source: trunk/src/STFitter.cpp@ 2549

Last change on this file since 2549 was 2455, checked in by Malte Marquarding, 13 years ago

casacore trunk has now been updated to use new casapy's component module
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 13.0 KB
Line 
1//#---------------------------------------------------------------------------
2//# Fitter.cc: A Fitter class for spectra
3//#--------------------------------------------------------------------------
4//# Copyright (C) 2004-2012
5//# ATNF
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//#
29//# $Id: STFitter.cpp 2455 2012-04-03 06:26:04Z MalteMarquarding $
30//#---------------------------------------------------------------------------
31#include <casa/aips.h>
32#include <casa/Arrays/ArrayMath.h>
33#include <casa/Arrays/ArrayLogical.h>
34#include <casa/Logging/LogIO.h>
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>
40#include <scimath/Functionals/Lorentzian1D.h>
41#include <scimath/Functionals/Sinusoid1D.h>
42#include <scimath/Functionals/Polynomial.h>
43#include <scimath/Mathematics/AutoDiff.h>
44#include <scimath/Mathematics/AutoDiffMath.h>
45#include <scimath/Fitting/NonLinearFitLM.h>
46#include <components/SpectralComponents/SpectralEstimate.h>
47
48#include "STFitter.h"
49
50using namespace asap;
51using namespace casa;
52
53Fitter::Fitter()
54{
55}
56
57Fitter::~Fitter()
58{
59 reset();
60}
61
62void Fitter::clear()
63{
64 for (uInt i=0;i< funcs_.nelements();++i) {
65 delete funcs_[i]; funcs_[i] = 0;
66 }
67 funcs_.resize(0,True);
68 parameters_.resize();
69 fixedpar_.resize();
70 error_.resize();
71 thefit_.resize();
72 estimate_.resize();
73 chisquared_ = 0.0;
74}
75
76void Fitter::reset()
77{
78 clear();
79 x_.resize();
80 y_.resize();
81 m_.resize();
82}
83
84
85bool Fitter::computeEstimate() {
86 if (x_.nelements() == 0 || y_.nelements() == 0)
87 throw (AipsError("No x/y data specified."));
88
89 if (dynamic_cast<Gaussian1D<Float>* >(funcs_[0]) == 0)
90 return false;
91 uInt n = funcs_.nelements();
92 SpectralEstimate estimator(n);
93 estimator.setQ(5);
94 Int mn,mx;
95 mn = 0;
96 mx = m_.nelements()-1;
97 for (uInt i=0; i<m_.nelements();++i) {
98 if (m_[i]) {
99 mn = i;
100 break;
101 }
102 }
103 // use Int to suppress compiler warning
104 for (Int j=m_.nelements()-1; j>=0;--j) {
105 if (m_[j]) {
106 mx = j;
107 break;
108 }
109 }
110 //mn = 0+x_.nelements()/10;
111 //mx = x_.nelements()-x_.nelements()/10;
112 estimator.setRegion(mn,mx);
113 //estimator.setWindowing(True);
114 SpectralList listGauss = estimator.estimate(x_, y_);
115 parameters_.resize(n*3);
116 Gaussian1D<Float>* g = 0;
117 for (uInt i=0; i<n;i++) {
118 g = dynamic_cast<Gaussian1D<Float>* >(funcs_[i]);
119 if (g) {
120 const GaussianSpectralElement *gauss =
121 dynamic_cast<const GaussianSpectralElement *>(listGauss[i]) ;
122 (*g)[0] = gauss->getAmpl();
123 (*g)[1] = gauss->getCenter();
124 (*g)[2] = gauss->getFWHM();
125 /*
126 (*g)[0] = listGauss[i].getAmpl();
127 (*g)[1] = listGauss[i].getCenter();
128 (*g)[2] = listGauss[i].getFWHM();
129 */
130 }
131 }
132 estimate_.resize();
133 listGauss.evaluate(estimate_,x_);
