source: trunk/src/SDFitter.cc@ 241

Last change on this file since 241 was 125, checked in by mar637, 20 years ago

Moved to casa namespace.
Adjusted the copyright to be ATNF.
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 10.2 KB
Line 
1//#---------------------------------------------------------------------------
2//# SDFitter.cc: A Fitter class for spectra
3//#--------------------------------------------------------------------------
4//# Copyright (C) 2004
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:
30//#---------------------------------------------------------------------------
31#include <casa/aips.h>
32#include <casa/Arrays/ArrayMath.h>
33#include <casa/Arrays/ArrayLogical.h>
34#include <scimath/Fitting.h>
35#include <scimath/Fitting/LinearFit.h>
36#include <scimath/Functionals/CompiledFunction.h>
37#include <scimath/Functionals/CompoundFunction.h>
38#include <scimath/Functionals/Gaussian1D.h>
39#include <scimath/Functionals/Polynomial.h>
40#include <scimath/Mathematics/AutoDiff.h>
41#include <scimath/Mathematics/AutoDiffMath.h>
42#include <scimath/Fitting/NonLinearFitLM.h>
43#include <components/SpectralComponents/SpectralEstimate.h>
44
45#include "SDFitter.h"
46using namespace asap;
47using namespace casa;
48
49SDFitter::SDFitter()
50{
51}
52
53SDFitter::~SDFitter()
54{
55 reset();
56}
57
58void SDFitter::clear()
59{
60 for (uInt i=0;i< funcs_.nelements();++i) {
61 delete funcs_[i]; funcs_[i] = 0;
62 };
63 funcs_.resize(0, True);
64 parameters_.resize();
65 error_.resize();
66 thefit_.resize();
67 estimate_.resize();
68 chisquared_ = 0.0;
69}
70void SDFitter::reset()
71{
72 clear();
73 x_.resize();
74 y_.resize();
75 m_.resize();
76}
77
78
79bool SDFitter::computeEstimate() {
80 if (x_.nelements() == 0 || y_.nelements() == 0)
81 throw (AipsError("No x/y data specified."));
82
83 if (dynamic_cast<Gaussian1D<Float>* >(funcs_[0]) == 0)
84 return false;
85 uInt n = funcs_.nelements();
86 SpectralEstimate estimator(n);
87 estimator.setQ(5);
88 Int mn,mx;
89 mn = 0;
90 mx = m_.nelements()-1;
91 for (uInt i=0; i<m_.nelements();++i) {
92 if (m_[i]) {
93 mn = i;
94 break;
95 }
96 }
97 for (uInt j=m_.nelements()-1; j>=0;--j) {
98 if (m_[j]) {
99 mx = j;
100 break;
101 }
102 }
103 //mn = 0+x_.nelements()/10;
104 //mx = x_.nelements()-x_.nelements()/10;
105 estimator.setRegion(mn,mx);
106 //estimator.setWindowing(True);
107 SpectralList listGauss = estimator.estimate(x_, y_);
108 Gaussian1D<Float>* g;
109 parameters_.resize(n*3);
110 uInt count = 0;
111 for (uInt i=0; i<n;i++) {
112 g = dynamic_cast<Gaussian1D<Float>* >(funcs_[i]);
113 if (g) {
114 (*g)[0] = listGauss[i].getAmpl();
115 (*g)[1] = listGauss[i].getCenter();
116 (*g)[2] = listGauss[i].getFWHM();
117 ++count;
118 }
119 }
120 estimate_.resize();
121 listGauss.evaluate(estimate_,x_);
122 return true;
123}
124
125std::vector<float> SDFitter::getEstimate() const
126{
127 if (estimate_.nelements() == 0)
128 throw (AipsError("No estimate set."));
129 std::vector<float> stlout;
130 estimate_.tovector(stlout);
131 return stlout;
132}
133
134
135bool SDFitter::setExpression(const std::string& expr, int ncomp)
136{
