1 |
/* jetfit.cpp - Package to fit multi-Gaussian distributions to histograms
|
2 |
* rewrite after accidental deletion 07-24-09
|
3 |
* Author: David Nisson
|
4 |
*/
|
5 |
|
6 |
#include <cstdlib>
|
7 |
#include <cmath>
|
8 |
#include <ctime>
|
9 |
#include <cstdio>
|
10 |
#include <cstring>
|
11 |
#include <cctype>
|
12 |
#include <iostream>
|
13 |
#include <fstream>
|
14 |
#include <sstream>
|
15 |
#include <string>
|
16 |
#include <vector>
|
17 |
#include <set>
|
18 |
#include <map>
|
19 |
#include <limits>
|
20 |
#include <exception>
|
21 |
|
22 |
#include <TMinuit.h>
|
23 |
#include <TH1.h>
|
24 |
#include <TH2.h>
|
25 |
#include <TF2.h>
|
26 |
#include <TRandom3.h>
|
27 |
#include <Rtypes.h>
|
28 |
#include <TFormula.h>
|
29 |
#include <TSystem.h>
|
30 |
|
31 |
#include "UserCode/JetFitAnalyzer/interface/jetfit.h"
|
32 |
|
33 |
#define PI 3.141592653
|
34 |
|
35 |
#define MAX_GAUSS 3
|
36 |
|
37 |
using namespace std;
|
38 |
|
39 |
namespace jetfit {
|
40 |
// configurable options
|
41 |
bool ignorezero = false; // ignore zero bins when fitting
|
42 |
|
43 |
int ngauss; // number of Gaussians in current model
|
44 |
|
45 |
struct histo {
|
46 |
vector< vector<double> > bins;
|
47 |
double Xlo, Xhi, Ylo, Yhi;
|
48 |
} energy;
|
49 |
|
50 |
|
51 |
model_def *mdef;
|
52 |
|
53 |
model_def& curr_model_def() {
|
54 |
return *mdef;
|
55 |
}
|
56 |
|
57 |
void set_model_def(model_def *_mdef) {
|
58 |
mdef = _mdef;
|
59 |
}
|
60 |
|
61 |
int get_ngauss() {
|
62 |
return ngauss;
|
63 |
}
|
64 |
|
65 |
void set_ngauss(int _ngauss) {
|
66 |
ngauss = _ngauss;
|
67 |
}
|
68 |
|
69 |
// fit function
|
70 |
double fit_fcn(double x, double y, double *xval) {
|
71 |
int npar_indiv = mdef->get_formula()->GetNpar();
|
72 |
double val = 0.0;
|
73 |
for (int i = 0; i < ngauss; i++) {
|
74 |
mdef->get_formula()->SetParameters(xval + i*npar_indiv);
|
75 |
val += mdef->get_formula()->Eval(x, y);
|
76 |
}
|
77 |
return val;
|
78 |
}
|
79 |
|
80 |
// TF2-compatible fit function
|
81 |
double fit_fcn_TF2(double *x, double *pval) {
|
82 |
double val = fit_fcn(x[0], x[1], pval);
|
83 |
return val;
|
84 |
}
|
85 |
|
86 |
// Integral of model formula / chisquare sigma
|
87 |
double formula_int(double xlo, double xhi, double ylo, double yhi,
|
88 |
double *pval, double XbinSize, double YbinSize,
|
89 |
double *xval) {
|
90 |
double xstep = (xhi - xlo) / static_cast<double>(20);
|
91 |
double ystep = (yhi - ylo) / static_cast<double>(20);
|
92 |
|
93 |
double fsum = 0.0;
|
94 |
double pval_old[256];
|
95 |
int npar = mdef->get_formula()->GetNpar();
|
96 |
if (npar > 256) {
|
97 |
cerr << "Parameter overload" << endl;
|
98 |
return 0.0;
|
99 |
}
|
100 |
mdef->get_formula()->GetParameters(pval_old);
|
101 |
mdef->get_formula()->SetParameters(pval);
|
102 |
|
103 |
for (int i = 0; i < 20; i++) {
|
104 |
for (int j = 0; j < 20; j++) {
|
105 |
double x = (static_cast<double>(i) + 0.5)*xstep + xlo;
|
106 |
double y = (static_cast<double>(j) + 0.5)*ystep + ylo;
