[91] | 1 | //#--------------------------------------------------------------------------- |
---|
[890] | 2 | //# Fitter.cc: A Fitter class for spectra |
---|
[91] | 3 | //#-------------------------------------------------------------------------- |
---|
| 4 | //# Copyright (C) 2004 |
---|
[125] | 5 | //# ATNF |
---|
[91] | 6 | //# |
---|
| 7 | //# This program is free software; you can redistribute it and/or modify it |
---|
| 8 | //# under the terms of the GNU General Public License as published by the Free |
---|
| 9 | //# Software Foundation; either version 2 of the License, or (at your option) |
---|
| 10 | //# any later version. |
---|
| 11 | //# |
---|
| 12 | //# This program is distributed in the hope that it will be useful, but |
---|
| 13 | //# WITHOUT ANY WARRANTY; without even the implied warranty of |
---|
| 14 | //# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General |
---|
| 15 | //# Public License for more details. |
---|
| 16 | //# |
---|
| 17 | //# You should have received a copy of the GNU General Public License along |
---|
| 18 | //# with this program; if not, write to the Free Software Foundation, Inc., |
---|
| 19 | //# 675 Massachusetts Ave, Cambridge, MA 02139, USA. |
---|
| 20 | //# |
---|
| 21 | //# Correspondence concerning this software should be addressed as follows: |
---|
| 22 | //# Internet email: Malte.Marquarding@csiro.au |
---|
| 23 | //# Postal address: Malte Marquarding, |
---|
| 24 | //# Australia Telescope National Facility, |
---|
| 25 | //# P.O. Box 76, |
---|
| 26 | //# Epping, NSW, 2121, |
---|
| 27 | //# AUSTRALIA |
---|
| 28 | //# |
---|
[891] | 29 | //# $Id: STFitter.cpp 1819 2010-08-02 07:28:20Z KanaSugimoto $ |
---|
[91] | 30 | //#--------------------------------------------------------------------------- |
---|
[125] | 31 | #include <casa/aips.h> |
---|
[91] | 32 | #include <casa/Arrays/ArrayMath.h> |
---|
| 33 | #include <casa/Arrays/ArrayLogical.h> |
---|
[1819] | 34 | #include <casa/Logging/LogIO.h> |
---|
[91] | 35 | #include <scimath/Fitting.h> |
---|
| 36 | #include <scimath/Fitting/LinearFit.h> |
---|
| 37 | #include <scimath/Functionals/CompiledFunction.h> |
---|
| 38 | #include <scimath/Functionals/CompoundFunction.h> |
---|
| 39 | #include <scimath/Functionals/Gaussian1D.h> |
---|
[1819] | 40 | #include "Lorentzian1D.h" |
---|
[91] | 41 | #include <scimath/Functionals/Polynomial.h> |
---|
| 42 | #include <scimath/Mathematics/AutoDiff.h> |
---|
| 43 | #include <scimath/Mathematics/AutoDiffMath.h> |
---|
| 44 | #include <scimath/Fitting/NonLinearFitLM.h> |
---|
| 45 | #include <components/SpectralComponents/SpectralEstimate.h> |
---|
| 46 | |
---|
[894] | 47 | #include "STFitter.h" |
---|
| 48 | |
---|
[91] | 49 | using namespace asap; |
---|
[125] | 50 | using namespace casa; |
