DLLEXPORT void XGBoosterBoostOneIter( void *handle, void *dtrain,
float *grad, float *hess, size_t len, int bst_group ){
Booster *bst = static_cast<Booster*>(handle);
DMatrix *dtr = static_cast<DMatrix*>(dtrain);
bst->CheckInit(); dtr->CheckInit();
bst->BoostOneIter( *dtr, grad, hess, len, bst_group );
}
double SQPInternal::quad_form(const std::vector<double>& x, const DMatrix& A){
// Assert dimensions
casadi_assert(x.size()==A.size1() && x.size()==A.size2());
// Access the internal data of A
const std::vector<int> &A_rowind = A.rowind();
const std::vector<int> &A_col = A.col();
const std::vector<double> &A_data = A.data();
// Return value
double ret=0;
// Loop over the rows of A
for(int i=0; i<x.size(); ++i){
// Loop over the nonzeros of A
for(int el=A_rowind[i]; el<A_rowind[i+1]; ++el){
// Get column
int j = A_col[el];
// Add contribution
ret += x[i]*A_data[el]*x[j];
}
}
return ret;
}
DLLEXPORT void XGBoosterUpdateInteract( void *handle, void *dtrain, const char *action ){
Booster *bst = static_cast<Booster*>(handle);
DMatrix *dtr = static_cast<DMatrix*>(dtrain);
bst->CheckInit(); dtr->CheckInit();
std::string act( action );
bst->UpdateInteract( act, *dtr );
}
//--------------------------------------------------------------------------
double Performance(const DVector &ytest_truth, const DMatrix &Ytest_predict,
int NumClasses) {
long n = ytest_truth.GetN();
if (n != Ytest_predict.GetM() || Ytest_predict.GetN() != NumClasses ) {
printf("Performance. Error: Size mismatch. Return NAN");
return NAN;
}
if (NumClasses <= 2) {
printf("Performance. Error: Not multiclass classification. Return NAN");
return NAN;
}
double perf = 0.0;
// Compute ytest_predict
DVector ytest_predict(n);
double *y = ytest_predict.GetPointer();
for (long i = 0; i < n; i++) {
DVector row(NumClasses);
Ytest_predict.GetRow(i, row);
long idx = -1;
row.Max(idx);
y[i] = (double)idx;
}
// Accuracy
double *y1 = ytest_truth.GetPointer();
double *y2 = ytest_predict.GetPointer();
for (long i = 0; i < n; i++) {
perf += ((int)y1[i])==((int)y2[i]) ? 1.0 : 0.0;
}
perf = perf/n * 100.0;
return perf;
}
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 ConstraintElement::get(Function& function_, DMatrix& lb_, DMatrix& ub_)
{
if ( fcn == NULL )
return RET_INITIALIZE_FIRST;
// This is not exactly bullet proof, but for now will serve the purpose
function_ = fcn[ 0 ];
int dimFcn = fcn[ 0 ].getDim();
if ( dimFcn == 0 )
{
lb_.init(0, 0);
ub_.init(0, 0);
return SUCCESSFUL_RETURN;
}
lb_.init(nB, dimFcn);
ub_.init(nB, dimFcn);
int i, j;
for (i = 0; i < nB; ++i)
for (j = 0; j < dimFcn; ++j)
{
lb_(i, j) = lb[ i ][ j ];
ub_(i, j) = ub[ i ][ j ];
}
return SUCCESSFUL_RETURN;
}
void GLEABC::set_BBT(const DMatrix& nBBT)
{
if (n==0) set_size(nBBT.rows());
if (A.rows()>0) fr_init=true;
if (n!=nBBT.rows()) ERROR("Use set_size if you want to change matrix size.");
if (n!=nBBT.cols()) ERROR("Square matrix expected.");
BBT=nBBT; fr_c=false;
}
int EigenVectors::TrimSystem(DMatrix &dmat, DArray &darr) const {
dmat.pop_back();
darr.pop_back();
int i, size = dmat.size();
for(i = 0; i < size; ++i) {
dmat[i].pop_back();
}
return size;
}
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;
}
void EigenVectors::InitMatrix(DMatrix &dmat, int size) const {
if(!dmat.empty()) {
dmat.clear();
}
DArray darr(size, 0.);
dmat.reserve(size);
for(int i = 0; i < size; ++i) {
dmat.push_back(darr);
}
}
//--------------------------------------------------------------------------
void ConvertYtrain(const DVector &ytrain, DMatrix &Ytrain, int NumClasses) {
Ytrain.Init(ytrain.GetN(), NumClasses);
// Lingfei: the y label is supposed to start from 0.
