JetFitAnalyzer/src/jetfit.cpp

/* jetfit.cpp - Package to fit multi-Gaussian distributions to histograms
 * Author: David Nisson
 */

#include <cstdlib>
#include <cmath>
#include <ctime>
#include <cstdio>
#include <cstring>
#include <cctype>
#include <iostream>
#include <fstream>
#include <sstream>
#include <string>
#include <vector>
#include <set>
#include <map>
#include <limits>
#include <exception>

#include <TMinuit.h>
#include <TH1.h>
#include <TH2.h>
#include <TF2.h>
#include <TRandom3.h>
#include <Rtypes.h>
#include <TFormula.h>
#include <TSystem.h>
#include <TMath.h>

#include "UserCode/JetFitAnalyzer/interface/jetfit.h"

#define PI 3.141592653

#define MAX_GAUSS 3

using namespace std;

namespace jetfit {
  // configurable options
  bool ignorezero = false; // ignore zero bins when fitting
  
  model_def *mdef;

  model_def& curr_model_def() {
    return *mdef;
  }

  void set_model_def(model_def *_mdef) {
    mdef = _mdef;
  }

  int model_def::get_ngauss() {
    return ngauss;
  }
  
  void model_def::set_ngauss(int _ngauss) {
    ngauss = _ngauss;
  }

  // fit function
  double model_def::fit_fcn(double x, double y, double *xval) {
    int npar_indiv = get_formula()->GetNpar();
    double val = 0.0;
    for (int i = 0; i < ngauss; i++) {
      get_formula()->SetParameters(xval + i*npar_indiv);
      val += mdef->get_formula()->Eval(x, y);
    }
    return val;
  }

  // TF2-compatible fit function
  double fit_fcn_TF2(double *x, double *pval) {
    double val = mdef->fit_fcn(x[0], x[1], pval);
    return val;
  }

  // Integral of (model formula)^2 / chisquare sigma
  double model_def::formula_int(double xlo, double xhi, double ylo, double yhi,
                     double *pval, double XbinSize, double YbinSize,
                     double *xval) {
    double xstep = (xhi - xlo) / static_cast<double>(20);
    double ystep = (yhi - ylo) / static_cast<double>(20);

    double fsum = 0.0;
    double pval_old[256];
    int npar = get_formula()->GetNpar();
    if (npar > 256) {
      cerr << "Parameter overload" << endl;
      return 0.0;
    }
    get_formula()->GetParameters(pval_old);
    get_formula()->SetParameters(pval);

    for (int i = 0; i < 20; i++) {
      for (int j = 0; j < 20; j++) {
        double x = (static_cast<double>(i) + 0.5)*xstep + xlo;
        double y = (static_cast<double>(j) + 0.5)*ystep + ylo;
        double sig_cut = 1.0e-5;
        double nu = XbinSize * YbinSize * fit_fcn(x, y, xval);
        double sigma = fabs(chisquare_error(0.0)) < sig_cut
          ? sig_cut : chisquare_error(0.0);

        fsum += pow(xstep * ystep * get_formula()->Eval(x, y), 2.0)
          / sigma / sigma;
      }
    }

    get_formula()->SetParameters(pval_old);
    return fsum;
  }

  // MINUIT objective function 
  void fcn(int &npar, double *grad, double &fcnval, double *xval, int iflag)
  {
    double chisquare = 0.0;
    double XbinSize = (energy.Xhi - energy.Xlo)
      / static_cast<double>(energy.bins.size());
    double YbinSize = (energy.Yhi - energy.Ylo)
      / static_cast<double>(energy.bins.at(0).size());
    try {
      // add errors in data points in histogram
      for (int i = 0; i < energy.bins.size(); i++) {
        for (int j = 0; j < energy.bins.at(i).size(); j++) {
          double x = (static_cast<double>(i) + 0.5) * (energy.Xhi - energy.Xlo)
            / static_cast<double>(energy.bins.size()) + energy.Xlo;
          double y = (static_cast<double>(j) + 0.5) * (energy.Yhi - energy.Ylo)
            / static_cast<double>(energy.bins.at(i).size()) + energy.Ylo;
          double nu = mdef->fit_fcn(x, y, xval) * XbinSize * YbinSize;
          double sig_cut = 1.0e-5;
          if (fabs(mdef->chisquare_error(energy.bins.at(i).at(j))) > sig_cut || !ignorezero) {
            chisquare += pow(energy.bins.at(i).at(j) - nu, 2.0) 
              / pow(mdef->chisquare_error(energy.bins.at(i).at(j)), 2.0);
          }
          else {
            chisquare += pow(energy.bins.at(i).at(j) - nu, 2.0) 
              / pow(sig_cut, 2.0);
          }
        }
      }
      