134 return true;
135}
136
137std::vector<float> Fitter::getEstimate() const
138{
139 if (estimate_.nelements() == 0)
140 throw (AipsError("No estimate set."));
141 std::vector<float> stlout;
142 estimate_.tovector(stlout);
143 return stlout;
144}
145
146
147bool Fitter::setExpression(const std::string& expr, int ncomp)
148{
149 clear();
150 if (expr == "gauss") {
151 if (ncomp < 1) throw (AipsError("Need at least one gaussian to fit."));
152 funcs_.resize(ncomp);
153 funcnames_.clear();
154 funccomponents_.clear();
155 for (Int k=0; k<ncomp; ++k) {
156 funcs_[k] = new Gaussian1D<Float>();
157 funcnames_.push_back(expr);
158 funccomponents_.push_back(3);
159 }
160 } else if (expr == "lorentz") {
161 if (ncomp < 1) throw (AipsError("Need at least one lorentzian to fit."));
162 funcs_.resize(ncomp);
163 funcnames_.clear();
164 funccomponents_.clear();
165 for (Int k=0; k<ncomp; ++k) {
166 funcs_[k] = new Lorentzian1D<Float>();
167 funcnames_.push_back(expr);
168 funccomponents_.push_back(3);
169 }
170 } else if (expr == "sinusoid") {
171 if (ncomp < 1) throw (AipsError("Need at least one sinusoid to fit."));
172 funcs_.resize(ncomp);
173 funcnames_.clear();
174 funccomponents_.clear();
175 for (Int k=0; k<ncomp; ++k) {
176 funcs_[k] = new Sinusoid1D<Float>();
177 funcnames_.push_back(expr);
178 funccomponents_.push_back(3);
179 }
180 } else if (expr == "poly") {
181 funcs_.resize(1);
182 funcnames_.clear();
183 funccomponents_.clear();
184 funcs_[0] = new Polynomial<Float>(ncomp);
185 funcnames_.push_back(expr);
186 funccomponents_.push_back(ncomp);
187 } else {
188 LogIO os( LogOrigin( "Fitter", "setExpression()", WHERE ) ) ;
189 os << LogIO::WARN << " compiled functions not yet implemented" << LogIO::POST;
190 //funcs_.resize(1);
191 //funcs_[0] = new CompiledFunction<Float>();
192 //funcs_[0]->setFunction(String(expr));
193 return false;
194 }
195 return true;
196}
197
198bool Fitter::setData(std::vector<float> absc, std::vector<float> spec,
199 std::vector<bool> mask)
200{
201 x_.resize();
202 y_.resize();
203 m_.resize();
204 // convert std::vector to casa Vector
205 Vector<Float> tmpx(absc);
206 Vector<Float> tmpy(spec);
207 Vector<Bool> tmpm(mask);
208 AlwaysAssert(tmpx.nelements() == tmpy.nelements(), AipsError);
209 x_ = tmpx;
210 y_ = tmpy;
211 m_ = tmpm;
212 return true;
213}
214
215std::vector<float> Fitter::getResidual() const
216{
217 if (residual_.nelements() == 0)
218 throw (AipsError("Function not yet fitted."));
219 std::vector<float> stlout;
220 residual_.tovector(stlout);
221 return stlout;
222}
223
224std::vector<float> Fitter::getFit() const
225{
226 Vector<Float> out = thefit_;
227 std::vector<float> stlout;
228 out.tovector(stlout);
229 return stlout;
230
231}
232
233std::vector<float> Fitter::getErrors() const
234{
235 Vector<Float> out = error_;
236 std::vector<float> stlout;
237 out.tovector(stlout);
238 return stlout;
239}
240
241bool Fitter::setParameters(std::vector<float> params)
242{
243 Vector<Float> tmppar(params);
244 if (funcs_.nelements() == 0)
245 throw (AipsError("Function not yet set."));