137 clear();
138 if (expr == "gauss") {
139 if (ncomp < 1) throw (AipsError("Need at least one gaussian to fit."));
140 funcs_.resize(ncomp);
141 for (Int k=0; k<ncomp; ++k) {
142 funcs_[k] = new Gaussian1D<Float>();
143 }
144 } else if (expr == "poly") {
145 funcs_.resize(1);
146 funcs_[0] = new Polynomial<Float>(ncomp);
147 } else {
148 cerr << " compiled functions not yet implemented" << endl;
149 //funcs_.resize(1);
150 //funcs_[0] = new CompiledFunction<Float>();
151 //funcs_[0]->setFunction(String(expr));
152 return false;
153 };
154 return true;
155}
156
157bool SDFitter::setData(std::vector<float> absc, std::vector<float> spec,
158 std::vector<bool> mask)
159{
160 x_.resize();
161 y_.resize();
162 m_.resize();
163 // convert std::vector to casa Vector
164 Vector<Float> tmpx(absc);
165 Vector<Float> tmpy(spec);
166 Vector<Bool> tmpm(mask);
167 AlwaysAssert(tmpx.nelements() == tmpy.nelements(), AipsError);
168 x_ = tmpx;
169 y_ = tmpy;
170 m_ = tmpm;
171 return true;
172}
173
174std::vector<float> SDFitter::getResidual() const
175{
176 if (residual_.nelements() == 0)
177 throw (AipsError("Function not yet fitted."));
178 std::vector<float> stlout;
179 residual_.tovector(stlout);
180 return stlout;
181}
182
183std::vector<float> SDFitter::getFit() const
184{
185 Vector<Float> out = thefit_;
186 std::vector<float> stlout;
187 out.tovector(stlout);
188 return stlout;
189
190}
191
192std::vector<float> SDFitter::getErrors() const
193{
194 Vector<Float> out = error_;
195 std::vector<float> stlout;
196 out.tovector(stlout);
197 return stlout;
198}
199
200bool SDFitter::setParameters(std::vector<float> params)
201{
202 Vector<Float> tmppar(params);
203 if (funcs_.nelements() == 0)
204 throw (AipsError("Function not yet set."));
205 if (parameters_.nelements() > 0 && tmppar.nelements() != parameters_.nelements())
206 throw (AipsError("Number of parameters inconsistent with function."));
207 if (parameters_.nelements() == 0)
208 parameters_.resize(tmppar.nelements());
209 fixedpar_.resize(tmppar.nelements());
210 fixedpar_ = False;
211 if (dynamic_cast<Gaussian1D<Float>* >(funcs_[0]) != 0) {
212 uInt count = 0;
213 for (uInt j=0; j < funcs_.nelements(); ++j) {
214 for (uInt i=0; i < funcs_[j]->nparameters(); ++i) {
215 (funcs_[j]->parameters())[i] = tmppar[count];
216 parameters_[count] = tmppar[count];
217 ++count;
218 }
219 }
220 } else if (dynamic_cast<Polynomial<Float>* >(funcs_[0]) != 0) {
221 for (uInt i=0; i < funcs_[0]->nparameters(); ++i) {
222 parameters_[i] = tmppar[i];
223 (funcs_[0]->parameters())[i] = tmppar[i];
224 }
225 }
226 return true;
227}
228
229bool SDFitter::setFixedParameters(std::vector<bool> fixed)
230{
231 Vector<Bool> tmp(fixed);
232 if (funcs_.nelements() == 0)
233 throw (AipsError("Function not yet set."));
234 if (fixedpar_.nelements() > 0 && tmp.nelements() != fixedpar_.nelements())
235 throw (AipsError("Number of mask elements inconsistent with function."));
236 if (dynamic_cast<Gaussian1D<Float>* >(funcs_[0]) != 0) {
237 uInt count = 0;
238 for (uInt j=0; j < funcs_.nelements(); ++j) {
239 for (uInt i=0; i < funcs_[j]->nparameters(); ++i) {