      // add errors due to Gaussians outside histogram
      double eps = 0.01; // accuracy set for this function
      for (int i = 0; i < mdef->get_ngauss(); i++) {
        double *pval = xval+i*(mdef->get_formula()->GetNpar());
        double par_x = pval[mdef->get_indiv_max_x()];
        double par_y = pval[mdef->get_indiv_max_y()];
        double par_sig = pval[3];
        double cutoff_rad = sqrt(-log(eps * 2.0 * PI * par_sig * par_sig
          / pval[0]) * 2.0 * par_sig * par_sig);

        bool left = par_x < energy.Xlo + cutoff_rad;
        bool right = par_x > energy.Xhi - cutoff_rad;
        bool top = par_y > energy.Yhi - cutoff_rad;
        bool bottom = par_y < energy.Ylo + cutoff_rad;

        if (left) {
          double xlo = par_x - cutoff_rad;
          double xhi = energy.Xlo;
          double ylo = energy.Ylo;
          double yhi = energy.Yhi;
          chisquare += 2.0*mdef->formula_int(xlo, xhi, ylo, yhi, pval,
                                   XbinSize, YbinSize, xval);
        }
        if (right) {
          double xlo = energy.Xhi;
          double xhi = par_x + cutoff_rad;
          double ylo = energy.Ylo;
          double yhi = energy.Yhi;
          chisquare += 2.0*mdef->formula_int(xlo, xhi, ylo, yhi, pval,
                                   XbinSize, YbinSize, xval);
        }
        if (top) {
          double xlo = left ? par_x - cutoff_rad : energy.Xlo;
          double xhi = right ? par_x + cutoff_rad : energy.Xhi;
          double ylo = energy.Yhi;
          double yhi = par_y + cutoff_rad;
          chisquare += 2.0*mdef->formula_int(xlo, xhi, ylo, yhi, pval,
                                   XbinSize, YbinSize, xval);
        }
        if (bottom) {
          double xlo = left ? par_x - cutoff_rad : energy.Xlo;
          double xhi = right ? par_x + cutoff_rad : energy.Xhi;
          double ylo = par_y - cutoff_rad;
          double yhi = energy.Ylo;
          chisquare += 2.0*mdef->formula_int(xlo, xhi, ylo, yhi, pval,
                                   XbinSize, YbinSize, xval);
        }
      }
      
      fcnval = chisquare;
    }
    catch (exception ex) {
      cerr << "Exception in jetfit::fcn" << endl;
    }
  }

  bool get_ignorezero() {
    return ignorezero;
  }

  void set_ignorezero(bool _iz) {
    ignorezero = _iz;
  }

  TFormula * model_def::get_formula() {
    return indiv_formula;
  }

  void model_def::set_formula(TFormula *new_formula) {
    indiv_formula = new_formula;
  }

  void model_def::get_indiv_par(int i, string &name, double &start, double &step,
                     double &lo, double &hi) {
    try {
      name = indiv_par_names.at(i);
      start = indiv_par_starts.at(i);
      step = indiv_par_start_steps.at(i);
      lo = indiv_par_lo.at(i);
      hi = indiv_par_hi.at(i);
    } catch (exception ex) {
      cerr << "exception in jetfit::model_def::get_indiv_par" << endl;
    }
  }

  void model_def::set_indiv_par(int i, string name, double start, double step,
                     double lo, double hi) {
    if (indiv_par_names.size() < i+1) {
      indiv_par_names.resize(i+1);
      indiv_par_starts.resize(i+1);
      indiv_par_start_steps.resize(i+1);
      indiv_par_lo.resize(i+1);
      indiv_par_hi.resize(i+1);
    }
    try {
      indiv_par_names.at(i) = name;
      indiv_par_starts.at(i) = start;
      indiv_par_start_steps.at(i) = step;
      indiv_par_lo.at(i) = lo;
      indiv_par_hi.at(i) = hi;
    }
    catch (exception ex) {
      cerr << "exception in jetfit::model_def::set_indiv_par" << endl;
    }
  }

