EnumerationConstraint* ModelEnumerator::doInit(SharedContext& ctx, MinimizeConstraint* min, int numModels) {
delete queue_;
queue_ = 0;
initProjection(ctx);
uint32 st = strategy();
if (detectStrategy() || (ctx.concurrency() > 1 && !ModelEnumerator::supportsParallel())) {
st = 0;
}
bool optOne = minimizer() && minimizer()->mode() == MinimizeMode_t::optimize;
bool trivial = optOne || std::abs(numModels) == 1;
if (optOne && project_) {
const SharedMinimizeData* min = minimizer();
for (const WeightLiteral* it = min->lits; !isSentinel(it->first) && trivial; ++it) {
trivial = ctx.varInfo(it->first.var()).project();
}
if (!trivial) { ctx.report(warning(Event::subsystem_prepare, "Projection: Optimization may depend on enumeration order.")); }
}
if (st == strategy_auto) { st = trivial || (project_ && ctx.concurrency() > 1) ? strategy_record : strategy_backtrack; }
if (trivial) { st |= trivial_flag; }
if (ctx.concurrency() > 1 && !trivial && st != strategy_backtrack) {
queue_ = new SolutionQueue(ctx.concurrency());
queue_->reserve(ctx.concurrency() + 1);
}
options_ &= ~uint32(strategy_opts_mask);
options_ |= st;
Solver& s = *ctx.master();
EnumerationConstraint* c = st == strategy_backtrack
? static_cast<ConPtr>(new BacktrackFinder(s, min, project_, projectOpts()))
: static_cast<ConPtr>(new RecordFinder(s, min, project_, queue_));
if (projectionEnabled()) { setIgnoreSymmetric(true); }
return c;
}
EnumerationConstraint* ModelEnumerator::doInit(SharedContext& ctx, SharedMinimizeData* opt, int numModels) {
initProjection(ctx);
uint32 st = strategy();
if (detectStrategy() || (ctx.concurrency() > 1 && !ModelEnumerator::supportsParallel())) {
st = 0;
}
bool optOne = opt && opt->mode() == MinimizeMode_t::optimize;
bool trivial = optOne || std::abs(numModels) == 1;
if (optOne && projectionEnabled()) {
for (const WeightLiteral* it = minimizer()->lits; !isSentinel(it->first) && trivial; ++it) {
trivial = ctx.varInfo(it->first.var()).project();
}
if (!trivial) { ctx.report(warning(Event::subsystem_prepare, "Projection: Optimization may depend on enumeration order.")); }
}
if (st == strategy_auto) { st = trivial || (projectionEnabled() && ctx.concurrency() > 1) ? strategy_record : strategy_backtrack; }
if (trivial) { st |= trivial_flag; }
options_ &= ~uint32(strategy_opts_mask);
options_ |= st;
const LitVec* dom = (projectOpts() & project_dom_lits) != 0 ? (ctx.heuristic.domRec = &domRec_) : 0;
EnumerationConstraint* c = st == strategy_backtrack
? static_cast<ConPtr>(new BacktrackFinder(projectOpts()))
: static_cast<ConPtr>(new RecordFinder(dom));
if (projectionEnabled()) { setIgnoreSymmetric(true); }
return c;
}
long NormalModeRelax::run(const long numTimesteps) {
if( numTimesteps < 1 ) return 0;
//check valid eigenvectors
if(*Q == NULL)
report << error << "No Eigenvectors for NormalMode integrator."<<endr;
//
//do minimization with local forces, max loop 100, set subSpace minimization true
itrs = minimizer(minLim, 100, simpleMin, reDiag, true, &forceCalc, &lastLambda, &app->energies, &app->positions, app->topology);
Real minPotEnergy = app->energies.potentialEnergy(); //save potential energy before random
//constraints?
app->energies.clear();
//run brownian if any remaining modes
if(testRemainingModes()) myNextIntegrator->run(myCycleLength); //cyclelength
(app->energies)[ScalarStructure::OTHER] = minPotEnergy; //store minimum potential energy
if(*Q == NULL){ //rediagonalize?
