hessian Nonzeros = " << numNonz << std::endl;
02774         i = 0;
02775         // now that we have the dimension create SparseHessianMatrix (upper triangular)
02776         m_LagrangianSparseHessian = new SparseHessianMatrix();
02777         m_LagrangianSparseHessian->bDeleteArrays = true;
02778         m_LagrangianSparseHessian->hessDimension = numNonz;
02779         m_LagrangianSparseHessian->hessRowIdx = new int[m_LagrangianSparseHessian->hessDimension];
02780         m_LagrangianSparseHessian->hessColIdx = new int[m_LagrangianSparseHessian->hessDimension];
02781         m_LagrangianSparseHessian->hessValues = new double[m_LagrangianSparseHessian->hessDimension];
02782         //std::cout << "HESSIAN DIMENSION = " << m_LagrangianSparseHessian->hessDimension << std::endl;
02783         numNonz = 0;
02784         for(posMap1 = m_mapAllNonlinearVariablesIndex.begin(); posMap1 != m_mapAllNonlinearVariablesIndex.end(); ++posMap1){
02785                 //std::cout << "posMap1->first  " << posMap1->first << std::endl;
02786                 j = i;
02787                 for(posMap2 = posMap1; posMap2 != m_mapAllNonlinearVariablesIndex.end(); ++posMap2){
02788                         //std::cout << "posMap2->first  " << posMap2->first << std::endl;
02789                         if(m_vbLagHessNonz[ i*m_iNumberOfNonlinearVariables + j] == true){
02790                                 *(m_LagrangianSparseHessian->hessRowIdx + numNonz) = posMap1->first;
02791                                 *(m_LagrangianSparseHessian->hessColIdx + numNonz) = posMap2->first;
02792                                 numNonz++;
02793                         }
02794                         //std::cout << m_vbLagHessNonz[ i*m_iNumberOfNonlinearVariables + j] <<  "  " << std::endl;
02795                         j++;
02796                 }
02797                 i++;
02798         }
02799         #ifdef DEBUG
02800         std::cout << "HESSIAN SPARSITY PATTERN" << std::endl;
02801         int kj;
02802         for(kj = 0; kj < m_LagrangianSparseHessian->hessDimension; kj++){
02803                 std::cout <<  "Row Index = " << *(m_LagrangianSparseHessian->hessRowIdx + kj) << std::endl;
02804                 std::cout <<  "Column Index = " << *(m_LagrangianSparseHessian->hessColIdx + kj) << std::endl;
02805         }
02806         #endif
02807         //
02808         m_bLagrangianSparseHessianCreated = true;
02809         return m_LagrangianSparseHessian;
02810 }//getLagrangianHessianSparsityPattern
02811 
02812 
02813 void OSInstance::duplicateExpressionTreesMap(){
02814         //std::cout << "I AM IN DUPLICATE EXPRSSION TREES MAP" << std::endl;
02815         // we do this so that we can keep the integrity of the original formulation
02816         if(m_bDuplicateExpressionTreesMap == false){ 
02817                 // first make sure the map was created
02818                 if( m_bProcessExpressionTrees == false) getAllNonlinearExpressionTrees();
02819                 m_mapExpressionTreesMod = m_mapExpressionTrees;
02820                 m_bDuplicateExpressionTreesMap = true;
02821                 return;
02822         }
02823         else{
02824                 return;
02825         }
02826 }//duplicateExpressionTreesMap
02827 
02828 
02829 bool OSInstance::createCppADFun(std::vector<double> vdX){
02830         if(m_bCppADFunIsCreated == true) return true;
02831         //if( m_bNonLinearStructuresInitialized == false) initializeNonLinearStructures( );
02832         if(m_binitForAlgDiff == false) initForAlgDiff();
02833         
02834         //if( m_bAllNonlinearVariablesIndex == false) getAllNonlinearVariablesIndexMap( );
02835         std::map<int, OSExpressionTree*>::iterator posMapExpTree;
02836         unsigned int i;
02837         size_t n = vdX.size();
02838         // declare a CppAD vector and fill it in
02839         CppADvector< AD<double> > vdaX( n );
02840         for(i = 0; i < n; i++){
02841                 vdaX[ i] = vdX[ i];
02842                 //std::cout << "vdX =  " << vdX[ i] << std::endl;
02843         }
02844         // declare the independent variables and start recording
02845         CppAD::Independent( vdaX);
02851         CppAD::vector< AD<double> > m_vFG;      
02852         int kount = 0;
02853         for(posMapExpTree = m_mapExpressionTreesMod.begin(); posMapExpTree != m_mapExpressionTreesMod.end(); ++posMapExpTree){  
02854                 m_vFG.push_back( (posMapExpTree->second)->m_treeRoot->constructCppADTape(&m_mapAllNonlinearVariablesIndex, &vdaX) );
02855                 //std::cout << "PUSHING BACK EXPRESSION NUMBER " << posMapExpTree->first << std::endl;
02856                 if( m_mapCppADFunRangeIndex.find( posMapExpTree->first) == m_mapCppADFunRangeIndex.end() ){
