returnValue ExportIRK4StageSimplifiedNewton::setTransformations( const DMatrix& _transf1, const DMatrix& _transf2 ) {
transf1 = _transf1;
transf2 = _transf2;
if( _transf1.getNumRows() != 4 || _transf1.getNumCols() != 4 ) return ACADOERROR( RET_INVALID_ARGUMENTS );
if( _transf2.getNumRows() != 4 || _transf2.getNumCols() != 4 ) return ACADOERROR( RET_INVALID_ARGUMENTS );
if( _transf1.isZero() || _transf2.isZero() ) return ACADOERROR( RET_INVALID_ARGUMENTS );
return SUCCESSFUL_RETURN;
}
VariablesGrid::VariablesGrid( const DMatrix& arg,
VariableType _type
) : MatrixVariablesGrid( arg.getNumCols()-1,1,arg.getNumRows(),_type )
{
uint i,j;
for( i=0; i<arg.getNumRows(); ++i )
{
setTime( i,arg( i,0 ) );
for( j=1; j<arg.getNumCols(); ++j )
operator()( i,j-1 ) = arg( i,j );
}
}
returnValue ModelData::setNARXmodel( const uint _delay, const DMatrix& _parms ) {
if( rhs_name.empty() && NX2 == 0 && NX3 == 0 ) {
NX2 = _parms.getNumRows();
delay = _delay;
uint numParms = 1;
uint n = delay*getNX();
if( delay >= 1 ) {
numParms = numParms + n;
}
for( uint i = 1; i < delay; i++ ) {
numParms = numParms + (n+1)*(uint)pow((double)n,(int)i)/2;
}
if( _parms.getNumCols() == numParms ) {
parms = _parms;
}
else {
return ACADOERROR( RET_UNABLE_TO_EXPORT_CODE );
}
export_rhs = BT_TRUE;
}
else {
return ACADOERROR( RET_INVALID_OPTION );
}
return SUCCESSFUL_RETURN;
}
returnValue ExportIRK4StageSimplifiedNewton::setEigenvalues( const DMatrix& _eig ) {
eig = _eig;
if( _eig.getNumRows() != 2 || _eig.getNumCols() != 2 ) return ACADOERROR( RET_INVALID_ARGUMENTS );
if( _eig.isZero() ) return ACADOERROR( RET_INVALID_ARGUMENTS );
// each row represents a complex conjugate pair of eigenvalues
return SUCCESSFUL_RETURN;
}
returnValue WeightGeneration::generateWeights( const int &n ,
DVector &weight ,
DMatrix &Weights ,
const DVector &weightsLB ,
const DVector &weightsUB ,
DVector &formers ,
const int &layer ,
int &lastOne ,
int ¤tOne,
double &step
) const{
int run1, run2;
if( n == layer ){
//printf("case1: n = %d layer = %d \n", n, layer );
double weight_test = 1.0;
for( run1 = 0; run1 < (int) weight.getDim()-1; run1++ )
weight_test -= weight(run1);
if( ( weight_test >= weightsLB(layer) - 100.0*EPS ) &&
( weight_test <= weightsUB(layer) + 100.0*EPS ) ){
if( currentOne >= 0 ){
DVector tmp( formers.getDim()+1 );
for( run1 = 0; run1 < (int) formers.getDim(); run1++ )
tmp(run1) = formers(run1);
tmp(formers.getDim()) = lastOne;
formers = tmp;
lastOne = currentOne;
currentOne = -1;
}
else{
DVector tmp( formers.getDim()+1 );
for( run1 = 0; run1 < (int) formers.getDim(); run1++ )
tmp(run1) = formers(run1);
tmp(formers.getDim()) = Weights.getNumCols();
formers = tmp;
}
weight(n) = weight_test;
DMatrix tmp( weight.getDim(), Weights.getNumCols()+1 );
for( run1 = 0; run1 < (int) Weights.getNumRows(); run1++ ){
for( run2 = 0; run2 < (int) Weights.getNumCols(); run2++ )
tmp(run1,run2) = Weights(run1,run2);
}
for( run2 = 0; run2 < (int) weight.getDim(); run2++ )
tmp(run2,Weights.getNumCols()) = weight(run2);
Weights = tmp;
}
}
else{
//printf("case2: n = %d layer = %d \n", n, layer );
double weight_test = weightsLB(n);
while( weight_test <= weightsUB(n) + 100.0*EPS ){
if( n == layer-2 ) currentOne = Weights.getNumCols()+1;
weight(n) = weight_test;
WeightGeneration child;
child.generateWeights( n+1, weight, Weights, weightsLB, weightsUB, formers, layer, lastOne, currentOne, step );
weight_test += (weightsUB(n)-weightsLB(n))*step;
}
}
return SUCCESSFUL_RETURN;
}