---|
[91] | 51 | |
---|
[890] | 52 | Fitter::Fitter() |
---|
[91] | 53 | { |
---|
| 54 | } |
---|
| 55 | |
---|
[890] | 56 | Fitter::~Fitter() |
---|
[91] | 57 | { |
---|
[517] | 58 | reset(); |
---|
[91] | 59 | } |
---|
| 60 | |
---|
[890] | 61 | void Fitter::clear() |
---|
[91] | 62 | { |
---|
[517] | 63 | for (uInt i=0;i< funcs_.nelements();++i) { |
---|
| 64 | delete funcs_[i]; funcs_[i] = 0; |
---|
| 65 | } |
---|
[612] | 66 | funcs_.resize(0,True); |
---|
[517] | 67 | parameters_.resize(); |
---|
[1232] | 68 | fixedpar_.resize(); |
---|
[517] | 69 | error_.resize(); |
---|
| 70 | thefit_.resize(); |
---|
| 71 | estimate_.resize(); |
---|
| 72 | chisquared_ = 0.0; |
---|
[91] | 73 | } |
---|
[517] | 74 | |
---|
[890] | 75 | void Fitter::reset() |
---|
[91] | 76 | { |
---|
[517] | 77 | clear(); |
---|
| 78 | x_.resize(); |
---|
| 79 | y_.resize(); |
---|
| 80 | m_.resize(); |
---|
[91] | 81 | } |
---|
| 82 | |
---|
| 83 | |
---|
[890] | 84 | bool Fitter::computeEstimate() { |
---|
[517] | 85 | if (x_.nelements() == 0 || y_.nelements() == 0) |
---|
| 86 | throw (AipsError("No x/y data specified.")); |
---|
[91] | 87 | |
---|
[517] | 88 | if (dynamic_cast<Gaussian1D<Float>* >(funcs_[0]) == 0) |
---|
| 89 | return false; |
---|
| 90 | uInt n = funcs_.nelements(); |
---|
| 91 | SpectralEstimate estimator(n); |
---|
| 92 | estimator.setQ(5); |
---|
| 93 | Int mn,mx; |
---|
| 94 | mn = 0; |
---|
| 95 | mx = m_.nelements()-1; |
---|
| 96 | for (uInt i=0; i<m_.nelements();++i) { |
---|
| 97 | if (m_[i]) { |
---|
| 98 | mn = i; |
---|
| 99 | break; |
---|
[108] | 100 | } |
---|
[517] | 101 | } |
---|
| 102 | for (uInt j=m_.nelements()-1; j>=0;--j) { |
---|
| 103 | if (m_[j]) { |
---|
| 104 | mx = j; |
---|
| 105 | break; |
---|
[108] | 106 | } |
---|
[517] | 107 | } |
---|
[1067] | 108 | //mn = 0+x_.nelements()/10; |
---|
| 109 | //mx = x_.nelements()-x_.nelements()/10; |
---|
[517] | 110 | estimator.setRegion(mn,mx); |
---|
| 111 | //estimator.setWindowing(True); |
---|
| 112 | SpectralList listGauss = estimator.estimate(x_, y_); |
---|
| 113 | parameters_.resize(n*3); |
---|
| 114 | Gaussian1D<Float>* g = 0; |
---|
| 115 | for (uInt i=0; i<n;i++) { |
---|
| 116 | g = dynamic_cast<Gaussian1D<Float>* >(funcs_[i]); |
---|
| 117 | if (g) { |
---|
| 118 | (*g)[0] = listGauss[i].getAmpl(); |
---|
| 119 | (*g)[1] = listGauss[i].getCenter(); |
---|
| 120 | (*g)[2] = listGauss[i].getFWHM(); |
---|
[91] | 121 | } |
---|
[517] | 122 | } |
---|
| 123 | estimate_.resize(); |
---|
| 124 | listGauss.evaluate(estimate_,x_); |
---|
| 125 | return true; |
---|
[91] | 126 | } |
---|
| 127 | |
---|
[890] | 128 | std::vector<float> Fitter::getEstimate() const |
---|
[91] | 129 | { |
---|
[517] | 130 | if (estimate_.nelements() == 0) |
---|
| 131 | throw (AipsError("No estimate set.")); |