02857                         // count which nonlinear obj/constraint this is
02858                         m_mapCppADFunRangeIndex[ posMapExpTree->first] = kount;
02859                         kount++;
02860                 }
02861         }       
02862         //create the function and stop recording
02863         std::cout << "create the function and stop recording"  << std::endl;
02864         Fad = new CppAD::ADFun<double>(vdaX, m_vFG);
02865         std::cout << "range space dimension =  " << m_vFG.size() << std::endl;
02866         // no forward sweeps done yet
02867         m_iHighestTaylorCoeffOrder = -1;
02868         m_bCppADFunIsCreated = true;
02869         return true;
02870 }//end createCppADFun
02871 
02872 
02873 std::vector<double> OSInstance::forwardAD(int p, std::vector<double> vdX){
02874         try{
02875                 // make sure a CppADFun has been created
02876                 if(m_bCppADFunIsCreated == false) createCppADFun( vdX);
02877                 if(p > (m_iHighestTaylorCoeffOrder + 1) ) throw 
02878                         ErrorClass( "trying to calculate a p order forward when p-1 Taylor coefficient not available");
02879                 // adjust the order of the Taylor coefficient
02880                 m_iHighestTaylorCoeffOrder = p; 
02881                 m_iHighestOrderEvaluated = p;
02882                 //for(int i  = 0; i < vdX.size(); i++){
02883                         //std::cout << "ForwardAD Primal Variables " << i   << " " << vdX[ i] << std::endl;
02884                 //}
02885                 return (*Fad).Forward(p, vdX);
02886         }
02887         catch(const ErrorClass& eclass){
02888                 throw ErrorClass( eclass.errormsg);
02889         }  
02890 }//end forwardAD
02891 
02892 
02893 std::vector<double> OSInstance::reverseAD(int p, std::vector<double> vdlambda){
02894         try{
02895                 if(p == 0) throw 
02896                         ErrorClass( "reverseAD must have p >= 1");
02897                 if(p > (m_iHighestTaylorCoeffOrder + 1) ) throw 
02898                         ErrorClass( "trying to calculate a p order reverse when p-1 Taylor coefficient not available");
02899                 //for(int i  = 0; i < vdlambda.size(); i++){
02900                 //      std::cout << "ReverseAD Multiplier " << i   << " " << vdlambda[ i] << std::endl;
02901                 //}
02902                 m_iHighestOrderEvaluated = p;
02903                 return (*Fad).Reverse(p, vdlambda);
02904         }
02905         catch(const ErrorClass& eclass){
02906                 throw ErrorClass( eclass.errormsg);
02907         }  
02908 }//end forwardAD
02909 
02910 bool OSInstance::getIterateResults( double *x, double *objLambda, double* conMultipliers, 
02911                 bool new_x, int highestOrder){
02912         try{ 
02913                 if( m_binitForAlgDiff == false) initForAlgDiff();
02914                 std::map<int, int>::iterator posVarIndexMap;
02915                 
02916                 if(new_x == true){
02917                         if( m_vdX.size() > 0) m_vdX.clear();
02918                         for(posVarIndexMap = m_mapAllNonlinearVariablesIndex.begin(); posVarIndexMap != m_mapAllNonlinearVariablesIndex.end(); ++posVarIndexMap){
02919                                 m_vdX.push_back( x[ posVarIndexMap->first]) ;
02920                         }
02921                         if( (m_bCppADFunIsCreated == false || m_bCppADMustReTape == true )  && (m_mapExpressionTreesMod.size() > 0) ) {
02922                                 createCppADFun( m_vdX);
02923                         }       
02924                 }       
02925                 switch( highestOrder){          
02926                         case 0: 
02927                                 if(new_x == true || m_iHighestOrderEvaluated < 0){      
02928                                         if(bUseExpTreeForFunEval == true){
02929                                                 calculateAllConstraintFunctionValues( x, new_x);
02930                                                 calculateAllObjectiveFunctionValues( x, new_x);
02931                                         }
02932                                         else{
02933                                                 getZeroOrderResults(x, objLambda, conMultipliers);
02934                                         }
02935 
02936                                 }
02937                                 break;  
02938                         case 1:
02939                                 if(new_x == true || m_iHighestOrderEvaluated < 0)       
02940                                         getZeroOrderResults(x, objLambda, conMultipliers);