246 if (parameters_.nelements() > 0 && tmppar.nelements() != parameters_.nelements())
247 throw (AipsError("Number of parameters inconsistent with function."));
248 if (parameters_.nelements() == 0) {
249 parameters_.resize(tmppar.nelements());
250 if (tmppar.nelements() != fixedpar_.nelements()) {
251 fixedpar_.resize(tmppar.nelements());
252 fixedpar_ = False;
253 }
254 }
255 if (dynamic_cast<Gaussian1D<Float>* >(funcs_[0]) != 0) {
256 uInt count = 0;
257 for (uInt j=0; j < funcs_.nelements(); ++j) {
258 for (uInt i=0; i < funcs_[j]->nparameters(); ++i) {
259 (funcs_[j]->parameters())[i] = tmppar[count];
260 parameters_[count] = tmppar[count];
261 ++count;
262 }
263 }
264 } else if (dynamic_cast<Lorentzian1D<Float>* >(funcs_[0]) != 0) {
265 uInt count = 0;
266 for (uInt j=0; j < funcs_.nelements(); ++j) {
267 for (uInt i=0; i < funcs_[j]->nparameters(); ++i) {
268 (funcs_[j]->parameters())[i] = tmppar[count];
269 parameters_[count] = tmppar[count];
270 ++count;
271 }
272 }
273 } else if (dynamic_cast<Sinusoid1D<Float>* >(funcs_[0]) != 0) {
274 uInt count = 0;
275 for (uInt j=0; j < funcs_.nelements(); ++j) {
276 for (uInt i=0; i < funcs_[j]->nparameters(); ++i) {
277 (funcs_[j]->parameters())[i] = tmppar[count];
278 parameters_[count] = tmppar[count];
279 ++count;
280 }
281 }
282 } else if (dynamic_cast<Polynomial<Float>* >(funcs_[0]) != 0) {
283 for (uInt i=0; i < funcs_[0]->nparameters(); ++i) {
284 parameters_[i] = tmppar[i];
285 (funcs_[0]->parameters())[i] = tmppar[i];
286 }
287 }
288 // reset
289 if (params.size() == 0) {
290 parameters_.resize();
291 fixedpar_.resize();
292 }
293 return true;
294}
295
296bool Fitter::setFixedParameters(std::vector<bool> fixed)
297{
298 if (funcs_.nelements() == 0)
299 throw (AipsError("Function not yet set."));
300 if (fixedpar_.nelements() > 0 && fixed.size() != fixedpar_.nelements())
301 throw (AipsError("Number of mask elements inconsistent with function."));
302 if (fixedpar_.nelements() == 0) {
303 fixedpar_.resize(parameters_.nelements());
304 fixedpar_ = False;
305 }
306 if (dynamic_cast<Gaussian1D<Float>* >(funcs_[0]) != 0) {
307 uInt count = 0;
308 for (uInt j=0; j < funcs_.nelements(); ++j) {
309 for (uInt i=0; i < funcs_[j]->nparameters(); ++i) {
310 funcs_[j]->mask(i) = !fixed[count];
311 fixedpar_[count] = fixed[count];
312 ++count;
313 }
314 }
315 } else if (dynamic_cast<Lorentzian1D<Float>* >(funcs_[0]) != 0) {
316 uInt count = 0;
317 for (uInt j=0; j < funcs_.nelements(); ++j) {
318 for (uInt i=0; i < funcs_[j]->nparameters(); ++i) {
319 funcs_[j]->mask(i) = !fixed[count];
320 fixedpar_[count] = fixed[count];
321 ++count;
322 }
323 }
324 } else if (dynamic_cast<Sinusoid1D<Float>* >(funcs_[0]) != 0) {
325 uInt count = 0;
326 for (uInt j=0; j < funcs_.nelements(); ++j) {
327 for (uInt i=0; i < funcs_[j]->nparameters(); ++i) {
328 funcs_[j]->mask(i) = !fixed[count];
329 fixedpar_[count] = fixed[count];
330 ++count;
331 }
332 }
333 } else if (dynamic_cast<Polynomial<Float>* >(funcs_[0]) != 0) {
334 for (uInt i=0; i < funcs_[0]->nparameters(); ++i) {
335 fixedpar_[i] = fixed[i];
336 funcs_[0]->mask(i) = !fixed[i];
337 }
338 }
339 return true;