---|
| 132 | std::vector<float> stlout; |
---|
| 133 | estimate_.tovector(stlout); |
---|
| 134 | return stlout; |
---|
[91] | 135 | } |
---|
| 136 | |
---|
| 137 | |
---|
[890] | 138 | bool Fitter::setExpression(const std::string& expr, int ncomp) |
---|
[91] | 139 | { |
---|
[517] | 140 | clear(); |
---|
| 141 | if (expr == "gauss") { |
---|
| 142 | if (ncomp < 1) throw (AipsError("Need at least one gaussian to fit.")); |
---|
| 143 | funcs_.resize(ncomp); |
---|
| 144 | for (Int k=0; k<ncomp; ++k) { |
---|
| 145 | funcs_[k] = new Gaussian1D<Float>(); |
---|
| 146 | } |
---|
| 147 | } else if (expr == "poly") { |
---|
| 148 | funcs_.resize(1); |
---|
| 149 | funcs_[0] = new Polynomial<Float>(ncomp); |
---|
[1819] | 150 | } else if (expr == "lorentz") { |
---|
| 151 | if (ncomp < 1) throw (AipsError("Need at least one lorentzian to fit.")); |
---|
| 152 | funcs_.resize(ncomp); |
---|
| 153 | for (Int k=0; k<ncomp; ++k) { |
---|
| 154 | funcs_[k] = new Lorentzian1D<Float>(); |
---|
| 155 | } |
---|
[517] | 156 | } else { |
---|
[1819] | 157 | //cerr << " compiled functions not yet implemented" << endl; |
---|
| 158 | LogIO os( LogOrigin( "Fitter", "setExpression()", WHERE ) ) ; |
---|
| 159 | os << LogIO::WARN << " compiled functions not yet implemented" << LogIO::POST; |
---|
[517] | 160 | //funcs_.resize(1); |
---|
| 161 | //funcs_[0] = new CompiledFunction<Float>(); |
---|
| 162 | //funcs_[0]->setFunction(String(expr)); |
---|
| 163 | return false; |
---|
| 164 | } |
---|
| 165 | return true; |
---|
[91] | 166 | } |
---|
| 167 | |
---|
[890] | 168 | bool Fitter::setData(std::vector<float> absc, std::vector<float> spec, |
---|
[91] | 169 | std::vector<bool> mask) |
---|
| 170 | { |
---|
| 171 | x_.resize(); |
---|
| 172 | y_.resize(); |
---|
| 173 | m_.resize(); |
---|
| 174 | // convert std::vector to casa Vector |
---|
| 175 | Vector<Float> tmpx(absc); |
---|
| 176 | Vector<Float> tmpy(spec); |
---|
| 177 | Vector<Bool> tmpm(mask); |
---|
| 178 | AlwaysAssert(tmpx.nelements() == tmpy.nelements(), AipsError); |
---|
| 179 | x_ = tmpx; |
---|
| 180 | y_ = tmpy; |
---|
| 181 | m_ = tmpm; |
---|
| 182 | return true; |
---|
| 183 | } |
---|
| 184 | |
---|
[890] | 185 | std::vector<float> Fitter::getResidual() const |
---|
[91] | 186 | { |
---|
| 187 | if (residual_.nelements() == 0) |
---|
| 188 | throw (AipsError("Function not yet fitted.")); |
---|
| 189 | std::vector<float> stlout; |
---|
| 190 | residual_.tovector(stlout); |
---|
| 191 | return stlout; |
---|
| 192 | } |
---|
| 193 | |
---|
[890] | 194 | std::vector<float> Fitter::getFit() const |
---|
[91] | 195 | { |
---|
| 196 | Vector<Float> out = thefit_; |
---|
| 197 | std::vector<float> stlout; |
---|
| 198 | out.tovector(stlout); |
---|
| 199 | return stlout; |
---|