for (int i = 0; i < NumClasses; i++) {
DVector y = ytrain;
double *my = y.GetPointer();
for (long j = 0; j < y.GetN(); j++) {
my[j] = ((int)my[j])==i ? 1.0 : -1.0;
}
Ytrain.SetColumn(i, y);
}
}
int main(int argc, char *argv[])
{
/*
DMatrix W { 5, 5,
{ 0, 4, inf, 2, inf,
inf, 0, 3, inf, 1,
5, inf, 0, 1, inf,
7, inf, inf, 0, 5,
3, inf, inf, inf, 0 } };
*/
// Checking that the value of n is given
if (argc != 2) {
cerr << "floyd-sequential n" << endl;
exit(1);
}
int n = atoi(argv[1]);
if (n == 0) {
cerr << argv[1] << " is a wrong parameter" << endl;
exit(1);
}
srand(10);
DMatrix W = generate_matrix(n);
DMatrix dist(W.rows(),W.cols(),inf);
IMatrix next(W.rows(),W.cols(),-1);
double start = omp_get_wtime();
floyd_warshall(W, dist, next);
path_t p = recover_path(dist,next,0,(rand()%(n-1))+1);
double end = omp_get_wtime();
if (p.size() == 0) {
cout << "No path" << endl;
} else {
copy(p.begin(), p.end(), ostream_iterator<int>(cout, " "));
cout << "\n";
double cost = 0;
#pragma omp parallel for reduction (+:cost)
for (uint32_t i = 0; i < p.size()-1; ++i) {
cost += dist(p[i],p[i+1]);
}
cout << "Cost: " << cost << "\n";
}
cout << "Elapsed time: " << (end-start) << endl;
}
void GLEABC::set_C(const DMatrix& nC)
{
if (n==0) set_size(nC.rows());
if (A.rows()>0) fr_init=true;
if (n!=nC.rows()) ERROR("Use set_size if you want to change matrix size.");
if (n!=nC.cols()) ERROR("Square matrix expected.");
C=nC; fr_c=true;
if (fr_init && fr_c)
{
//update BBT, it's so cheap we can do it routinely;
DMatrix T; mult(A, C, BBT); transpose(BBT, T); incr(BBT,T);
}
}
returnValue RungeKuttaExport::initializeButcherTableau( const DMatrix& _AA, const DVector& _bb, const DVector& _cc ) {
if( _cc.isEmpty() || !_AA.isSquare() || _AA.getNumRows() != _bb.getDim() || _bb.getDim() != _cc.getDim() ) return RET_INVALID_OPTION;
numStages = _cc.getDim();
is_symmetric = checkSymmetry( _cc );
// std::cout << "Symmetry of the chosen method: " << is_symmetric << "\n";
AA = _AA;
bb = _bb;
cc = _cc;
return SUCCESSFUL_RETURN;
}
void CSparseCholeskyInternal::prepare() {
prepared_ = false;
// Get a reference to the nonzeros of the linear system
const vector<double>& linsys_nz = input().data();
// Make sure that all entries of the linear system are valid
for (int k=0; k<linsys_nz.size(); ++k) {
casadi_assert_message(!isnan(linsys_nz[k]), "Nonzero " << k << " is not-a-number");
casadi_assert_message(!isinf(linsys_nz[k]), "Nonzero " << k << " is infinite");
}
if (verbose()) {
userOut() << "CSparseCholeskyInternal::prepare: numeric factorization" << endl;
userOut() << "linear system to be factorized = " << endl;
input(0).printSparse();
}
if (L_) cs_nfree(L_);
L_ = cs_chol(&AT_, S_) ; // numeric Cholesky factorization
if (L_==0) {
DMatrix temp = input();
temp.makeSparse();
if (temp.sparsity().issingular()) {
stringstream ss;
ss << "CSparseCholeskyInternal::prepare: factorization failed due "
"to matrix being singular. Matrix contains numerical zeros which are"
" structurally non-zero. Promoting these zeros to be structural "
"zeros, the matrix was found to be structurally rank deficient. "
"sprank: " << sprank(temp.sparsity()) << " <-> " << temp.size2() << endl;
if (verbose()) {
ss << "Sparsity of the linear system: " << endl;
input(LINSOL_A).sparsity().print(ss); // print detailed
}
throw CasadiException(ss.str());
