1 | //#---------------------------------------------------------------------------
|
---|
2 | //# Fitter.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"
|
---|
46 | using namespace asap;
|
---|
47 | using namespace casa;
|
---|
48 |
|
---|
49 | Fitter::Fitter()
|
---|
50 | {
|
---|
51 | }
|
---|
52 |
|
---|
53 | Fitter::~Fitter()
|
---|
54 | {
|
---|
55 | reset();
|
---|
56 | }
|
---|
57 |
|
---|
58 | void Fitter::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 | }
|
---|
70 |
|
---|
71 | void Fitter::reset()
|
---|
72 | {
|
---|
73 | clear();
|
---|
74 | x_.resize();
|
---|
75 | y_.resize();
|
---|
76 | m_.resize();
|
---|
77 | }
|
---|
78 |
|
---|
79 |
|
---|
80 | bool Fitter::computeEstimate() {
|
---|
81 | if (x_.nelements() == 0 || y_.nelements() == 0)
|
---|
82 | throw (AipsError("No x/y data specified."));
|
---|
83 |
|
---|
84 | if (dynamic_cast<Gaussian1D<Float>* >(funcs_[0]) == 0)
|
---|
85 | return false;
|
---|
86 | uInt n = funcs_.nelements();
|
---|
87 | SpectralEstimate estimator(n);
|
---|
88 | estimator.setQ(5);
|
---|
89 | Int mn,mx;
|
---|
90 | mn = 0;
|
---|
91 | mx = m_.nelements()-1;
|
---|
92 | for (uInt i=0; i<m_.nelements();++i) {
|
---|
93 | if (m_[i]) {
|
---|
94 | mn = i;
|
---|
95 | break;
|
---|
96 | }
|
---|
97 | }
|
---|
98 | for (uInt j=m_.nelements()-1; j>=0;--j) {
|
---|
99 | if (m_[j]) {
|
---|
100 | mx = j;
|
---|
101 | break;
|
---|
102 | }
|
---|
103 | }
|
---|
104 | mn = 0+x_.nelements()/10;
|
---|
105 | mx = x_.nelements()-x_.nelements()/10;
|
---|
106 | estimator.setRegion(mn,mx);
|
---|
107 | //estimator.setWindowing(True);
|
---|
108 | SpectralList listGauss = estimator.estimate(x_, y_);
|
---|
109 | parameters_.resize(n*3);
|
---|
110 | Gaussian1D<Float>* g = 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 | }
|
---|
118 | }
|
---|
119 | estimate_.resize();
|
---|
120 | listGauss.evaluate(estimate_,x_);
|
---|
121 | return true;
|
---|
122 | }
|
---|
123 |
|
---|
124 | std::vector<float> Fitter::getEstimate() const
|
---|
125 | {
|
---|
126 | if (estimate_.nelements() == 0)
|
---|
127 | throw (AipsError("No estimate set."));
|
---|
128 | std::vector<float> stlout;
|
---|
129 | estimate_.tovector(stlout);
|
---|
130 | return stlout;
|
---|
131 | }
|
---|
132 |
|
---|
133 |
|
---|
134 | bool Fitter::setExpression(const std::string& expr, int ncomp)
|
---|
135 | {
|
---|
136 | clear();
|
---|
137 | if (expr == "gauss") {
|
---|
138 | if (ncomp < 1) throw (AipsError("Need at least one gaussian to fit."));
|
---|
139 | funcs_.resize(ncomp);
|
---|
140 | for (Int k=0; k<ncomp; ++k) {
|
---|
141 | funcs_[k] = new Gaussian1D<Float>();
|
---|
142 | }
|
---|
143 | } else if (expr == "poly") {
|
---|
144 | funcs_.resize(1);
|
---|
145 | funcs_[0] = new Polynomial<Float>(ncomp);
|
---|
146 | } else {
|
---|
147 | cerr << " compiled functions not yet implemented" << endl;
|
---|
148 | //funcs_.resize(1);
|
---|
149 | //funcs_[0] = new CompiledFunction<Float>();
|
---|
150 | //funcs_[0]->setFunction(String(expr));
|
---|
151 | return false;
|
---|
152 | }
|
---|
153 | return true;
|
---|
154 | }
|
---|
155 |
|
---|