| 200 | |
---|
| 201 | } |
---|
| 202 | |
---|
[890] | 203 | std::vector<float> Fitter::getErrors() const |
---|
[91] | 204 | { |
---|
| 205 | Vector<Float> out = error_; |
---|
| 206 | std::vector<float> stlout; |
---|
| 207 | out.tovector(stlout); |
---|
| 208 | return stlout; |
---|
| 209 | } |
---|
| 210 | |
---|
[890] | 211 | bool Fitter::setParameters(std::vector<float> params) |
---|
[91] | 212 | { |
---|
| 213 | Vector<Float> tmppar(params); |
---|
| 214 | if (funcs_.nelements() == 0) |
---|
| 215 | throw (AipsError("Function not yet set.")); |
---|
| 216 | if (parameters_.nelements() > 0 && tmppar.nelements() != parameters_.nelements()) |
---|
| 217 | throw (AipsError("Number of parameters inconsistent with function.")); |
---|
[1232] | 218 | if (parameters_.nelements() == 0) { |
---|
[91] | 219 | parameters_.resize(tmppar.nelements()); |
---|
[1232] | 220 | if (tmppar.nelements() != fixedpar_.nelements()) { |
---|
| 221 | fixedpar_.resize(tmppar.nelements()); |
---|
| 222 | fixedpar_ = False; |
---|
| 223 | } |
---|
| 224 | } |
---|
[91] | 225 | if (dynamic_cast<Gaussian1D<Float>* >(funcs_[0]) != 0) { |
---|
| 226 | uInt count = 0; |
---|
| 227 | for (uInt j=0; j < funcs_.nelements(); ++j) { |
---|
| 228 | for (uInt i=0; i < funcs_[j]->nparameters(); ++i) { |
---|
| 229 | (funcs_[j]->parameters())[i] = tmppar[count]; |
---|
| 230 | parameters_[count] = tmppar[count]; |
---|
| 231 | ++count; |
---|
| 232 | } |
---|
| 233 | } |
---|
| 234 | } else if (dynamic_cast<Polynomial<Float>* >(funcs_[0]) != 0) { |
---|
| 235 | for (uInt i=0; i < funcs_[0]->nparameters(); ++i) { |
---|
| 236 | parameters_[i] = tmppar[i]; |
---|
| 237 | (funcs_[0]->parameters())[i] = tmppar[i]; |
---|
| 238 | } |
---|
[1819] | 239 | } else if (dynamic_cast<Lorentzian1D<Float>* >(funcs_[0]) != 0) { |
---|
| 240 | uInt count = 0; |
---|
| 241 | for (uInt j=0; j < funcs_.nelements(); ++j) { |
---|
| 242 | for (uInt i=0; i < funcs_[j]->nparameters(); ++i) { |
---|
| 243 | (funcs_[j]->parameters())[i] = tmppar[count]; |
---|
| 244 | parameters_[count] = tmppar[count]; |
---|
| 245 | ++count; |
---|
| 246 | } |
---|
| 247 | } |
---|
[91] | 248 | } |
---|
[1232] | 249 | // reset |
---|
| 250 | if (params.size() == 0) { |
---|
| 251 | parameters_.resize(); |
---|
| 252 | fixedpar_.resize(); |
---|
| 253 | } |
---|
[91] | 254 | return true; |
---|
| 255 | } |
---|
| 256 | |
---|
[890] | 257 | bool Fitter::setFixedParameters(std::vector<bool> fixed) |
---|
[91] | 258 | { |
---|
| 259 | if (funcs_.nelements() == 0) |
---|
| 260 | throw (AipsError("Function not yet set.")); |
---|
[1232] | 261 | if (fixedpar_.nelements() > 0 && fixed.size() != fixedpar_.nelements()) |
---|
[91] | 262 | throw (AipsError("Number of mask elements inconsistent with function.")); |
---|