if(myPreviousIntegrator == NULL)
report << error << "[NormalModeRelax::Run] Re-diagonalization forced with NormalModeRelax as outermost Integrator. Aborting."<<endr;
return numTimesteps;
}
//
postStepModify();
return numTimesteps;
}
gp_Circ LeastSquaresFitting::fitLeastSquaresCircleToPoints(const Handle_TColgp_HArray1OfPnt& points)
{
//Get initial guess for least squares fit
InitialGuessForLeastSquaresFitting initialGuess = findInitialGuess(points);
math_Vector startingPointForBFGS(1,3);
startingPointForBFGS(1) = initialGuess.myCenterPoint.X();
startingPointForBFGS(2) = initialGuess.myCenterPoint.Y();
startingPointForBFGS(3) = initialGuess.myRadius;
CircleFitCostFunction costFunction(points);
math_BFGS minimizer(costFunction,startingPointForBFGS);
math_Vector minimum(1,3);
if (minimizer.IsDone())
{
std::cout << "Minimizer is done, result is:" << std::endl;
minimizer.Location(minimum);
std::cout << minimum(1) << " " << minimum(2) << " " << minimum(3) << std::endl;
} else
{
std::cout << "Error: minimizer failed!" << std::endl;
}
gp_Ax2 axis = gp::XOY();
axis.SetLocation(gp_Pnt(minimum(1),minimum(2),0.0));
gp_Circ result(axis,minimum(3));
return result;
}
EllipseFit2<Real>::EllipseFit2 (int numPoints, const Vector2<Real>* points,
Vector2<Real>& center, Matrix2<Real>& rotate, Real diag[2],
Real& error)
:
mNumPoints(numPoints),
mPoints(points)
{
// Energy function is E : R^5 -> R where
// V = (V0, V1, V2, V3, V4)
// = (D[0], D[1], U.X(), U.Y(), atan2(R[1][0],R[1][1])).
mTemp = new1<Vector2<Real> >(numPoints);
MinimizeN<Real> minimizer(5, Energy, 8, 8, 32, this);
InitialGuess(numPoints, mPoints, center, rotate, diag);
Real angle = Math<Real>::ACos(rotate[0][0]);
Real e0 = diag[0]*Math<Real>::FAbs(rotate[0][0]) +
diag[1]*Math<Real>::FAbs(rotate[0][1]);
Real e1 = diag[0]*Math<Real>::FAbs(rotate[1][0]) +
diag[1]*Math<Real>::FAbs(rotate[1][1]);
Real v0[5] =
{
((Real)0.5)*diag[0],
((Real)0.5)*diag[1],
center.X() - e0,
center.Y() - e1,
(Real)0
};
Real v1[5] =
{
((Real)2)*diag[0],
((Real)2)*diag[1],
center.X() + e0,
center.Y() + e1,
Math<Real>::PI
};
Real vInitial[5] =
{
diag[0],
diag[1],
center.X(),
center.Y(),
angle
};
Real vMin[5];
minimizer.GetMinimum(v0, v1, vInitial, vMin, error);
diag[0] = vMin[0];
diag[1] = vMin[1];
center.X() = vMin[2];
center.Y() = vMin[3];
rotate.MakeRotation(vMin[4]);
delete1(mTemp);
}
int RigidRegistration::StartRegistration(vnl_vector<double>& params) {
vnl_lbfgsb minimizer(*func_);
SetRigidTransformBound(this->d_, &minimizer);
//double fxval;
func_->SetBase(this);
for (unsigned int k = 0; k < this->level_; ++k) {
func_->SetScale(this->v_scale_[k]);
SetParam(params);
minimizer.set_max_function_evals(this->v_func_evals_[k]);
// For more options, see
// http://public.kitware.com/vxl/doc/release/core/vnl/html/vnl__nonlinear__minimizer_8h-source.html
minimizer.minimize(params);
if (minimizer.get_failure_code() < 0) {
/*
fxval = func->f( params );
vcl_cout << "break Minimized to " << fxval << vcl_endl
<< "Iterations: " << minimizer.get_num_iterations() << "; "
<< "Evaluations: " << minimizer.get_num_evaluations() << vcl_endl;
vcl_cout << params << vcl_endl;
*/
return -1;
}
/*