|
107 |
double sig_cut = 1.0e-5;
|
108 |
double nu = XbinSize * YbinSize * fit_fcn(x, y, xval);
|
109 |
double sigma = mdef->chisquare_error(nu) < sig_cut
|
110 |
? sig_cut : mdef->chisquare_error(nu);
|
111 |
|
112 |
fsum += xstep * ystep * mdef->get_formula()->Eval(x, y)
|
113 |
/ sigma / sigma;
|
114 |
}
|
115 |
}
|
116 |
|
117 |
mdef->get_formula()->SetParameters(pval_old);
|
118 |
return fsum;
|
119 |
}
|
120 |
|
121 |
// MINUIT objective function
|
122 |
void fcn(int &npar, double *grad, double &fcnval, double *xval, int iflag)
|
123 |
{
|
124 |
double chisquare = 0.0;
|
125 |
double XbinSize = (energy.Xhi - energy.Xlo)
|
126 |
/ static_cast<double>(energy.bins.size());
|
127 |
double YbinSize = (energy.Yhi - energy.Ylo)
|
128 |
/ static_cast<double>(energy.bins.at(0).size());
|
129 |
try {
|
130 |
// add errors in data points in histogram
|
131 |
for (int i = 0; i < energy.bins.size(); i++) {
|
132 |
for (int j = 0; j < energy.bins.at(i).size(); j++) {
|
133 |
double x = (static_cast<double>(i) + 0.5) * (energy.Xhi - energy.Xlo)
|
134 |
/ static_cast<double>(energy.bins.size()) + energy.Xlo;
|
135 |
double y = (static_cast<double>(j) + 0.5) * (energy.Yhi - energy.Ylo)
|
136 |
/ static_cast<double>(energy.bins.at(i).size()) + energy.Ylo;
|
137 |
double nu = fit_fcn(x, y, xval) * XbinSize * YbinSize;
|
138 |
double sig_cut = 1.0e-5;
|
139 |
if (mdef->chisquare_error(nu) > sig_cut || !ignorezero)
|
140 |
chisquare += pow(energy.bins.at(i).at(j) - nu, 2.0)
|
141 |
/ pow(mdef->chisquare_error(nu), 2.0);
|
142 |
else
|
143 |
chisquare += pow(energy.bins.at(i).at(j) - nu, 2.0)
|
144 |
/ pow(sig_cut, 2.0);
|
145 |
}
|
146 |
}
|
147 |
|
148 |
// add errors due to Gaussians outside histogram
|
149 |
double eps = 0.1; // accuracy set for this function
|
150 |
for (int i = 0; i < ngauss; i++) {
|
151 |
double *pval = xval+i*mdef->get_formula()->GetNpar();
|
152 |
double par_x = pval[mdef->get_indiv_max_x()];
|
153 |
double par_y = pval[mdef->get_indiv_max_y()];
|
154 |
double par_sig = pval[3];
|
155 |
double cutoff_rad = -log(eps * 2.0 * PI * par_sig * par_sig
|
156 |
/ pval[0]);
|
157 |
|
158 |
bool left = par_x < energy.Xlo + cutoff_rad;
|
159 |
bool right = par_x > energy.Xhi - cutoff_rad;
|
160 |
bool top = par_y > energy.Yhi - cutoff_rad;
|
161 |
bool bottom = par_y > energy.Ylo + cutoff_rad;
|
162 |
|
163 |
if (left) {
|
164 |
double xlo = par_x - cutoff_rad;
|
165 |
double xhi = energy.Xlo;
|
166 |
double ylo = energy.Ylo;
|
167 |
double yhi = energy.Yhi;
|
168 |
chisquare += formula_int(xlo, xhi, ylo, yhi, pval,
|
169 |
XbinSize, YbinSize, xval);
|
170 |
}
|
171 |
if (right) {
|
172 |
double xlo = energy.Xhi;
|
173 |
double xhi = par_x + cutoff_rad;
|
174 |
double ylo = energy.Ylo;
|
175 |
double yhi = energy.Yhi;
|
176 |
chisquare += formula_int(xlo, xhi, ylo, yhi, pval,
|
177 |
XbinSize, YbinSize, xval);
|
178 |
}
|
179 |
if (top) {
|
180 |
double xlo = left ? par_x - cutoff_rad : energy.Xlo;
|
181 |