  int model_def::get_indiv_max_E() {
    return indiv_max_E;
  }

  void model_def::set_indiv_max_E(int _new_max_E) {
    indiv_max_E = _new_max_E;
  }

  int model_def::get_indiv_max_x() {
    return indiv_max_x;
  }

  void model_def::set_indiv_max_x(int new_max_x) {
    indiv_max_x = new_max_x;
  }

  int model_def::get_indiv_max_y() {
    return indiv_max_y;
  }

  void model_def::set_indiv_max_y(int new_max_y) {
    indiv_max_y = new_max_y;
  }

  unsigned model_def::get_n_special_par_sets() {
    return special_par_starts.size();
  }

  void model_def::get_special_par(int g, int i, double &pstart,
                                  double &perr, double &plo, double &phi) {
    try {
      pstart = special_par_starts.at(g).at(i);
      perr = special_par_start_steps.at(g).at(i);
      plo = special_par_lo.at(g).at(i);
      phi = special_par_hi.at(g).at(i);
    } catch (exception ex) {
      cerr << "Exception in model_def::get_special_par" << endl;
    }
  }

  void model_def::set_special_par(int g, int i, double pstart,
                                  double perr, double plo, double phi) {
    if (g+1 > special_par_starts.size()) {
      special_par_starts.resize(g+1);
      special_par_start_steps.resize(g+1);
      special_par_lo.resize(g+1);
      special_par_hi.resize(g+1);
    }

    try {
      if (i+1 > special_par_starts.at(g).size()) {
        special_par_starts.at(g).resize(i+1);
        special_par_start_steps.at(g).resize(i+1);
        special_par_lo.at(g).resize(i+1);
        special_par_hi.at(g).resize(i+1);
      }

      special_par_starts.at(g).at(i) = pstart;
      special_par_start_steps.at(g).at(i) = perr;
      special_par_lo.at(g).at(i) = plo;
      special_par_hi.at(g).at(i) = phi;
    }
    catch (exception ex) {
      cerr << "exception in jetfit::model_def::set_special_par" << endl;
    }
  }

  void fill_histo_with_vec(TH2 *hist, vector< vector<double> > &vec) {
    for (int i = 0; i < vec.size(); i++) {
      for (int j = 0; j < vec[i].size(); j++) {
        hist->SetBinContent(i+1, j+1, vec[i][j]);
      }
    }
  }

  void model_def::par_init(TH2 *hist, TMinuit *gMinuit, vector<TString> &pars,
                double *pval, double *perr, double *plo, double *phi,
                int npar, FitResults r) {
    int npar1 = npar - get_formula()->GetNpar();
    bool init_spec_pars = false;
    if (ngauss <= get_n_special_par_sets()) {
      init_spec_pars = true;
      for (int k = 0; k < ngauss; k++) {
        int ipar0 = k*get_formula()->GetNpar(); // index of indiv par 0
        for (int l = 0; l < get_formula()->GetNpar(); l++) {
          int ipar = ipar0 + l;
          if (ipar < npar)
            get_special_par(k, l, pval[ipar], perr[ipar], plo[ipar],
                                 phi[ipar]);
          else
            cerr << "WARNING: Attempt to set parameter out of index range!"
                 << endl;
        }
      }
    }
    else {
      double XbinSize = (energy.Xhi - energy.Xlo)
        / static_cast<double>(energy.bins.size());
      double YbinSize = (energy.Xhi - energy.Xlo)
        / static_cast<double>(energy.bins.at(0).size());
      vector< vector<double> > sub_energy(energy.bins.size());
      double max_sub, max_x = 0.0, max_y = 0.0;
      double pval_other[256];
      max_sub = -(numeric_limits<double>::max());
      for (int i = 0; i < energy.bins.size(); i++) {
        sub_energy[i].resize(energy.bins[i].size());
        for (int j = 0; j < energy.bins[i].size(); j++) {
          double x = static_cast<double>(i)*(energy.Xhi - energy.Xlo)
            / static_cast<double>(energy.bins.size()) + energy.Xlo;
          double y = static_cast<double>(j)*(energy.Yhi - energy.Ylo)
            / static_cast<double>(energy.bins[i].size()) + energy.Ylo;
          if (ngauss > 1) {
            // subtract integral of fit function
            try {
              int npval_other = r.pval.at(r.pval.size()-1).size();
              if (npval_other > 256) {
                cerr << "Parameter overload" << endl;
                return;
              }
              for (int k = 0; k < npval_other; k++) {
                pval_other[k] = r.pval[r.pval.size()-1][k];
              }
              ngauss--;
              double nu = fit_fcn(x, y, pval_other) * XbinSize * YbinSize;
              ngauss++;
              sub_energy[i][j] = energy.bins[i][j] - nu;
            }
            catch (exception ex) {
              cerr << "Exception in par_init" << endl;
            }
          }
          else {
            sub_energy[i][j] = energy.bins[i][j];
          }
        }
      }
      TH2D *hist_sub = new TH2D("hist_sub", "Subtracted histo",
                                sub_energy.size(), energy.Xlo,
                                energy.Xhi,
                                sub_energy[0].size(), energy.Ylo,
                                energy.Yhi);
      fill_histo_with_vec(hist_sub, sub_energy);
      max_sub = 0.0;
      max_x = 0.0;
      max_y = 0.0;
      for (int i = 1; i <= hist_sub->GetXaxis()->GetNbins(); i++) {
        for (int j = 1; j <= hist_sub->GetYaxis()->GetNbins(); j++) {
          double nu = energy.bins[i-1][j-1] - sub_energy[i-1][j-1];
          if (hist_sub->GetBinContent(i, j) 
              - 3.0*pow(chisquare_error(nu), 2.0) > max_sub) {
            max_sub = hist_sub->GetBinContent(i, j);
            max_x = static_cast<double>(i-1)*XbinSize
              + hist_sub->GetXaxis()->GetXmin();
            max_y = static_cast<double>(j-1)*YbinSize
              + hist_sub->GetYaxis()->GetXmin();
          }
        }
      }  