02941                                 if(new_x == true || m_iHighestOrderEvaluated < 1)       
02942                                         getFirstOrderResults(x, objLambda, conMultipliers);
02943                                 break;
02944                         case 2: 
02945                                 if(new_x == true || m_iHighestOrderEvaluated < 0)       
02946                                         getZeroOrderResults(x, objLambda, conMultipliers);
02947                                 if(new_x == true || m_iHighestOrderEvaluated < 2)       
02948                                         getSecondOrderResults(x, objLambda, conMultipliers);
02949                                 break;
02950                         default:
02951                                 throw ErrorClass("Derivative should be order 0, 1, or 2");      
02952                 }//end switch
02953                 return true;
02954         }
02955         catch(const ErrorClass& eclass){
02956                 throw ErrorClass( eclass.errormsg);
02957         }  
02958 }//end getIterateResults
02959 
02960 
02961 bool OSInstance::getZeroOrderResults(double *x, double *objLambda, double *conMultipliers){
02962         try{ 
02963                 // initialize everything
02964                 int i, j, rowNum, objNum;
02965                 if( m_mapExpressionTreesMod.size() > 0){
02966                         m_vdYval = this->forwardAD(0, m_vdX);   
02967                 }
02968                 // now get all function and constraint values using forward result
02969                 for(rowNum = 0; rowNum < m_iConstraintNumber; rowNum++){
02970                         m_mdConstraintFunctionValues[ rowNum] = 0.0;
02971                         if( m_mapExpressionTreesMod.find( rowNum) != m_mapExpressionTreesMod.end() ){
02972                                 m_mdConstraintFunctionValues[ rowNum] = m_vdYval[  m_mapCppADFunRangeIndex[ rowNum]];
02973                         }
02974                         // now the linear part
02975                         // be careful, loop over only the constant terms in sparseJacMatrix
02976                         i = m_sparseJacMatrix->starts[ rowNum];
02977                         j = m_sparseJacMatrix->starts[ rowNum + 1 ];
02978                         while ( (i - m_sparseJacMatrix->starts[ rowNum])  < m_sparseJacMatrix->conVals[ rowNum] ){
02979                                 m_mdConstraintFunctionValues[ rowNum] += m_sparseJacMatrix->values[ i]*x[ m_sparseJacMatrix->indexes[ i] ];
02980                                 i++;
02981                         }       
02982                         // add in the constraint function constant
02983                         m_mdConstraintFunctionValues[ rowNum] += m_mdConstraintConstants[ rowNum ];
02984                         #ifdef DEBUG
02985                         std::cout << "Constraint " <<  rowNum << " function value =  " << m_mdConstraintFunctionValues[ rowNum ] << std::endl;
02986                         #endif
02987                 }
02988                 // now get the objective function values from the forward result
02989                 for(objNum = 0; objNum < m_iObjectiveNumber; objNum++){
02990                         m_mdObjectiveFunctionValues[ objNum] = 0.0;
02991                         if( m_mapExpressionTreesMod.find( -objNum -1) != m_mapExpressionTreesMod.end() ){
02992                                 m_mdObjectiveFunctionValues[ objNum] = m_vdYval[ objNum];
02993                         }
02994                         for(i = 0; i < m_iVariableNumber; i++){
02995                                 m_mdObjectiveFunctionValues[ objNum] += m_mmdDenseObjectiveCoefficients[ objNum][i]*x[ i];
02996                         }
02997                         #ifdef DEBUG
02998                         std::cout << "Objective " << objNum << " function value =  " << m_mdObjectiveFunctionValues[ objNum] << std::endl;
02999                         #endif
03000                 }
03001         return true;
03002         }//end try
03003         catch(const ErrorClass& eclass){
03004                 throw ErrorClass( eclass.errormsg);
03005         }  
03006 }//end getZeroOrderResults
03007 
03008 
03009 
03010 bool OSInstance::getFirstOrderResults(double *x, double *objLambda, double *conMultipliers){
03011         try{
03012                 // initialize everything
03013                 unsigned int i, j;
03014                 int rowNum,  jacIndex;
03015                 unsigned int jstart, jend;
03016                 int idx;
03017                 OSExpressionTree *expTree = NULL;
03018                 int domainIdx = 0;      