double xhi = right ? par_x + cutoff_rad : energy.Xhi;
|
182 |
double ylo = energy.Yhi;
|
183 |
double yhi = par_y + cutoff_rad;
|
184 |
chisquare += formula_int(xlo, xhi, ylo, yhi, pval,
|
185 |
XbinSize, YbinSize, xval);
|
186 |
}
|
187 |
if (bottom) {
|
188 |
double xlo = left ? par_x - cutoff_rad : energy.Xlo;
|
189 |
double xhi = right ? par_x + cutoff_rad : energy.Xhi;
|
190 |
double ylo = par_y - cutoff_rad;
|
191 |
double yhi = energy.Ylo;
|
192 |
chisquare += formula_int(xlo, xhi, ylo, yhi, pval,
|
193 |
XbinSize, YbinSize, xval);
|
194 |
}
|
195 |
}
|
196 |
|
197 |
fcnval = chisquare;
|
198 |
}
|
199 |
catch (exception ex) {
|
200 |
cerr << "Exception in jetfit::fcn" << endl;
|
201 |
}
|
202 |
}
|
203 |
|
204 |
bool get_ignorezero() {
|
205 |
return ignorezero;
|
206 |
}
|
207 |
|
208 |
void set_ignorezero(bool _iz) {
|
209 |
ignorezero = _iz;
|
210 |
}
|
211 |
|
212 |
TFormula * model_def::get_formula() {
|
213 |
return indiv_formula;
|
214 |
}
|
215 |
|
216 |
void model_def::set_formula(TFormula *new_formula) {
|
217 |
indiv_formula = new_formula;
|
218 |
}
|
219 |
|
220 |
void model_def::get_indiv_par(int i, string &name, double &start, double &step,
|
221 |
double &lo, double &hi) {
|
222 |
try {
|
223 |
name = indiv_par_names.at(i);
|
224 |
start = indiv_par_starts.at(i);
|
225 |
step = indiv_par_start_steps.at(i);
|
226 |
lo = indiv_par_lo.at(i);
|
227 |
hi = indiv_par_hi.at(i);
|
228 |
} catch (exception ex) {
|
229 |
cerr << "exception in jetfit::model_def::get_indiv_par" << endl;
|
230 |
}
|
231 |
}
|
232 |
|
233 |
void model_def::set_indiv_par(int i, string name, double start, double step,
|
234 |
double lo, double hi) {
|
235 |
if (indiv_par_names.size() < i+1) {
|
236 |
indiv_par_names.resize(i+1);
|
237 |
indiv_par_starts.resize(i+1);
|
238 |
indiv_par_start_steps.resize(i+1);
|
239 |
indiv_par_lo.resize(i+1);
|
240 |
indiv_par_hi.resize(i+1);
|
241 |
}
|
242 |
try {
|
243 |
indiv_par_names.at(i) = name;
|
244 |
indiv_par_starts.at(i) = start;
|
245 |
indiv_par_start_steps.at(i) = step;
|
246 |
indiv_par_lo.at(i) = lo;
|
247 |
indiv_par_hi.at(i) = hi;
|
248 |
}
|
249 |
catch (exception ex) {
|
250 |
cerr << "exception in jetfit::model_def::set_indiv_par" << endl;
|
251 |
}
|
252 |
}
|
253 |
|
254 |
int model_def::get_indiv_max_E() {
|
255 |
return indiv_max_E;
|
256 |
}
|
257 |
|
258 |
void model_def::set_indiv_max_E(int _new_max_E) {
|
259 |
indiv_max_E = _new_max_E;
|
260 |
}
|
261 |
|
262 |
int model_def::get_indiv_max_x() {
|
263 |
return indiv_max_x;
|
264 |
}
|
265 |
|
266 |
void model_def::set_indiv_max_x(int new_max_x) {
|
267 |
indiv_max_x = new_max_x;
|
268 |
}
|
269 |
|
270 |
int model_def::get_indiv_max_y() {
|
271 |
return indiv_max_y;
|
272 |
}
|
273 |
|
274 |
void model_def::set_indiv_max_y(int new_max_y) {
|
275 |
indiv_max_y = new_max_y;
|
276 |
}
|
277 |
|
278 |
unsigned model_def::get_n_special_par_sets() {
|
279 |
return special_par_starts.size();
|
280 |
}
|
281 |
|
282 |