156 | bool Fitter::setData(std::vector<float> absc, std::vector<float> spec,
|
---|
157 | std::vector<bool> mask)
|
---|
158 | {
|
---|
159 | x_.resize();
|
---|
160 | y_.resize();
|
---|
161 | m_.resize();
|
---|
162 | // convert std::vector to casa Vector
|
---|
163 | Vector<Float> tmpx(absc);
|
---|
164 | Vector<Float> tmpy(spec);
|
---|
165 | Vector<Bool> tmpm(mask);
|
---|
166 | AlwaysAssert(tmpx.nelements() == tmpy.nelements(), AipsError);
|
---|
167 | x_ = tmpx;
|
---|
168 | y_ = tmpy;
|
---|
169 | m_ = tmpm;
|
---|
170 | return true;
|
---|
171 | }
|
---|
172 |
|
---|
173 | std::vector<float> Fitter::getResidual() const
|
---|
174 | {
|
---|
175 | if (residual_.nelements() == 0)
|
---|
176 | throw (AipsError("Function not yet fitted."));
|
---|
177 | std::vector<float> stlout;
|
---|
178 | residual_.tovector(stlout);
|
---|
179 | return stlout;
|
---|
180 | }
|
---|
181 |
|
---|
182 | std::vector<float> Fitter::getFit() const
|
---|
183 | {
|
---|
184 | Vector<Float> out = thefit_;
|
---|
185 | std::vector<float> stlout;
|
---|
186 | out.tovector(stlout);
|
---|
187 | return stlout;
|
---|
188 |
|
---|
189 | }
|
---|
190 |
|
---|
191 | std::vector<float> Fitter::getErrors() const
|
---|
192 | {
|
---|
193 | Vector<Float> out = error_;
|
---|
194 | std::vector<float> stlout;
|
---|
195 | out.tovector(stlout);
|
---|
196 | return stlout;
|
---|
197 | }
|
---|
198 |
|
---|
199 | bool Fitter::setParameters(std::vector<float> params)
|
---|
200 | {
|
---|
201 | Vector<Float> tmppar(params);
|
---|
202 | if (funcs_.nelements() == 0)
|
---|
203 | throw (AipsError("Function not yet set."));
|
---|
204 | if (parameters_.nelements() > 0 && tmppar.nelements() != parameters_.nelements())
|
---|
205 | throw (AipsError("Number of parameters inconsistent with function."));
|
---|
206 | if (parameters_.nelements() == 0)
|
---|
207 | parameters_.resize(tmppar.nelements());
|
---|
208 | fixedpar_.resize(tmppar.nelements());
|
---|
209 | fixedpar_ = False;
|
---|
210 | if (dynamic_cast<Gaussian1D<Float>* >(funcs_[0]) != 0) {
|
---|
211 | uInt count = 0;
|
---|
212 | for (uInt j=0; j < funcs_.nelements(); ++j) {
|
---|
213 | for (uInt i=0; i < funcs_[j]->nparameters(); ++i) {
|
---|
214 | (funcs_[j]->parameters())[i] = tmppar[count];
|
---|
215 | parameters_[count] = tmppar[count];
|
---|
216 | ++count;
|
---|
217 | }
|
---|
218 | }
|
---|
219 | } else if (dynamic_cast<Polynomial<Float>* >(funcs_[0]) != 0) {
|
---|
220 | for (uInt i=0; i < funcs_[0]->nparameters(); ++i) {
|
---|
221 | parameters_[i] = tmppar[i];
|
---|
222 | (funcs_[0]->parameters())[i] = tmppar[i];
|
---|
223 | }
|
---|
224 | }
|
---|
225 | return true;
|
---|
226 | }
|
---|
227 |
|
---|
228 | bool Fitter::setFixedParameters(std::vector<bool> fixed)
|
---|
229 | {
|
---|
230 | Vector<Bool> tmp(fixed);
|
---|
231 | if (funcs_.nelements() == 0)
|
---|
232 | throw (AipsError("Function not yet set."));
|
---|
233 | if (fixedpar_.nelements() > 0 && tmp.nelements() != fixedpar_.nelements())
|
---|
234 | throw (AipsError("Number of mask elements inconsistent with function."));
|
---|
235 | if (dynamic_cast<Gaussian1D<Float>* >(funcs_[0]) != 0) {
|
---|
236 | uInt count = 0;