fxval = func->f( params );
vcl_cout << "Minimized to " << fxval << vcl_endl
<< "Iterations: " << minimizer.get_num_iterations() << "; "
<< "Evaluations: " << minimizer.get_num_evaluations() << vcl_endl;
*/
}
vcl_cout << "Solution: " << params << vcl_endl;
return 0;
}
// Finds parameters b, n in the confidence region induced by confidence
// which maximize wblinv(c, b, n)
WeibullEstimates
estimate_weibull_upper_bounds(std::map<double,double> distribution,
const WeibullEstimates& best_params,
double confidence,
double c)
{
if (has_key(distribution, 0))
distribution[0.0001] += distribution[0];
distribution.erase(0);
if (distribution.size() < 5)
return best_params;
weibull_upper_bounds_function f(distribution, best_params, confidence, c);
// vnl_lbfgs minimizer(f);
vnl_powell minimizer(&f);
// vnl_conjugate_gradient minimizer(f);
vnl_vector<double> estimate(2);
estimate[0] = best_params.beta;
estimate[1] = best_params.etha;
ntk_assert(minimizer.minimize(estimate), "");
if (std::abs(estimate[0]-best_params.beta) > (best_params.beta*0.8)
|| std::abs(estimate[1]-best_params.etha) > (best_params.etha*0.8))
{
std::cerr << "Warning: algorithm did not converge." << std::endl;
return best_params;
}
WeibullEstimates estimates;
estimates.beta = estimate[0];
estimates.etha = estimate[1];
estimates.gamma = best_params.gamma;
estimates.logl = 0;
return estimates;
}
FunctionMinimum MnApplication::operator()(unsigned int maxfcn, double toler) {
assert(theState.isValid());
unsigned int npar = variableParameters();
// assert(npar > 0);
if(maxfcn == 0) maxfcn = 200 + 100*npar + 5*npar*npar;
FunctionMinimum min = minimizer().minimize( fcnbase(), theState, theStrategy, maxfcn, toler);
theNumCall += min.nfcn();
theState = min.userState();
return min;
}
void itk::BiExpFitFunctor::operator()(vnl_matrix<double> & newSignal,const vnl_matrix<double> & SignalMatrix, const double & S0)
{
vnl_vector<double> initalGuess(3);
// initialize Least Squres Function
// SignalMatrix.cols() defines the number of shells points
lestSquaresFunction model(SignalMatrix.cols());
model.set_bvalues(m_BValueList);// set BValue Vector e.g.: [1000, 2000, 3000] <- shell b Values
// initialize Levenberg Marquardt
vnl_levenberg_marquardt minimizer(model);
minimizer.set_max_function_evals(1000); // Iterations
minimizer.set_f_tolerance(1e-10); // Function tolerance
// for each Direction calculate LSF Coeffs ADC & AKC
for(unsigned int i = 0 ; i < SignalMatrix.rows(); i++)
{
model.set_measurements(SignalMatrix.get_row(i));
model.set_reference_measurement(S0);
initalGuess.put(0, 0.f); // ADC_slow
initalGuess.put(1, 0.009f); // ADC_fast
initalGuess.put(2, 0.7f); // lambda
// start Levenberg-Marquardt
minimizer.minimize_without_gradient(initalGuess);
const double & ADC_slow = initalGuess.get(0);
const double & ADC_fast = initalGuess.get(1);
const double & lambda = initalGuess(2);
newSignal.put(i, 0, S0 * (lambda * std::exp(-m_TargetBvalue * ADC_slow) + (1-lambda)* std::exp(-m_TargetBvalue * ADC_fast)));
newSignal.put(i, 1, minimizer.get_end_error()); // RMS Error
//OUTPUT FOR EVALUATION
/*std::cout << std::scientific << std::setprecision(5)