void model_def::get_special_par(int g, int i, double &pstart,
|
283 |
double &perr, double &plo, double &phi) {
|
284 |
try {
|
285 |
pstart = special_par_starts.at(g).at(i);
|
286 |
perr = special_par_start_steps.at(g).at(i);
|
287 |
plo = special_par_lo.at(g).at(i);
|
288 |
phi = special_par_hi.at(g).at(i);
|
289 |
} catch (exception ex) {
|
290 |
cerr << "Exception in model_def::get_special_par" << endl;
|
291 |
}
|
292 |
}
|
293 |
|
294 |
void model_def::set_special_par(int g, int i, double pstart,
|
295 |
double perr, double plo, double phi) {
|
296 |
if (g+1 > special_par_starts.size()) {
|
297 |
special_par_starts.resize(g+1);
|
298 |
special_par_start_steps.resize(g+1);
|
299 |
special_par_lo.resize(g+1);
|
300 |
special_par_hi.resize(g+1);
|
301 |
}
|
302 |
|
303 |
try {
|
304 |
if (i+1 > special_par_starts.at(g).size()) {
|
305 |
special_par_starts.at(g).resize(i+1);
|
306 |
special_par_start_steps.at(g).resize(i+1);
|
307 |
special_par_lo.at(g).resize(i+1);
|
308 |
special_par_hi.at(g).resize(i+1);
|
309 |
}
|
310 |
|
311 |
special_par_starts.at(g).at(i) = pstart;
|
312 |
special_par_start_steps.at(g).at(i) = perr;
|
313 |
special_par_lo.at(g).at(i) = plo;
|
314 |
special_par_hi.at(g).at(i) = phi;
|
315 |
}
|
316 |
catch (exception ex) {
|
317 |
cerr << "exception in jetfit::model_def::set_special_par" << endl;
|
318 |
}
|
319 |
}
|
320 |
|
321 |
// translation of CERN's ERFC function
|
322 |
double erfc(double x) {
|
323 |
bool lef = false;
|
324 |
double p2[5], q2[5];
|
325 |
|
326 |
const double z1 = 1.0, hf = z1/2.0, c1 = 0.56418958;
|
327 |
const double vmax = 7.0;
|
328 |
const double switch_ = 4.0;
|
329 |
|
330 |
double p10 = 3.6767877, q10 = 3.2584593, p11 = -9.7970465e-2;
|
331 |
p2[0] = 7.3738883;
|
332 |
q2[0] = 7.3739609;
|
333 |
p2[1] = 6.8650185;
|
334 |
q2[1] = 1.5184908e1;
|
335 |
p2[2] = 3.0317993;
|
336 |
q2[2] = 1.2795530e1;
|
337 |
p2[3] = 5.6316962e-1;
|
338 |
q2[3] = 5.3542168;
|
339 |
p2[4] = 4.3187787e-5;
|
340 |
q2[4] = 1.0;
|
341 |
|
342 |
double p30 = -1.2436854e-1, q30 = 4.4091706e-1, p31 = -9.6821036e-2;
|
343 |
|
344 |
double v = fabs(x);
|
345 |
double y, h, hc, ap, aq;
|
346 |
if (v < hf) {
|
347 |
y = v*v;
|
348 |
h = x*(p10 + p11*y)/(q10 + y);
|
349 |
hc = 1.0 - h;
|
350 |
}
|
351 |
else {
|
352 |
if (v < switch_) {
|
353 |
ap = p2[4];
|
354 |
aq = q2[4];
|
355 |
for (int i = 3; i >= 0; i--) {
|
356 |
ap = p2[i] + v*ap;
|
357 |
aq = q2[i] + v*aq;
|
358 |
}
|
359 |
hc = exp(-v*v)*ap/aq;
|
360 |
h = 1.0 - hc;
|
361 |
}
|
362 |
else if (v < vmax) {
|
363 |
y = 1/v/v;
|
364 |
hc = exp(-v*v)*(c1 + y*(p30 + p31*y)/(q30 + y))/v;
|
365 |
h = 1.0 - hc;
|
366 |
}
|
367 |
// for very big values we can save us any calculation
|
368 |
// and the FP-exceptions from exp
|
369 |
else {
|
370 |
h = 1.0;
|
371 |
hc = 0.0;
|
372 |
}
|
373 |
if (x < 0) {
|
374 |
h = -h;
|
375 |
hc = 2.0 - hc;
|
376 |
}
|
377 |
}
|
378 |
if (lef) {
|
379 |
return h;
|
380 |
}
|
381 |
else {
|
382 |
return hc;
|
383 |
}
|