340}
341
342std::vector<float> Fitter::getParameters() const {
343 Vector<Float> out = parameters_;
344 std::vector<float> stlout;
345 out.tovector(stlout);
346 return stlout;
347}
348
349std::vector<bool> Fitter::getFixedParameters() const {
350 Vector<Bool> out(parameters_.nelements());
351 if (fixedpar_.nelements() == 0) {
352 return std::vector<bool>();
353 //throw (AipsError("No parameter mask set."));
354 } else {
355 out = fixedpar_;
356 }
357 std::vector<bool> stlout;
358 out.tovector(stlout);
359 return stlout;
360}
361
362float Fitter::getChisquared() const {
363 return chisquared_;
364}
365
366bool Fitter::fit() {
367 NonLinearFitLM<Float> fitter;
368 CompoundFunction<Float> func;
369
370 uInt n = funcs_.nelements();
371 for (uInt i=0; i<n; ++i) {
372 func.addFunction(*funcs_[i]);
373 }
374
375 fitter.setFunction(func);
376 fitter.setMaxIter(50+n*10);
377 // Convergence criterium
378 fitter.setCriteria(0.001);
379
380 // Fit
381 Vector<Float> sigma(x_.nelements());
382 sigma = 1.0;
383
384 parameters_.resize();
385 parameters_ = fitter.fit(x_, y_, sigma, &m_);
386 if ( !fitter.converged() ) {
387 return false;
388 }
389 std::vector<float> ps;
390 parameters_.tovector(ps);
391 setParameters(ps);
392
393 error_.resize();
394 error_ = fitter.errors();
395
396 chisquared_ = fitter.getChi2();
397
398 residual_.resize();
399 residual_ = y_;
400 fitter.residual(residual_,x_);
401 // use fitter.residual(model=True) to get the model
402 thefit_.resize(x_.nelements());
403 fitter.residual(thefit_,x_,True);
404 return true;
405}
406
407bool Fitter::lfit() {
408 LinearFit<Float> fitter;
409 CompoundFunction<Float> func;
410
411 uInt n = funcs_.nelements();
412 for (uInt i=0; i<n; ++i) {
413 func.addFunction(*funcs_[i]);
414 }
415
416 fitter.setFunction(func);
417 //fitter.setMaxIter(50+n*10);
418 // Convergence criterium
419 //fitter.setCriteria(0.001);
420
421 // Fit
422 Vector<Float> sigma(x_.nelements());
423 sigma = 1.0;
424
425 parameters_.resize();
426 parameters_ = fitter.fit(x_, y_, sigma, &m_);
427 std::vector<float> ps;
428 parameters_.tovector(ps);
429 setParameters(ps);
430
431 error_.resize();
432 error_ = fitter.errors();
433
434 chisquared_ = fitter.getChi2();
435
436 residual_.resize();
437 residual_ = y_;
438 fitter.residual(residual_,x_);
439 // use fitter.residual(model=True) to get the model
440 thefit_.resize(x_.nelements());
441 fitter.residual(thefit_,x_,True);
442 return true;
443}
444
445std::vector<float> Fitter::evaluate(int whichComp) const
446{
447 std::vector<float> stlout;
448 uInt idx = uInt(whichComp);
449 Float y;
450 if ( idx < funcs_.nelements() ) {
451 for (uInt i=0; i<x_.nelements(); ++i) {
452 y = (*funcs_[idx])(x_[i]);
453 stlout.push_back(float(y));
454 }
455 }
456 return stlout;
457}
458
459STFitEntry Fitter::getFitEntry() const
460{
461 STFitEntry fit;
462 fit.setParameters(getParameters());
463 fit.setErrors(getErrors());
464 fit.setComponents(funccomponents_);
465 fit.setFunctions(funcnames_);
466 fit.setParmasks(getFixedParameters());
467 return fit;
468}
Note: See TracBrowser for help on using the repository browser.