<< ADC_slow << "," // lambda
<< ADC_fast << "," // alpha
<< lambda << "," // lambda
<< S0 << "," // S0 value
<< minimizer.get_end_error() << ","; // End error
for(unsigned int j = 0; j < SignalMatrix.get_row(i).size(); j++ ){
std::cout << std::scientific << std::setprecision(5) << SignalMatrix.get_row(i)[j]; // S_n Values corresponding to shell 1 to shell n
if(j != SignalMatrix.get_row(i).size()-1) std::cout << ",";
}
std::cout << std::endl;*/
}
}
void estimate_weibull3(std::map<double,double> distribution,
WeibullEstimates& e)
{
if (has_key(distribution, 0))
distribution[0.0001] += distribution[0];
distribution.erase(0);
weibull3_likelihood_function f(distribution, e);
// vnl_lbfgs minimizer(f);
// vnl_powell minimizer(&f);
vnl_amoeba minimizer(f);
vnl_vector<double> estimate(1);
estimate[0] = e.gamma;
minimizer.minimize(estimate);
ntk_assert(flt_eq(e.gamma, estimate[0]), "");
}
int main() {
double q = 2.0;
double g0 = 2.0;
for(int i = 1, j = 0; i <= N; i = 2*(++j) + 1) { /* i in {2k+1} */
std::cout<<"#Gaussian: "<<i<<" Result: ";
GMinimizer minimizer(i); /*minimizer*/
const gsl_vector *result = &minimizer.solve(q, g0, 1.0e-6);
q = gsl_vector_get(result,0);
g0 =gsl_vector_get(result, 1);
std::cout<<"q = "<<q<< " g_0 = "<<g0<<" E_0 = "<<gsl_vector_get(result, 2)<<std::endl;
}
return EXIT_SUCCESS;
}
void TpsRegistration::StartRegistration(vnl_vector<double>& params) {
vnl_lbfgs minimizer(*func_);
func_->SetBase(this);
for (unsigned int k = 0; k < level_; ++k) {
func_->SetScale(v_scale_[k]);
func_->SetLambda(v_lambda_[k]);
bool b_fix_affine = (v_affine_[k] == 1);
func_->SetFixAffine(b_fix_affine);
func_->PrepareParamGradient();
SetParam(params);
int n_max_func_evals = v_func_evals_[k];
minimizer.set_max_function_evals(n_max_func_evals);
// For more options, see
// http://public.kitware.com/vxl/doc/release/core/vnl/html/vnl__nonlinear__minimizer_8h-source.html
minimizer.minimize(params);
if (minimizer.get_failure_code() < 0) {
break;
}
}
vcl_cout << "registration done" << vcl_endl;
}
void estimate_weibull(std::map<double,double> distribution,
WeibullEstimates& e)
{
if (has_key(distribution, 0))
distribution[0.0001] += distribution[0];
distribution.erase(0);
weibull_likelihood_function f(distribution, e.gamma);
// vnl_lbfgs minimizer(f);
// vnl_powell minimizer(&f);
vnl_amoeba minimizer(f);
vnl_vector<double> estimate(2);
estimate[0] = e.beta;
estimate[1] = e.etha;
// Q_ASSERT(minimizer.minimize(estimate));
minimizer.minimize(estimate);
WeibullEstimates estimates;
e.beta = estimate[0];
e.etha = estimate[1];
WeibullDistrib m (e.beta, e.etha, e.gamma);
e.logl = model_log_likelihood(distribution,
distrib_size(distribution),
m);
}
bool CamAugmentation::OptimizeCurrentView( int iter, double eps ){
// Get number parameters and observations:
int parameter_number = 6;
int observation_number = GetObservationNumber();
ls_minimizer2 minimizer(parameter_number);
minimizer.set_user_data(0, this);
minimizer.set_state_change_callback(updateCB);
minimizer.lm_set_max_iterations(iter);
// Feed with initial parameters:
GetParameterSet();
//FillLsqParams( lsqData, &v_opt_param[0], 0, 0 );
// Add the observations for ...