384 |
}
|
385 |
|
386 |
// translation of CERN's PROB function
|
387 |
double prob(double x, int n) {
|
388 |
const char *name = "PROB";
|
389 |
char errtxt[80];
|
390 |
|
391 |
const double r1 = 1.0,
|
392 |
hf = r1/2.0, th = r1/3.0, f1 = 2.0*r1/9.0;
|
393 |
const double c1 = 1.128379167095513;
|
394 |
const int nmax = 300;
|
395 |
// maximum chi2 per df for df >= 2., if chi2/df > chipdf prob=0
|
396 |
const double chipdf = 100.0;
|
397 |
const double xmax = 174.673, xmax2 = 2.0*xmax;
|
398 |
const double xlim = 24.0;
|
399 |
const double eps = 1e-30;
|
400 |
|
401 |
double y = x;
|
402 |
double u = hf*y;
|
403 |
double h, w, s, t, fi, e;
|
404 |
int m;
|
405 |
if (n < 0) {
|
406 |
h = 0.0;
|
407 |
sprintf(errtxt, "n = %d < 1", n);
|
408 |
cerr << "PROB: G100.1: "<<errtxt;
|
409 |
}
|
410 |
else if (y < 0.0) {
|
411 |
h = 0.0;
|
412 |
sprintf(errtxt, "x = %f < 0", n);
|
413 |
cerr << "PROB: G100.2: "<<errtxt;
|
414 |
}
|
415 |
else if (y == 0.0 || n/20 > y) {
|
416 |
h = 1.0;
|
417 |
}
|
418 |
else if (n == 1) {
|
419 |
w = sqrt(u);
|
420 |
if (w < xlim) {
|
421 |
h = erfc(w);
|
422 |
}
|
423 |
else {
|
424 |
h = 0.0;
|
425 |
}
|
426 |
}
|
427 |
else if (n > nmax) {
|
428 |
s = r1 / ((double)n);
|
429 |
t = f1 * s;
|
430 |
w = (pow(y*s, th) - (1.0 - t)) / sqrt(2.0*t);
|
431 |
if (w < -xlim) {
|
432 |
h = 1.0;
|
433 |
}
|
434 |
else if (w < xlim) {
|
435 |
h = hf * erfc(w);
|
436 |
}
|
437 |
else {
|
438 |
h = 0.0;
|
439 |
}
|
440 |
}
|
441 |
else {
|
442 |
m = n/2;
|
443 |
if (u < xmax2 && (y/n) <= chipdf) {
|
444 |
s = exp(-hf*u);
|
445 |
t = s;
|
446 |
e = s;
|
447 |
if (2*m == n) {
|
448 |
fi = 0.0;
|
449 |
for (int i = 1; i < m; i++) {
|
450 |
fi += 1.0;
|
451 |
t = u*t/fi;
|
452 |
s += t;
|
453 |
}
|
454 |
h = s*e;
|
455 |
}
|
456 |
else {
|
457 |
fi = 1.0;
|
458 |
for (int i = 1; i < m; i++) {
|
459 |
fi += 2.0;
|
460 |
t = t*y/fi;
|
461 |
s += t;
|
462 |
}
|
463 |
w = sqrt(u);
|
464 |
if (w < xlim) {
|
465 |
h = c1*w*s*e + erfc(w);
|
466 |
}
|
467 |
else {
|
468 |
h = 0.0;
|
469 |
}
|
470 |
}
|
471 |
}
|
472 |
else {
|
473 |
h = 0.0;
|
474 |
}
|
475 |
}
|
476 |
if (h > eps) {
|
477 |
return h;
|
478 |
}
|
479 |
else {
|
480 |
return 0.0;
|
481 |
}
|
482 |
}
|
483 |
|
484 |
void par_init(TH2 *hist, TMinuit *gMinuit, vector<TString> &pars,
|
485 |
double *pval, double *perr, double *plo, double *phi,
|
486 |
int npar, results r) {
|
487 |
int npar1 = npar - mdef->get_formula()->GetNpar();
|
488 |
bool init_spec_pars = false;
|
489 |
if (ngauss <= mdef->get_n_special_par_sets()) {
|
490 |
init_spec_pars = true;
|
491 |
for (int k = 0; k < ngauss; k++) {
|
492 |
int ipar0 = k*mdef->get_formula()->GetNpar(); // index of par 0
|
493 |
for (int l = 0; l < mdef->get_formula()->GetNpar(); l++) {
|
494 |
int ipar = ipar0 + l;
|
495 |
if (ipar < npar)
|
496 |
mdef->get_special_par(k, l, pval[ipar], perr[ipar], plo[ipar],
|
497 |
phi[ipar]);
|
498 |
else
|
499 |
cerr << "WARNING: Attempt to set parameter out of index range!"