      for (int i = npar1; i < npar; i++) {
        if (get_indiv_max_E() == i - npar1) {
          pval[i] = max_sub / XbinSize / YbinSize;
          perr[i] = mdef->chisquare_error(pval[i])*0.5;
          plo[i] = 0.0;
          phi[i] = 1.0e6;
        }
        else if (get_indiv_max_x() == i - npar1) {
          pval[i] = max_x;
          perr[i] = (energy.Xhi - energy.Xlo) / static_cast<double>(hist->GetXaxis()->GetNbins());
          plo[i] = 0.0;
          phi[i] = 0.0;
        }
        else if (get_indiv_max_y() == i - npar1) {
          pval[i] = max_y;
          perr[i] = (energy.Yhi - energy.Ylo) / static_cast<double>(hist->GetYaxis()->GetNbins());
          plo[i] = 0.0;
          phi[i] = 0.0;
        }
        else {
          string dummy;
          get_indiv_par(i-npar1, dummy, pval[i], perr[i], plo[i], phi[i]);
        }
      }
    }
    int n_pars_to_set = init_spec_pars ? ngauss*get_formula()->GetNpar()
      : npar - npar1;
    int init_par_to_set = init_spec_pars ? 0 : npar1;
    for (int i = 0; i < n_pars_to_set; i++) {
      double dummy;
      string s;
      int ipar = i + init_par_to_set;
      get_indiv_par(i % get_formula()->GetNpar(), s,
                         dummy, dummy, dummy, dummy);
      ostringstream par_oss;
      par_oss << s << ipar / (get_formula()->GetNpar()) << flush;
      try {
        pars.at(ipar) = TString(par_oss.str());
      }
      catch (exception ex) {
        cerr << "exception 2 in par_init" << endl;
      }
      int error_flag;
      try {
        gMinuit->mnparm(ipar, pars.at(ipar),
                        pval[ipar], perr[ipar], plo[ipar], phi[ipar], error_flag);
      } catch (exception ex) {
        cerr << "exception 3 in par_init" << endl;
      }
    }
  }

  FitResults fit_histo(TH2 *hist, vector<trouble> &t_vec,
                               void (*cc_minuit)(TMinuit *, TH2 *, int),
                               int start_ngauss,
                               int rebinX, int rebinY,
                               double P_cutoff_val) {
    TMinuit *gMinuit = new TMinuit();
    int npar_indiv = mdef->get_formula()->GetNpar();
    int istat, erflg;
    FitResults r;

    gMinuit->SetFCN(fcn);
    gMinuit->mninit(5, 6, 7);

    // set error def and machine accuracy
    double cs_err_def = 1.0;
    double fcn_eps = 0.2;
    gMinuit->mnexcm("SET ERR", &cs_err_def, 1, erflg);
    gMinuit->mnexcm("SET EPS", &fcn_eps, 1, erflg);