03019                 std::map<int, OSExpressionTree*>::iterator posMapExpTree;
03020                 std::map<int, int>::iterator posVarIdx;
03021                         
03027                 if(m_iNumberOfNonlinearVariables >= m_mapExpressionTreesMod.size() ){
03028                         // calculate the gradient by doing a reverse sweep over each row
03029                         // loop over the constraints that have a nonlinear term and get their gradients
03030                         for(posMapExpTree = m_mapExpressionTreesMod.begin(); posMapExpTree != m_mapExpressionTreesMod.end(); ++posMapExpTree){
03031                                 idx = posMapExpTree->first;
03032                                 // we are considering only constraints, not objective function
03033                                 if(idx >= 0){
03034                                         m_vdRangeUnitVec[ domainIdx] = 1.;
03035                                         m_mapExpressionTreesMod[ idx]->getVariableIndiciesMap(); 
03036                                         m_vdYjacval = this->reverseAD(1, m_vdRangeUnitVec);
03037                                         // check size
03038                                         jstart = m_miJacStart[ idx] + m_miJacNumConTerms[ idx];
03039                                         jend = m_miJacStart[ idx + 1 ];
03040                                         if( (*m_mapExpressionTreesMod[ idx]->mapVarIdx).size() != (jend - jstart)) throw 
03041                                         ErrorClass("number of partials not consistent");
03042                                         j = 0;
03043                                         jacIndex = 0;
03044                                         for(posVarIdx = m_mapAllNonlinearVariablesIndex.begin(); posVarIdx 
03045                                                 != m_mapAllNonlinearVariablesIndex.end(); ++posVarIdx){
03046                                                 //std::cout << "Constraint Function Jacobian Values" << "For Constraint  " << idx  << std::endl;
03047                                                 //std::cout << "Jac Val for index " << posVarIdx->first  << " = " << m_vdYjacval[ jacIndex] << std::endl;
03048                                                 //if(m_miJacIndex[ jstart] != posVarIdx->first) throw ErrorClass("error calculating Jacobian matrix");
03049                                                 // we are working with variable posVarIdx->first in the original variable space
03050                                                 // we need to see which variable this is in the individual constraint map
03051                                                 if( (*m_mapExpressionTreesMod[ idx]->mapVarIdx).find( posVarIdx->first) != (*m_mapExpressionTreesMod[ idx]->mapVarIdx).end()){
03052                                                         m_mdJacValue[ jstart] = m_vdYjacval[ jacIndex];
03053                                                         jstart++;
03054                                                         j++;
03055                                                 }
03056                                                 jacIndex++;
03057                                         }
03058                                         
03059                                         m_vdRangeUnitVec[ domainIdx] = 0.;
03060                                         domainIdx++;
03061                                 }
03062                                 else{    // we have an objective function
03063                                         m_vdRangeUnitVec[ domainIdx] = 1.;
03064                                         m_vdYjacval = this->reverseAD(1, m_vdRangeUnitVec);
03065                                         for(i = 0; i < m_iNumberOfNonlinearVariables; i++){
03066                                                 //kipp fix when more than one obj
03067                                                         m_mmdObjGradient[  (abs( idx) - 1)][ m_miNonLinearVarsReverseMap[ i]] = m_vdYjacval[ i] + 
03068                                                                 m_mmdDenseObjectiveCoefficients[  (abs( idx) - 1)][ m_miNonLinearVarsReverseMap[ i]];
03069                                         }                                                                       
03070                                         m_vdRangeUnitVec[ domainIdx] = 0.;
03071                                         domainIdx++;
03072                                 }
03073                         }
03074                 }
03075                 else{  
03076                         // calculate the gradients using a forward sweep over all the variables.                