|
500 |
<< endl;
|
501 |
}
|
502 |
}
|
503 |
}
|
504 |
else {
|
505 |
double XbinSize = (energy.Xhi - energy.Xlo)
|
506 |
/ static_cast<double>(energy.bins.size());
|
507 |
double YbinSize = (energy.Xhi - energy.Xlo)
|
508 |
/ static_cast<double>(energy.bins.at(0).size());
|
509 |
vector< vector<double> > sub_energy(energy.bins.size());
|
510 |
double max_sub, max_x = 0.0, max_y = 0.0;
|
511 |
double pval_other[256];
|
512 |
max_sub = -(numeric_limits<double>::infinity());
|
513 |
for (int i = 0; i < energy.bins.size(); i++) {
|
514 |
sub_energy[i].resize(energy.bins[i].size());
|
515 |
for (int j = 0; j < energy.bins[i].size(); j++) {
|
516 |
double x = static_cast<double>(i)*(energy.Xhi - energy.Xlo)
|
517 |
/ static_cast<double>(energy.bins.size()) + energy.Xlo;
|
518 |
double y = static_cast<double>(j)*(energy.Yhi - energy.Ylo)
|
519 |
/ static_cast<double>(energy.bins[i].size()) + energy.Ylo;
|
520 |
if (ngauss > 1) {
|
521 |
// subtract 4*sigma plus integral of fit function
|
522 |
try {
|
523 |
int npval_other = r.pval.at(r.pval.size()-1).size();
|
524 |
if (npval_other > 256) {
|
525 |
cerr << "Parameter overload" << endl;
|
526 |
return;
|
527 |
}
|
528 |
for (int k = 0; k < npval_other; k++) {
|
529 |
pval_other[k] = r.pval[r.pval.size()-1][k];
|
530 |
}
|
531 |
ngauss--;
|
532 |
double nu = fit_fcn(x, y, pval_other) * XbinSize * YbinSize;
|
533 |
ngauss++;
|
534 |
sub_energy[i][j] = energy.bins[i][j] - nu
|
535 |
- 4.0*mdef->chisquare_error(nu);
|
536 |
}
|
537 |
catch (exception ex) {
|
538 |
cerr << "Exception in par_init" << endl;
|
539 |
}
|
540 |
}
|
541 |
else {
|
542 |
sub_energy[i][j] = energy.bins[i][j];
|
543 |
}
|
544 |
if (sub_energy[i][j] > max_sub) {
|
545 |
max_sub = sub_energy[i][j];
|
546 |
max_x = x;
|
547 |
max_y = y;
|
548 |
}
|
549 |
}
|
550 |
}
|
551 |
|
552 |
for (int i = npar1; i < npar; i++) {
|
553 |
if (mdef->get_indiv_max_E() == i - npar1) {
|
554 |
double nu = 0.0;
|
555 |
if (ngauss > 1) {
|
556 |
nu = fit_fcn(max_x, max_y, pval_other) * XbinSize * YbinSize;
|
557 |
}
|
558 |
pval[i] = max_sub + (ngauss > 1 ? 4.0*mdef->chisquare_error(nu)
|
559 |
: 0.0);
|
560 |
perr[i] = mdef->chisquare_error(pval[i])*0.1;
|
561 |
plo[i] = 0.0;
|
562 |
phi[i] = 1.0e6;
|
563 |
}
|
564 |
else if (mdef->get_indiv_max_x() == i - npar1) {
|
565 |
pval[i] = max_x;
|
566 |
perr[i] = (energy.Xhi - energy.Xlo) / static_cast<double>(hist->GetXaxis()->GetNbins());
|
567 |
plo[i] = 0.0;
|
568 |
phi[i] = 0.0;
|
569 |
}
|
570 |
else if (mdef->get_indiv_max_y() == i - npar1) {
|
571 |
pval[i] = max_y;
|
572 |
perr[i] = (energy.Yhi - energy.Ylo) / static_cast<double>(hist->GetYaxis()->GetNbins());
|
573 |
plo[i] = 0.0;
|
574 |
phi[i] = 0.0;
|
575 |
}
|
576 |
else {
|
577 |
string dummy;
|
578 |
mdef->get_indiv_par(i-npar1, dummy, pval[i], perr[i], plo[i], phi[i]);
|
579 |
}
|
580 |
}
|
581 |
}
|
582 |
int n_pars_to_set = init_spec_pars ? ngauss*mdef->get_formula()->GetNpar()
|
583 |
: npar - npar1;
|
584 |
int init_par_to_set = init_spec_pars ? 0 : npar1;
|
585 |
for (int i = 0; i < n_pars_to_set; i++) {
|
586 |
double dummy;
|
587 |
string s;
|
588 |