|
---|
237 | for (uInt j=0; j < funcs_.nelements(); ++j) {
|
---|
238 | for (uInt i=0; i < funcs_[j]->nparameters(); ++i) {
|
---|
239 | funcs_[j]->mask(i) = !tmp[count];
|
---|
240 | fixedpar_[count] = !tmp[count];
|
---|
241 | ++count;
|
---|
242 | }
|
---|
243 | }
|
---|
244 | } else if (dynamic_cast<Polynomial<Float>* >(funcs_[0]) != 0) {
|
---|
245 | for (uInt i=0; i < funcs_[0]->nparameters(); ++i) {
|
---|
246 | fixedpar_[i] = tmp[i];
|
---|
247 | funcs_[0]->mask(i) = tmp[i];
|
---|
248 | }
|
---|
249 | }
|
---|
250 | //fixedpar_ = !tmpmsk;
|
---|
251 | return true;
|
---|
252 | }
|
---|
253 |
|
---|
254 | std::vector<float> Fitter::getParameters() const {
|
---|
255 | Vector<Float> out = parameters_;
|
---|
256 | std::vector<float> stlout;
|
---|
257 | out.tovector(stlout);
|
---|
258 | return stlout;
|
---|
259 | }
|
---|
260 |
|
---|
261 | std::vector<bool> Fitter::getFixedParameters() const {
|
---|
262 | Vector<Bool> out(parameters_.nelements());
|
---|
263 | if (fixedpar_.nelements() == 0) {
|
---|
264 | out = False;
|
---|
265 | //throw (AipsError("No parameter mask set."));
|
---|
266 | } else {
|
---|
267 | out = fixedpar_;
|
---|
268 | }
|
---|
269 | std::vector<bool> stlout;
|
---|
270 | out.tovector(stlout);
|
---|
271 | return stlout;
|
---|
272 | }
|
---|
273 |
|
---|
274 | float Fitter::getChisquared() const {
|
---|
275 | return chisquared_;
|
---|
276 | }
|
---|
277 |
|
---|
278 | bool Fitter::fit() {
|
---|
279 | NonLinearFitLM<Float> fitter;
|
---|
280 | CompoundFunction<Float> func;
|
---|
281 |
|
---|
282 | uInt n = funcs_.nelements();
|
---|
283 | for (uInt i=0; i<n; ++i) {
|
---|
284 | func.addFunction(*funcs_[i]);
|
---|
285 | }
|
---|
286 |
|
---|
287 | fitter.setFunction(func);
|
---|
288 | fitter.setMaxIter(50+n*10);
|
---|
289 | // Convergence criterium
|
---|
290 | fitter.setCriteria(0.001);
|
---|
291 |
|
---|
292 | // Fit
|
---|
293 | Vector<Float> sigma(x_.nelements());
|
---|
294 | sigma = 1.0;
|
---|
295 |
|
---|
296 | parameters_.resize();
|
---|
297 | parameters_ = fitter.fit(x_, y_, sigma, &m_);
|
---|
298 | std::vector<float> ps;
|
---|
299 | parameters_.tovector(ps);
|
---|
300 | setParameters(ps);
|
---|
301 |
|
---|
302 | error_.resize();
|
---|
303 | error_ = fitter.errors();
|
---|
304 |
|
---|
305 | chisquared_ = fitter.getChi2();
|
---|
306 |
|
---|
307 | residual_.resize();
|
---|
308 | residual_ = y_;
|
---|
309 | fitter.residual(residual_,x_);
|
---|
310 |
|
---|
311 | // use fitter.residual(model=True) to get the model
|
---|
312 | thefit_.resize(x_.nelements());
|
---|
313 | fitter.residual(thefit_,x_,True);
|
---|
314 | return true;
|
---|
315 | }
|
---|
316 |
|
---|
317 |
|
---|
318 | std::vector<float> Fitter::evaluate(int whichComp) const
|
---|
319 | {
|
---|
320 | std::vector<float> stlout;
|
---|
321 | uInt idx = uInt(whichComp);
|
---|
322 | Float y;
|
---|
323 | if ( idx < funcs_.nelements() ) {
|
---|
324 | for (uInt i=0; i<x_.nelements(); ++i) {
|
---|
325 | y = (*funcs_[idx])(x_[i]);
|
---|
326 | stlout.push_back(float(y));
|
---|
327 | }
|
---|
328 | }
|
---|
329 | return stlout;
|
---|
330 | }
|
---|
331 |
|
---|