03077                         for(i = 0; i < m_iNumberOfNonlinearVariables; i++){
03078                                 m_vdDomainUnitVec[i] = 1.;     
03079                                 rowNum = 0;
03080                                 if( m_mapExpressionTreesMod.size() > 0){          
03081                                         m_vdYjacval = this->forwardAD(1, m_vdDomainUnitVec);
03082                                 } 
03083                                 // fill in Jacobian here, we have column i 
03084                                 // start Jacobian calculation
03085                                 for(posMapExpTree = m_mapExpressionTreesMod.begin(); posMapExpTree != m_mapExpressionTreesMod.end(); ++posMapExpTree){
03086                                         idx = posMapExpTree->first;
03087                                         // we are considering only constraints, not objective function
03088                                         if(idx >= 0){
03089                                                 //figure out original variable this corresponds to
03090                                                 //then use (*m_mapExpressionTreesMod[ idx]->mapVarIdx) to figure out which variable it is within row idx
03091                                                 //m_mapAllNonlinearVariablesIndex
03092                                                 //std::cout << "This is the following variable in the expression tree  " <<  (*m_mapExpressionTreesMod[ idx]->mapVarIdx)[ m_miNonLinearVarsReverseMap[ i]]<< std::endl;                         
03093                                                 expTree = m_mapExpressionTreesMod[ idx];                
03094                                                 if( (*expTree->mapVarIdx).find( m_miNonLinearVarsReverseMap[ i]) != (*expTree->mapVarIdx).end()  ){             
03095                                                         jacIndex = (*m_mapExpressionTreesMod[ idx]->mapVarIdx)[ m_miNonLinearVarsReverseMap[ i]];
03096                                                         jstart = m_miJacStart[ idx] + m_miJacNumConTerms[ idx];
03097                                                         // kipp change 1 to number of objective functions
03098                                                         m_mdJacValue[ jstart + jacIndex] = m_vdYjacval[m_iObjectiveNumber + rowNum];
03099                                                 }
03100                                                 rowNum++;
03101                                         }//end Jacobian calculation
03102                                         else{
03103                                                 // see if we have the objective function of interest
03104                                                 // kipp -- fix if more than one obj
03105                                                         m_mmdObjGradient[  (abs( idx) - 1)][ m_miNonLinearVarsReverseMap[ i]] = m_vdYjacval[ (abs( idx) - 1)] + 
03106                                                         m_mmdDenseObjectiveCoefficients[  (abs( idx) - 1)][ m_miNonLinearVarsReverseMap[ i]];                                   
03107                                 }//end Obj gradient calculation 
03108                         }                       
03109                         //
03110                         m_vdDomainUnitVec[i] = 0.;
03111                         }
03112                 }
03113                 #ifdef DEBUG
03114                 int k;
03115                 std::cout  << "JACOBIAN DATA " << std::endl;
03116                 for(idx = 0; idx < m_iConstraintNumber; idx++){
03117                         for(k = *(m_sparseJacMatrix->starts + idx); k < *(m_sparseJacMatrix->starts + idx + 1); k++){
03118                                 std::cout << "row idx = " << idx <<  "  col idx = "<< *(m_sparseJacMatrix->indexes + k)
03119                                 << " value = " << *(m_sparseJacMatrix->values + k) << std::endl;
03120                         }
03121                 }
03122                 std::cout  << "OBJECTIVE FUNCTION DATA " << std::endl;
03123                 for(i = 0; i < m_iObjectiveNumber; i++){
03124                         for(idx = 0; idx < m_iVariableNumber; idx++){