int ipar = i + init_par_to_set;
|
589 |
mdef->get_indiv_par(i % mdef->get_formula()->GetNpar(), s,
|
590 |
dummy, dummy, dummy, dummy);
|
591 |
ostringstream par_oss;
|
592 |
par_oss << s << ipar / (mdef->get_formula()->GetNpar()) << flush;
|
593 |
try {
|
594 |
pars.at(ipar) = TString(par_oss.str());
|
595 |
}
|
596 |
catch (exception ex) {
|
597 |
cerr << "exception 2 in par_init" << endl;
|
598 |
}
|
599 |
int error_flag;
|
600 |
try {
|
601 |
gMinuit->mnparm(ipar, pars.at(ipar),
|
602 |
pval[ipar], perr[ipar], plo[ipar], phi[ipar], error_flag);
|
603 |
} catch (exception ex) {
|
604 |
cerr << "exception 3 in par_init" << endl;
|
605 |
}
|
606 |
}
|
607 |
}
|
608 |
|
609 |
results fit_histo(TH2 *hist, vector<trouble> &t_vec,
|
610 |
void (*cc_minuit)(TMinuit *, TH2 *, int),
|
611 |
int start_ngauss,
|
612 |
int rebinX, int rebinY,
|
613 |
double P_cutoff_val) {
|
614 |
TMinuit *gMinuit = new TMinuit();
|
615 |
int npar_indiv = mdef->get_formula()->GetNpar();
|
616 |
int istat, erflg;
|
617 |
results r;
|
618 |
|
619 |
gMinuit->SetFCN(fcn);
|
620 |
gMinuit->mninit(5, 6, 7);
|
621 |
|
622 |
// set error def and machine accuracy
|
623 |
double cs_err_def = 1.0;
|
624 |
double fcn_eps = 0.2;
|
625 |
gMinuit->mnexcm("SET ERR", &cs_err_def, 1, erflg);
|
626 |
gMinuit->mnexcm("SET EPS", &fcn_eps, 1, erflg);
|
627 |
|
628 |
// rebin histogram
|
629 |
TH2 *hist_rebinned;
|
630 |
if (rebinX > 1 && rebinY > 1)
|
631 |
hist_rebinned = hist->Rebin2D(rebinX, rebinY);
|
632 |
else
|
633 |
hist_rebinned = hist;
|
634 |
|
635 |
r.hist_rebinned = hist_rebinned;
|
636 |
// load histogram into energies vector
|
637 |
energy.bins.resize(hist_rebinned->GetXaxis()->GetNbins());
|
638 |
for (int i = 0; i < energy.bins.size(); i++) {
|
639 |
energy.bins[i].resize(hist_rebinned->GetYaxis()->GetNbins());
|
640 |
for (int j = 0; j < energy.bins[i].size(); j++) {
|
641 |
energy.bins[i][j] = hist_rebinned->GetBinContent(i+1, j+1);
|
642 |
}
|
643 |
}
|
644 |
|
645 |
energy.Xlo = hist->GetXaxis()->GetXmin();
|
646 |
energy.Xhi = hist->GetXaxis()->GetXmax();
|
647 |
energy.Ylo = hist->GetYaxis()->GetXmin();
|
648 |
energy.Yhi = hist->GetYaxis()->GetXmax();
|
649 |
|
650 |
if (start_ngauss < 1) {
|
651 |
start_ngauss = 1;
|
652 |
}
|
653 |
|
654 |
for (ngauss = start_ngauss;
|
655 |
ngauss <= (MAX_GAUSS > start_ngauss ? MAX_GAUSS : start_ngauss);
|
656 |
ngauss++) {
|
657 |
t_vec.resize(t_vec.size() + 1);
|
658 |
int npar = npar_indiv*ngauss;
|
659 |
double pval[256], perr[256], plo[256], phi[256];
|
660 |
if (npar > 256) {
|
661 |
cerr << "Parameter overload" << endl;
|
662 |
return r;
|
663 |
}
|
664 |
vector<TString> pars(npar);
|
665 |
|
666 |
// initialize parameters
|
667 |
par_init(hist, gMinuit, pars, pval, perr, plo, phi, npar, r);
|
668 |
|
669 |
// minimize
|
670 |
double chisquare, edm, errdef;
|
671 |
int nvpar, nparx;
|
672 |
trouble t;
|
673 |
t.occ = T_NULL;
|
674 |
t.istat = 3;
|
675 |
// fix the N values and sigmas of the Gaussians
|
676 |
for (int i = 0; i < ngauss; i++) {
|
677 |
double parno[2];
|
678 |
parno[0] = i*npar_indiv + 1;
|
679 |
parno[1] = i*npar_indiv + 4;
|
680 |