int test = 0;
for( int c = 0; c < (int)v_homography.size(); c++ ) // ... cameras
if( v_homography[c]->m_homography ){
int points = (int)v_homography[c]->v_point.size();
for( int point = 0; point < points; point++ ){ // ... points
PoseObs *o = new PoseObs();
o->cam = c;
o->point = point;
o->target[0] = v_homography[c]->v_point[point].u;
o->target[1] = v_homography[c]->v_point[point].v;
minimizer.add_observation(o, true, false);
}
}
minimizer.minimize_using_levenberg_marquardt_from(&v_opt_param[0]);
void *ptr = this;
updateCB(minimizer.state, &ptr);
return true;
}
void NormalModeMinimizer::run(int numTimesteps) {
if( numTimesteps < 1 )
return;
//check valid eigenvectors
if(*Q == NULL)
report << error << "No Eigenvectors for NormalMode integrator."<<endr;
//
preStepModify();
//remove last random pertubation
if(randforce) app->positions.intoSubtract(gaussRandCoord1);
//(*myPositions).intoWeightedAdd(-randStp,gaussRandCoord1);
//do minimization with local forces, max loop rediagOnMaxMinSteps, set subSpace minimization true
itrs = minimizer(minLim, rediagOnMaxMinSteps>0 ? rediagOnMaxMinSteps : 100 , simpleMin, reDiag, true, &forceCalc, &lastLambda, &app->energies, &app->positions, app->topology);
//flag excessive minimizations
if(itrs > 10) report << debug(1) << "[NormalModeMinimizer::run] iterations = " << itrs << "." << endr;
else report << debug(3) << "[NormalModeMinimizer::run] iterations = " << itrs << "." << endr;
avItrs += itrs;
//rediagonalize if minimization steps exceeds 'rediagOnMaxMinSteps'
if(reDiag && rediagOnMaxMinSteps > 0 && itrs >= rediagOnMaxMinSteps){
report << debug(1) << "[NormalModeMinimizer::run] Minimization steps ("
<< itrs << ") exceeded maximum (" << rediagOnMaxMinSteps << "), forcing re-diagonalize." << endr;
itrs = -1; //force re-diag
}
numSteps++;
avMinForceCalc += forceCalc;
report <<debug(5)<<"[NormalModeMinimizer::run] iterations = "<<itrs<<" average = "<<
(float)avItrs/(float)numSteps<<" force calcs = "<<forceCalc<<" average = "<<(float)avMinForceCalc/(float)numSteps<<endl;
if(randforce && itrs != -1 && lastLambda > 0){ //add random force, but not if rediagonalizing
//add random force
if(myPreviousNormalMode->genCompNoise) gaussRandCoord1 = myPreviousNormalMode->tempV3DBlk;
else genProjGauss(&gaussRandCoord1, app->topology);
//lambda = eUFactor / *eigValP; //user factor
randStp = sqrt(2 * Constant::BOLTZMANN * myTemp * lastLambda); //step length
//(*myPositions).intoWeightedAdd(randStp,gaussRandCoord1);
gaussRandCoord1.intoWeighted(randStp,gaussRandCoord1);
app->positions.intoAdd(gaussRandCoord1);
//additional random steps?
if(randforce > 1){
eUFactor = 0.5;
Real lambda;
for(int i=1;i<randforce;i++){
utilityCalculateForces();
nonSubspaceForce(myForces, myForces);
for(int j=0;j<_N;j++) (*myForces)[j] /= app->topology->atoms[j].scaledMass;
lambda = eUFactor / *eigValP; //user factor
//update positions
app->positions.intoWeightedAdd(lambda,*myForces);
gaussRandCoord1.intoWeightedAdd(lambda,*myForces); //add to grc1 so can be removed at the next step
//random force
genProjGauss(&gaussRandCoord2, app->topology);
randStp = sqrt(2 * Constant::BOLTZMANN * myTemp * lambda);
app->positions.intoWeightedAdd(randStp,gaussRandCoord2);
//add to grc1 so can be removed at the next step
gaussRandCoord1.intoWeightedAdd(randStp,gaussRandCoord2);
}
}
}else{
gaussRandCoord1.zero(-1);
//flag re-diagonalize if detected
if(itrs == -1) app->eigenInfo.reDiagonalize = true;
}
//
postStepModify();
}