240 funcs_[j]->mask(i) = !tmp[count];
241 fixedpar_[count] = !tmp[count];
242 ++count;
243 }
244 }
245 } else if (dynamic_cast<Polynomial<Float>* >(funcs_[0]) != 0) {
246 for (uInt i=0; i < funcs_[0]->nparameters(); ++i) {
247 fixedpar_[i] = tmp[i];
248 funcs_[0]->mask(i) = tmp[i];
249 }
250 }
251 //fixedpar_ = !tmpmsk;
252 return true;
253}
254
255std::vector<float> SDFitter::getParameters() const {
256 Vector<Float> out = parameters_;
257 std::vector<float> stlout;
258 out.tovector(stlout);
259 return stlout;
260}
261
262std::vector<bool> SDFitter::getFixedParameters() const {
263 Vector<Bool> out(parameters_.nelements());
264 if (fixedpar_.nelements() == 0) {
265 out = False;
266 //throw (AipsError("No parameter mask set."));
267 } else {
268 out = fixedpar_;
269 }
270 std::vector<bool> stlout;
271 out.tovector(stlout);
272 return stlout;
273}
274
275float SDFitter::getChisquared() const {
276 return chisquared_;
277}
278
279bool SDFitter::fit() {
280 NonLinearFitLM<Float> fitter;
281 //CompiledFunction<AutoDiff<Float> > comp;
282 //Polynomial<AutoDiff<Float> > poly;
283 CompoundFunction<AutoDiff<Float> > func;
284 if (dynamic_cast<Gaussian1D<Float>* >(funcs_[0]) != 0) {
285 //computeEstimates();
286 for (uInt i=0; i<funcs_.nelements(); i++) {
287 Gaussian1D<AutoDiff<Float> > gauss;
288 for (uInt j=0; j<funcs_[i]->nparameters(); j++) {
289 gauss[j] = AutoDiff<Float>((*funcs_[i])[j], gauss.nparameters(), j);
290 gauss.mask(j) = funcs_[i]->mask(j);
291 }
292 func.addFunction(gauss);
293 }
294 } else if (dynamic_cast<Polynomial<Float>* >(funcs_[0]) != 0) {
295 Polynomial<AutoDiff<Float> > poly(funcs_[0]->nparameters()-1);
296 for (uInt j=0; j<funcs_[0]->nparameters(); j++) {
297 poly[j] = AutoDiff<Float>(0, poly.nparameters(), j);
298 poly.mask(j) = funcs_[0]->mask(j);
299 }
300 func.addFunction(poly);
301 } else if (dynamic_cast<CompiledFunction<Float>* >(funcs_[0]) != 0) {
302
303// CompiledFunction<AutoDiff<Float> > comp;
304// for (uInt j=0; j<funcs_[0]->nparameters(); j++) {
305// comp[j] = AutoDiff<Float>(0, comp.nparameters(), j);
306// comp.mask(j) = funcs_[0]->mask(j);
307// }
308// func.addFunction(comp);
309
310 cout << "NYI." << endl;
311 } else {
312 throw (AipsError("Fitter not set up correctly."));
313 }
314 fitter.setFunction(func);
315 fitter.setMaxIter(50+funcs_.nelements()*10);
316 // Convergence criterium
317 fitter.setCriteria(0.001);
318 // Fit
319 Vector<Float> sigma(x_.nelements());
320 sigma = 1.0;
321 //Vector<Float> sol;
322 parameters_.resize();
323 //Vector<Float> err;
324 error_.resize();
325 parameters_ = fitter.fit(x_, y_, sigma, &m_);
326 /*
327 CompoundFunction<Float> f;
328 for (uInt i=0; i<funcs_.nelements(); i++) {
329 f.addFunction(*funcs_[i]);
330 }
331 f.parameters().setParameters(parameters_);
332 */
333 error_ = fitter.errors();
334 chisquared_ = fitter.getChi2();
335 residual_.resize();
336 residual_ = y_;
337 fitter.residual(residual_,x_);
338 // use fitter.residual(model=True) to get the model
339 thefit_.resize(x_.nelements());
340 fitter.residual(thefit_,x_,True);
341 return true;
342}
Note: See TracBrowser for help on using the repository browser.