} else {
stringstream ss;
ss << "CSparseCholeskyInternal::prepare: factorization failed, "
"check if Jacobian is singular" << endl;
if (verbose()) {
ss << "Sparsity of the linear system: " << endl;
input(LINSOL_A).sparsity().print(ss); // print detailed
}
throw CasadiException(ss.str());
}
}
casadi_assert(L_!=0);
prepared_ = true;
}
void Solver::applyIntegrationOnRigidBodies(DMatrix& s_next, DMatrix& u_next)
{
for (int i = 0; i< m_rigidBodies.size(); i++)
{
RigidBody_c* rb = m_rigidBodies[i];
rb->m_position.x = s_next.Get(i * 7 + 0);
rb->m_position.y = s_next.Get(i * 7 + 1);
rb->m_position.z = s_next.Get(i * 7 + 2);
rb->m_orientation.w = s_next.Get(i * 7 + 3);
rb->m_orientation.x = s_next.Get(i * 7 + 4);
rb->m_orientation.y = s_next.Get(i * 7 + 5);
rb->m_orientation.z = s_next.Get(i * 7 + 6);
rb->m_linearVelocity.x = u_next.Get(i * 6 + 0);
rb->m_linearVelocity.y = u_next.Get(i * 6 + 1);
rb->m_linearVelocity.z = u_next.Get(i * 6 + 2);
rb->m_angularVelocity.x = u_next.Get(i * 6 + 3);
rb->m_angularVelocity.y = u_next.Get(i * 6 + 4);
rb->m_angularVelocity.z = u_next.Get(i * 6 + 5);
rb->m_force = XMFLOAT3(0, 0, 0);
rb->m_torque = XMFLOAT3(0, 0, 0);
XMVECTOR normalQuaternion = XMQuaternionNormalize(XMLoadFloat4(&rb->m_orientation));
XMStoreFloat4(&rb->m_orientation, normalQuaternion);
}
}
DMatrix readVariable(const std::string &folder, const std::string filename)
{
std::ifstream file;
DMatrix variable;
file.open(folder + filename);
if(!file)
throw std::runtime_error("Couldn't open the file " + folder + filename + ".");
variable.read(file);
file.close();
return variable;
}
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 );
}
}
/**
* save a real matrix into a text file
*
* row_index col_index matrix_element
*/
void saveMatrixText(std::string filename, DMatrix& m, bool exactValue) {
std::ofstream out(filename.c_str());
// print to full precision
if (exactValue) {
out.precision(dbl::digits10 + 2);
}
// use default precision to save space
for (int row=0; row< m.rows(); ++row) {
for (int col=0; col<m.cols(); ++col) {
double v = m(row, col);
out << row <<" " << col << " "
<< v << std::endl;
}
}
out.close();
}
void floyd_warshall(const DMatrix& W,
DMatrix& dist,
IMatrix& next)
{
uint32_t nvertices = W.rows();
//#pragma omp parallel for
for (uint32_t i = 0; i < nvertices; ++i) {
for (uint32_t j = 0; j < nvertices; ++j) {
dist(i,j) = W(i,j);
}
}
//#pragma omp parallel for
for (uint32_t k=0; k < nvertices; ++k) {
for (uint32_t i=0; i < nvertices; ++i) {
for (uint32_t j=0; j < nvertices; ++j) {
double tmp = dist(i,k) + dist(k,j);
if (tmp < dist(i,j)) {
dist(i,j) = tmp;
next(i,j) = k;
}
}
}
}
}
//integrates the peak of the velocity-velocity correlation function from w*(1-d) to w*(1+d)
void harm_peak(const DMatrix& A, const DMatrix& BBT, double w, double d, double &pi)
{
unsigned long n=A.rows();
//prepares extended matrices
toolbox::FMatrix<double> xA(n+1,n+1), xBBT(n+1,n+1), xC;
xA*=0.; xBBT=xA;
for (int i=0; i<n;++i)for (int j=0; j<n;++j)
{ xA(i+1,j+1)=A(i,j); xBBT(i+1,j+1)=BBT(i,j); }
xA(0,1)=-1; xA(1,0)=w*w; //sets the harmonic hamiltonian part
GLEABC abc; abc.set_A(xA); abc.set_BBT(xBBT);
//std::cerr<<" --- C in tauw "<<w<<" ----\n"<<xC<<" ---------- \n";
abc.get_C(xC);
//get power spectrum peak intensity (should program a better way to integrate....)