[1232] | 263 | if (fixedpar_.nelements() == 0) { |
---|
| 264 | fixedpar_.resize(parameters_.nelements()); |
---|
| 265 | fixedpar_ = False; |
---|
| 266 | } |
---|
[91] | 267 | if (dynamic_cast<Gaussian1D<Float>* >(funcs_[0]) != 0) { |
---|
| 268 | uInt count = 0; |
---|
| 269 | for (uInt j=0; j < funcs_.nelements(); ++j) { |
---|
| 270 | for (uInt i=0; i < funcs_[j]->nparameters(); ++i) { |
---|
[1232] | 271 | funcs_[j]->mask(i) = !fixed[count]; |
---|
| 272 | fixedpar_[count] = fixed[count]; |
---|
[91] | 273 | ++count; |
---|
| 274 | } |
---|
| 275 | } |
---|
| 276 | } else if (dynamic_cast<Polynomial<Float>* >(funcs_[0]) != 0) { |
---|
| 277 | for (uInt i=0; i < funcs_[0]->nparameters(); ++i) { |
---|
[1232] | 278 | fixedpar_[i] = fixed[i]; |
---|
| 279 | funcs_[0]->mask(i) = !fixed[i]; |
---|
[91] | 280 | } |
---|
[1819] | 281 | } else if (dynamic_cast<Lorentzian1D<Float>* >(funcs_[0]) != 0) { |
---|
| 282 | uInt count = 0; |
---|
| 283 | for (uInt j=0; j < funcs_.nelements(); ++j) { |
---|
| 284 | for (uInt i=0; i < funcs_[j]->nparameters(); ++i) { |
---|
| 285 | funcs_[j]->mask(i) = !fixed[count]; |
---|
| 286 | fixedpar_[count] = fixed[count]; |
---|
| 287 | ++count; |
---|
| 288 | } |
---|
| 289 | } |
---|
[91] | 290 | } |
---|
| 291 | return true; |
---|
| 292 | } |
---|
| 293 | |
---|
[890] | 294 | std::vector<float> Fitter::getParameters() const { |
---|
[91] | 295 | Vector<Float> out = parameters_; |
---|
| 296 | std::vector<float> stlout; |
---|
| 297 | out.tovector(stlout); |
---|
| 298 | return stlout; |
---|
| 299 | } |
---|
| 300 | |
---|
[890] | 301 | std::vector<bool> Fitter::getFixedParameters() const { |
---|
[108] | 302 | Vector<Bool> out(parameters_.nelements()); |
---|
| 303 | if (fixedpar_.nelements() == 0) { |
---|
[1232] | 304 | return std::vector<bool>(); |
---|
[108] | 305 | //throw (AipsError("No parameter mask set.")); |
---|
| 306 | } else { |
---|
| 307 | out = fixedpar_; |
---|
| 308 | } |
---|
| 309 | std::vector<bool> stlout; |
---|
| 310 | out.tovector(stlout); |
---|
| 311 | return stlout; |
---|
[91] | 312 | } |
---|
| 313 | |
---|
[890] | 314 | float Fitter::getChisquared() const { |
---|
[91] | 315 | return chisquared_; |
---|
| 316 | } |
---|
| 317 | |
---|
[890] | 318 | bool Fitter::fit() { |
---|
[517] | 319 | NonLinearFitLM<Float> fitter; |
---|
| 320 | CompoundFunction<Float> func; |
---|
[612] | 321 | |
---|
| 322 | uInt n = funcs_.nelements(); |
---|
[517] | 323 | for (uInt i=0; i<n; ++i) { |
---|
| 324 | func.addFunction(*funcs_[i]); |
---|
| 325 | } |
---|
[612] | 326 | |
---|
[517] | 327 | fitter.setFunction(func); |
---|
| 328 | fitter.setMaxIter(50+n*10); |
---|
| 329 | // Convergence criterium |
---|
| 330 | fitter.setCriteria(0.001); |
---|
[612] | 331 | |
---|
[517] | 332 | // Fit |
---|