03125                                 std::cout << "var idx = " << idx <<  "  value = "<< m_mmdObjGradient[ i][idx] << std::endl;
03126                         }
03127                 }
03128                 #endif
03129                 return true;
03130         }//end try
03131         catch(const ErrorClass& eclass){
03132                 throw ErrorClass( eclass.errormsg);
03133         } 
03134 }// end getFirstOrderResults
03135                         
03136 
03137 bool OSInstance::getSecondOrderResults(double *x, double *objLambda, double *conMultipliers){
03138         try{
03139                 // initialize everything
03140                 unsigned int i, j;
03141                 int rowNum,  jacIndex;
03142                 int jstart,  idx;
03143                 OSExpressionTree *expTree = NULL;
03144                 int hessValuesIdx = 0;  
03145                 std::map<int, OSExpressionTree*>::iterator posMapExpTree;
03146                 std::map<int, int>::iterator posVarIndexMap;
03147                 if( conMultipliers == NULL) throw ErrorClass("cannot have a null vector of lagrange multipliers when calculating Hessian of Lagrangian");
03148                 if( m_vdLambda.size() > 0) m_vdLambda.clear();
03149                 for(posMapExpTree = m_mapExpressionTreesMod.begin(); posMapExpTree != m_mapExpressionTreesMod.end(); ++posMapExpTree){  
03150                         if( posMapExpTree->first >= 0){
03151                                 m_vdLambda.push_back( conMultipliers[ posMapExpTree->first]);
03152                         }
03153                         else{
03154                                 // kipp correct when there is more than one obj
03155                                 m_vdLambda.push_back( objLambda[ abs(posMapExpTree->first) - 1] );
03156                         }
03157                 }
03158                 for(i = 0; i < m_iNumberOfNonlinearVariables; i++){
03159                         m_vdDomainUnitVec[i] = 1.;     
03160                         rowNum = 0;
03161                         if( m_mapExpressionTreesMod.size() > 0){          
03162                                 m_vdYjacval = this->forwardAD(1, m_vdDomainUnitVec);
03163                         } 
03164                         // fill in Jacobian here, we have column i 
03165                         // start Jacobian calculation
03166                         for(posMapExpTree = m_mapExpressionTreesMod.begin(); posMapExpTree != m_mapExpressionTreesMod.end(); ++posMapExpTree){
03167                                 idx = posMapExpTree->first;
03168                                 // we are considering only constraints, not objective function
03169                                 if(idx >= 0){
03170                                         //figure out original variable this corresponds to
03171                                         //then use (*m_mapExpressionTreesMod[ idx]->mapVarIdx) to figure out which variable it is within row idx
03172                                         //m_mapAllNonlinearVariablesIndex
03173                                         //std::cout << "This is the following variable in the expression tree  " <<  (*m_mapExpressionTreesMod[ idx]->mapVarIdx)[ m_miNonLinearVarsReverseMap[ i]]<< std::endl;                         
03174                                         expTree = m_mapExpressionTreesMod[ idx];                
03175                                         if( (*expTree->mapVarIdx).find( m_miNonLinearVarsReverseMap[ i]) != (*expTree->mapVarIdx).end()  ){             
03176                                                 jacIndex = (*m_mapExpressionTreesMod[ idx]->mapVarIdx)[ m_miNonLinearVarsReverseMap[ i]];
03177                                                 jstart = m_miJacStart[ idx] + m_miJacNumConTerms[ idx];
03178                                                 m_mdJacValue[ jstart + jacIndex] = m_vdYjacval[m_iObjectiveNumber + rowNum];
03179                                         }
03180                                         rowNum++;
03181                                 }//end Jacobian calculation
03182                                 else{
03183                                         // see if we have the objective function of interest
03184                                         //kipp fix if more than one obj