//the total integral under the peak is Pi/2
FMatrix<double> AW1(xA), AW2(xA), atAW1, atAW2;
AW1*=(1./(w*(1.-d/2.)));
AW2*=(1./(w*(1.+d/2.)));
MatrixFunction(AW1,&atan,atAW1);
MatrixFunction(AW2,&atan,atAW2);
mult(atAW1,xC,AW1); mult(atAW2,xC,AW2);
pi=2./toolbox::constant::pi*(AW1(1,1)-AW2(1,1))/xC(1,1);
}
// xgboost implementation
DLLEXPORT void *XGBoosterCreate( void *dmats[], size_t len ){
std::vector<xgboost::regrank::DMatrix*> mats;
for( size_t i = 0; i < len; ++i ){
DMatrix *dtr = static_cast<DMatrix*>(dmats[i]);
dtr->CheckInit();
log_debug("XGBoosterCreate: * i=%u", (unsigned int)i);
mats.push_back( dtr );
log_debug("XGBoosterCreate: ** i=%u", (unsigned int)i);
}
log_debug("XGBoosterCreate: xxxx");
log_debug("XGBoosterCreate: mats=%u", (unsigned int)mats.size());
void *booster = new Booster( mats );
log_debug("XGBoosterCreate: booster=%p", booster);
return booster;
}
returnValue OCP::minimizeLSQEndTerm( const DMatrix & S,
const Function & m,
const DVector & r ){
if ( S.isPositiveSemiDefinite() == BT_FALSE )
return ACADOERROR( RET_NONPOSITIVE_WEIGHT );
return objective->addLSQEndTerm( S, m, r );
}
returnValue OCP::minimizeLSQ( const DMatrix &S,
const Function &h,
const VariablesGrid &r ){
if ( S.isPositiveSemiDefinite() == BT_FALSE )
return ACADOERROR( RET_NONPOSITIVE_WEIGHT );
MatrixVariablesGrid tmpS(S);
return objective->addLSQ( &tmpS, h, &r );
}
void Legendre::FindPolynomial(DArray &dArray) {
DMatrix dMatrix;
dMatrix.reserve(nn);
dArray.clear();
dArray.reserve(nn);
//double dbg[6] = { -.7746, .5556, 0., .8889, .7746, .5556 };
for(int i = 0; i < nn; ++i) {
double rat = static_cast<double>(i + 1) / static_cast<double>(nn);
dArray.push_back((i % 2) ? 1. - fabs(dArray[i - 1]) / (ul - ll) * 2. : (ll + (ul - ll) * rat));
//dArray.push_back(rat);
//dArray.push_back(dbg[i]);
cerr << "Initially " << i << ": " << dArray[i] << endl;
dMatrix.push_back(dIntegrals);
}
do {
Increment(dMatrix, dArray);
SolveSytem(dMatrix, dArray);
} while(ComputeError(dArray) > precision);
ExtractPolynomial(dArray);
}
EigenValues::EigenValues(const DMatrix& input) {
int size = input.size();
input_.reserve(size);
CDArray cdarr(size);
for(int i = 0; i < size; ++i) {
for(int j = 0; j < size; ++j) {
cdarr[j] = cd(input[i][j], 0.);
}
input_.push_back(cdarr);
}
}
void harm_check(const DMatrix& A, const DMatrix& BBT, double w, double &tq2, double &tp2, double& th, double& q2, double& p2, double& pq, double& dwq, double& dwp, double& lambdafp)
{
unsigned long n=A.rows();
double w2=w*w, w4=w2*w2;
//prepares extended matrices
toolbox::FMatrix<double> xA(n+1,n+1), xBBT(n+1,n+1), xC;
xA*=0.; xBBT=xA;
for (int i=0; i<n;++i)for (int j=0; j<n;++j)
{ xA(i+1,j+1)=A(i,j); xBBT(i+1,j+1)=BBT(i,j); }
xA(0,1)=-1; xA(1,0)=w2; //sets the harmonic hamiltonian part
GLEABC abc; abc.set_A(xA); abc.set_BBT(xBBT);
std::valarray<std::complex<double> >eva; abc.get_evA(eva);
lambdafp=eva[0].real();
for (int i=0; i<n+1; i++) lambdafp=(eva[i].real()<lambdafp?eva[i].real():lambdafp);
// std::cerr<<" --- C in tauw "<<w<<" ----\n"<<xC<<" ---------- \n";
abc.get_C(xC);
double c00=xC(0,0), c01=xC(0,1), c11=xC(1,1);
q2=c00*w2;
p2=c11;
pq=c01*w;
double t00, t01, t11;
abc.get_tau2(0,0,0,0,t00); abc.get_tau2(0,0,1,1,t01); abc.get_tau2(1,1,1,1,t11);
double qqqq=t00*w4;
double qqpp=t01*w2;
double ppqq=qqpp; //come on, the simmetry is evident. let's save time!