| 333 | Vector<Float> sigma(x_.nelements()); |
---|
| 334 | sigma = 1.0; |
---|
[890] | 335 | |
---|
[517] | 336 | parameters_.resize(); |
---|
| 337 | parameters_ = fitter.fit(x_, y_, sigma, &m_); |
---|
[1067] | 338 | if ( !fitter.converged() ) { |
---|
| 339 | return false; |
---|
| 340 | } |
---|
[517] | 341 | std::vector<float> ps; |
---|
| 342 | parameters_.tovector(ps); |
---|
| 343 | setParameters(ps); |
---|
[612] | 344 | |
---|
[517] | 345 | error_.resize(); |
---|
| 346 | error_ = fitter.errors(); |
---|
[612] | 347 | |
---|
[517] | 348 | chisquared_ = fitter.getChi2(); |
---|
[890] | 349 | |
---|
[517] | 350 | residual_.resize(); |
---|
| 351 | residual_ = y_; |
---|
| 352 | fitter.residual(residual_,x_); |
---|
| 353 | // use fitter.residual(model=True) to get the model |
---|
| 354 | thefit_.resize(x_.nelements()); |
---|
| 355 | fitter.residual(thefit_,x_,True); |
---|
| 356 | return true; |
---|
| 357 | } |
---|
[483] | 358 | |
---|
[1391] | 359 | bool Fitter::lfit() { |
---|
| 360 | LinearFit<Float> fitter; |
---|
| 361 | CompoundFunction<Float> func; |
---|
[483] | 362 | |
---|
[1391] | 363 | uInt n = funcs_.nelements(); |
---|
| 364 | for (uInt i=0; i<n; ++i) { |
---|
| 365 | func.addFunction(*funcs_[i]); |
---|
| 366 | } |
---|
| 367 | |
---|
| 368 | fitter.setFunction(func); |
---|
| 369 | //fitter.setMaxIter(50+n*10); |
---|
| 370 | // Convergence criterium |
---|
| 371 | //fitter.setCriteria(0.001); |
---|
| 372 | |
---|
| 373 | // Fit |
---|
| 374 | Vector<Float> sigma(x_.nelements()); |
---|
| 375 | sigma = 1.0; |
---|
| 376 | |
---|
| 377 | parameters_.resize(); |
---|
| 378 | parameters_ = fitter.fit(x_, y_, sigma, &m_); |
---|
| 379 | std::vector<float> ps; |
---|
| 380 | parameters_.tovector(ps); |
---|
| 381 | setParameters(ps); |
---|
| 382 | |
---|
| 383 | error_.resize(); |
---|
| 384 | error_ = fitter.errors(); |
---|
| 385 | |
---|
| 386 | chisquared_ = fitter.getChi2(); |
---|
| 387 | |
---|
| 388 | residual_.resize(); |
---|
| 389 | residual_ = y_; |
---|
| 390 | fitter.residual(residual_,x_); |
---|
| 391 | // use fitter.residual(model=True) to get the model |
---|
| 392 | thefit_.resize(x_.nelements()); |
---|
| 393 | fitter.residual(thefit_,x_,True); |
---|
| 394 | return true; |
---|
| 395 | } |
---|
| 396 | |
---|
[890] | 397 | std::vector<float> Fitter::evaluate(int whichComp) const |
---|
| 398 | { |
---|
[517] | 399 | std::vector<float> stlout; |
---|
[890] | 400 | uInt idx = uInt(whichComp); |
---|
[517] | 401 | Float y; |
---|
| 402 | if ( idx < funcs_.nelements() ) { |
---|
| 403 | for (uInt i=0; i<x_.nelements(); ++i) { |
---|
| 404 | y = (*funcs_[idx])(x_[i]); |
---|
| 405 | stlout.push_back(float(y)); |
---|
| 406 | } |
---|
| 407 | } |
---|
| 408 | return stlout; |
---|
| 409 | } |
---|
[483] | 410 | |
---|