gMinuit->mnexcm("FIX", parno, 2, erflg);
|
681 |
}
|
682 |
gMinuit->mnexcm("SIMPLEX", 0, 0, erflg);
|
683 |
gMinuit->mnexcm("MIGRAD", 0, 0, erflg);
|
684 |
// release N values and sigmas, fix mu_x and mu_y
|
685 |
for (int i = 0; i < ngauss; i++) {
|
686 |
double parno[4];
|
687 |
parno[0] = i*npar_indiv + 1;
|
688 |
parno[1] = i*npar_indiv + 2;
|
689 |
parno[2] = i*npar_indiv + 3;
|
690 |
parno[3] = i*npar_indiv + 4;
|
691 |
gMinuit->mnexcm("RELEASE", parno, 1, erflg);
|
692 |
gMinuit->mnexcm("RELEASE", parno+3, 1, erflg);
|
693 |
gMinuit->mnexcm("FIX", parno+1, 2, erflg);
|
694 |
}
|
695 |
gMinuit->mnexcm("MIGRAD", 0, 0, erflg);
|
696 |
// release mu_x and mu_y
|
697 |
for (int i = 0; i < ngauss; i++) {
|
698 |
double parno[4];
|
699 |
parno[0] = i*npar_indiv + 1;
|
700 |
parno[1] = i*npar_indiv + 2;
|
701 |
parno[2] = i*npar_indiv + 3;
|
702 |
parno[3] = i*npar_indiv + 4;
|
703 |
gMinuit->mnexcm("RELEASE", parno+1, 2, erflg);
|
704 |
}
|
705 |
gMinuit->mnexcm("MIGRAD", 0, 0, erflg);
|
706 |
gMinuit->mnstat(chisquare, edm, errdef, nvpar, nparx, istat);
|
707 |
if (istat == 3) {
|
708 |
/* we're not concerned about MINOS errors right now
|
709 |
gMinuit->mnexcm("MINOS", 0, 0, erflg);
|
710 |
gMinuit->mnstat(chisquare, edm, errdef, nvpar, nparx, istat);
|
711 |
if (istat != 3) {
|
712 |
t.occ = T_MINOS;
|
713 |
t.istat = istat;
|
714 |
}
|
715 |
*/
|
716 |
}
|
717 |
else {
|
718 |
t.occ = T_MIGRAD;
|
719 |
t.istat = istat;
|
720 |
}
|
721 |
|
722 |
try {
|
723 |
if (t_vec.size() < ngauss) {
|
724 |
t_vec.resize(ngauss);
|
725 |
}
|
726 |
t_vec.at(ngauss-1) = t;
|
727 |
}
|
728 |
catch (exception ex) {
|
729 |
cerr << "exception in fit_histo" << endl;
|
730 |
}
|
731 |
|
732 |
// put parameters in map
|
733 |
for (int i = 0; i < npar && i < 256; i++) {
|
734 |
int iuint; // internal parameter number
|
735 |
gMinuit->mnpout(i, pars[i], pval[i], perr[i], plo[i], phi[i], iuint);
|
736 |
}
|
737 |
vector<double> pval_copy(npar);
|
738 |
vector<double> perr_copy(npar);
|
739 |
vector<double> plo_copy(npar);
|
740 |
vector<double> phi_copy(npar);
|
741 |
for (int i = 0; i < npar && i < 256; i++) {
|
742 |
pval_copy[i] = pval[i];
|
743 |
perr_copy[i] = perr[i];
|
744 |
plo_copy[i] = plo[i];
|
745 |
phi_copy[i] = phi[i];
|
746 |
}
|
747 |
r.pars.push_back(pars);
|
748 |
r.pval.push_back(pval_copy);
|
749 |
r.perr.push_back(perr_copy);
|
750 |
r.plo.push_back(plo_copy);
|
751 |
r.phi.push_back(phi_copy);
|
752 |
|
753 |
// execute user minuit code
|
754 |
if (cc_minuit != 0)
|
755 |
(*cc_minuit)(gMinuit, hist_rebinned, ngauss);
|
756 |
|
757 |
int ndof = 0;
|
758 |
if (!ignorezero)
|
759 |
ndof = energy.bins.size() * energy.bins[0].size();
|
760 |
else {
|
761 |
for (int i = 0; i < energy.bins.size(); i++) {
|
762 |
for (int j = 0; j < energy.bins[i].size(); j++) {
|
763 |
if (energy.bins[i][j] > 1.0e-30)
|
764 |
ndof++;
|
765 |
}
|
766 |
}
|
767 |
}
|
768 |
ndof -= npar;
|
769 |
|
770 |
double P = prob(chisquare, ndof);
|
771 |
cout << "P = "<<P << endl;
|
772 |
if (P > P_cutoff_val) {
|
773 |
break;
|
774 |
}
|
775 |
}
|
776 |
|
777 |
delete gMinuit;
|
778 |
return r;
|
779 |
}
|
780 |
}
|