double pppp=t11;
// std::cerr<<"NEW: "<<qqqq<<","<<qqpp<<","<<ppqq<<","<<pppp<<"\n";
double qqqq0=q2*q2;
double qqpp0=pq*pq;
double ppqq0=qqpp0;
double pppp0=p2*p2;
th=(pppp+qqqq+ppqq+qqpp)/(pppp0+qqqq0+ppqq0+qqpp0);
tp2=pppp/pppp0;
tq2=qqqq/qqqq0;
//get power spectrum peak widths
toolbox::FMatrix<double> t1, t2, xDELTA;
mult(xA,xA,t1); //A^2
for (int i=0; i<n+1;++i) t1(i,i)+=w2; //A^2+w^2
MatrixInverse(t1,t2); //1/(A^2+w^2)
mult(xA,t2,t1); //A/(A^2+w^2)
mult(t1,xC,xDELTA); //A/(A^2+w^2)C
dwq=xC(0,0)/xDELTA(0,0);
dwp=xC(1,1)/xDELTA(1,1);
}
inline void BoostOneIter( const DMatrix &train,
float *grad, float *hess, size_t len, int bst_group ){
this->grad_.resize( len ); this->hess_.resize( len );
memcpy( &this->grad_[0], grad, sizeof(float)*len );
memcpy( &this->hess_[0], hess, sizeof(float)*len );
if( grad_.size() == train.Size() ){
if( bst_group < 0 ) bst_group = 0;
base_gbm.DoBoost(grad_, hess_, train.data, train.info.root_index, bst_group);
}else{
utils::Assert( bst_group == -1, "must set bst_group to -1 to support all group boosting" );
int ngroup = base_gbm.NumBoosterGroup();
utils::Assert( grad_.size() == train.Size() * (size_t)ngroup, "BUG: UpdateOneIter: mclass" );
std::vector<float> tgrad( train.Size() ), thess( train.Size() );
for( int g = 0; g < ngroup; ++ g ){
memcpy( &tgrad[0], &grad_[g*tgrad.size()], sizeof(float)*tgrad.size() );
memcpy( &thess[0], &hess_[g*tgrad.size()], sizeof(float)*tgrad.size() );
base_gbm.DoBoost(tgrad, thess, train.data, train.info.root_index, g );
}
}
}
int main() {
DMatrix dmat;
#if 1
ifstream ifs("data.dat");
string line;
while(getline(ifs, line)) {
DArray darr;
double val;
stringstream iss(line);
while(iss >> val) {
darr.push_back(val);
}
dmat.push_back(darr);
}
#else
DArray darr;
darr.push_back(2.);
darr.push_back(7.);
darr.push_back(1.);
dmat.push_back(darr);
darr[0] = 3.;
darr[1] = 4.;
darr[2] = 8.;
dmat.push_back(darr);
darr[0] = 4.;
darr[1] = 3.;
darr[2] = 4.;
dmat.push_back(darr);
#endif
EigenValues egv(dmat);
egv.ComputeEigenValues();
egv.GetDeterminant();
Laplacian lpl(dmat);
cout << "Matrix determinant (original): " << lpl.GetDeterminant() << endl << endl;
return 0;
}
void EigenVectors::PopulateSystem(DMatrix &dmat, DArray &darr, const DMatrix &appld, int ind) const {
int size = appld.size();
for(int i1 = 0, i2 = 0; i1 < size; ++i1) {
if(i1 != ind) {
darr[i2] -= appld[i1][ind];
for(int j1 = 0, j2 = 0; j1 < size; ++j1) {
if(j1 != ind) {
dmat[i2][j2] = appld[i1][j1];
++j2;
}
}
++i2;
}
}
}