03185                                                 m_mmdObjGradient[  (abs( idx) - 1)][ m_miNonLinearVarsReverseMap[ i]] = m_vdYjacval[ (abs( idx) - 1)] + 
03186                                                 m_mmdDenseObjectiveCoefficients[  (abs( idx) - 1)][ m_miNonLinearVarsReverseMap[ i]];                                   
03187                         }//end Obj gradient calculation 
03188                 }                       
03189                 // now calculate the Hessian
03190                 if( m_mapExpressionTreesMod.size() > 0){   
03191                         m_vdw = reverseAD(2, m_vdLambda);   // derivtative of partial
03192                 }
03193                 for(j = i; j < m_iNumberOfNonlinearVariables; j++){
03194                         if( m_vbLagHessNonz[i*m_iNumberOfNonlinearVariables + j] == true){
03195                                 m_LagrangianSparseHessian->hessValues[ hessValuesIdx] =  m_vdw[  j*2 + 1];
03196                                 #ifdef DEBUG
03197                                 std::cout << "reverse 2 " << m_LagrangianSparseHessian->hessValues[ hessValuesIdx] << std::endl;
03198                                 #endif
03199                                 hessValuesIdx++;
03200                         }
03201                 }
03202                 //
03203                 //
03204                 m_vdDomainUnitVec[i] = 0.;
03205         }
03206 
03207         #ifdef DEBUG
03208         int k;
03209         std::cout  << "JACOBIAN DATA " << std::endl;
03210         for(idx = 0; idx < m_iConstraintNumber; idx++){
03211                 for(k = *(m_sparseJacMatrix->starts + idx); k < *(m_sparseJacMatrix->starts + idx + 1); k++){
03212                         std::cout << "row idx = " << idx <<  "  col idx = "<< *(m_sparseJacMatrix->indexes + k)
03213                         << " value = " << *(m_sparseJacMatrix->values + k) << std::endl;
03214                 }
03215         }
03216         #endif
03217         return true;
03218         }//end try
03219         catch(const ErrorClass& eclass){
03220                 throw ErrorClass( eclass.errormsg);
03221         } 
03222 }// end getSecondOrderResults
03223 
03224 bool OSInstance::initForAlgDiff(){
03225         if( m_binitForAlgDiff == true ) return true;
03226         initializeNonLinearStructures( );
03227         initObjGradients();
03228         getAllNonlinearVariablesIndexMap( );
03229         if(m_bSparseJacobianCalculated  == false) getJacobianSparsityPattern();
03230         //see if we need to retape 
03231         //loop over expression tree and see if one requires it
03232         std::map<int, OSExpressionTree*>::iterator posMapExpTree;
03233         for(posMapExpTree = m_mapExpressionTreesMod.begin(); posMapExpTree != m_mapExpressionTreesMod.end(); ++posMapExpTree){
03234                 if(posMapExpTree->second->bCppADMustReTape == true) m_bCppADMustReTape = true;
03235         }                               
03236 
03237         #ifdef DEBUG
03238         std::cout << "RETAPE ==  " << m_bCppADMustReTape << std::endl;
03239         #endif
03240         unsigned int i;
03241         for(i = 0; i < m_iNumberOfNonlinearVariables; i++){
03242                 m_vdDomainUnitVec.push_back( 0.0 );
03243         }
03244         for(i = 0; i < m_mapExpressionTreesMod.size(); i++){
03245                 m_vdRangeUnitVec.push_back( 0.0 );
03246         }
03247         m_binitForAlgDiff = true;
03248         return true;
03249 }//end initForAlgDiff
03250 
03251 bool OSInstance::initObjGradients(){
03252         int i, j;
03253         int m, n;
03254         m = getObjectiveNumber();
03255         n = getVariableNumber();
03256         getDenseObjectiveCoefficients();
03257         m_mmdObjGradient = new double*[m];
03258         for(i = 0; i < m; i++){
03259                 m_mmdObjGradient[i] = new double[n];
03260                 for(j = 0; j < n; j++){
03261                         m_mmdObjGradient[i][j] =  m_mmdDenseObjectiveCoefficients[ i][j];
03262                         #ifdef DEBUG
03263                         std::cout << "m_mmdObjGradient[i][j] = " << m_mmdObjGradient[i][j]  << std::endl;
03264                         #endif
03265                 }
03266         }
03267         return true;
03268 }//end initObjGradients

Generated on Sat Mar 29 22:38:01 2008 by  doxygen 1.5.3