    // rebin histogram
    TH2 *hist_rebinned;
    if (rebinX > 1 && rebinY > 1)
      hist_rebinned = hist->Rebin2D(rebinX, rebinY);
    else
      hist_rebinned = hist;

    r.hist_rebinned = hist_rebinned;
    // load histogram into energies vector
    energy.bins.resize(hist_rebinned->GetXaxis()->GetNbins());
    for (int i = 0; i < energy.bins.size(); i++) {
      energy.bins[i].resize(hist_rebinned->GetYaxis()->GetNbins());
      for (int j = 0; j < energy.bins[i].size(); j++) {
        energy.bins[i][j] = hist_rebinned->GetBinContent(i+1, j+1);
      }
    }

    energy.Xlo = hist->GetXaxis()->GetXmin();
    energy.Xhi = hist->GetXaxis()->GetXmax();
    energy.Ylo = hist->GetYaxis()->GetXmin();
    energy.Yhi = hist->GetYaxis()->GetXmax();
    
    if (start_ngauss < 1) {
      start_ngauss = 1;
    }

    for (mdef->set_ngauss(start_ngauss);
         mdef->get_ngauss() <=
           (MAX_GAUSS > start_ngauss ? MAX_GAUSS : start_ngauss);
         mdef->set_ngauss(mdef->get_ngauss()+1)) {
     
      int ngauss = mdef->get_ngauss();
      t_vec.resize(t_vec.size() + 1);
      int npar = npar_indiv*mdef->get_ngauss();
      double pval[256], perr[256], plo[256], phi[256];
      if (npar > 256) {
        cerr << "Parameter overload" << endl;
        return r;
      }
      vector<TString> pars(npar);

      // initialize parameters
      mdef->par_init(hist, gMinuit, pars, pval, perr, plo, phi, npar, r);

      // minimize
      double chisquare, edm, errdef;
      int nvpar, nparx;
      trouble t;
      t.occ = T_NULL;
      t.istat = 3;
      // fix sigmas of the Gaussians
      for (int i = 0; i < ngauss; i++) {
        double parno[2];
        parno[0] = i*npar_indiv + 4;
        gMinuit->mnexcm("FIX", parno, 1, erflg);
      }
      gMinuit->mnexcm("SIMPLEX", 0, 0, erflg);
      gMinuit->mnexcm("MIGRAD", 0, 0, erflg);
      gMinuit->mnstat(chisquare, edm, errdef, nvpar, nparx, istat);
      if (istat == 3) {
        /* we're not concerned about MINOS errors right now
        gMinuit->mnexcm("MINOS", 0, 0, erflg);
        gMinuit->mnstat(chisquare, edm, errdef, nvpar, nparx, istat);
        if (istat != 3) {
          t.occ = T_MINOS;
          t.istat = istat;
        }
        */
      }
      else {
        t.occ = T_MIGRAD;
        t.istat = istat;
      }

      try {
        if (t_vec.size() < ngauss) {
          t_vec.resize(ngauss);
        }
        t_vec.at(ngauss-1) = t;
      }
      catch (exception ex) {
        cerr << "exception in fit_histo" << endl;
      }

      // put parameters in map
      for (int i = 0; i < npar && i < 256; i++) {
        int iuint; // internal parameter number
        gMinuit->mnpout(i, pars[i], pval[i], perr[i], plo[i], phi[i], iuint);
      }
      vector<double> pval_copy(npar);
      vector<double> perr_copy(npar);
      vector<double> plo_copy(npar);
      vector<double> phi_copy(npar);
      for (int i = 0; i < npar && i < 256; i++) {
        pval_copy[i] = pval[i];
        perr_copy[i] = perr[i];
        plo_copy[i] = plo[i];
        phi_copy[i] = phi[i];
      }
      r.pars.push_back(pars);
      r.pval.push_back(pval_copy);
      r.perr.push_back(perr_copy);
      r.plo.push_back(plo_copy);
      r.phi.push_back(phi_copy);

      // execute user minuit code
      if (cc_minuit != 0)
        (*cc_minuit)(gMinuit, hist_rebinned, ngauss);

      int ndof = 0;
      if (!ignorezero)
        ndof = energy.bins.size() * energy.bins[0].size();
      else {
        for (int i = 0; i < energy.bins.size(); i++) {
          for (int j = 0; j < energy.bins[i].size(); j++) {
            if (energy.bins[i][j] > 1.0e-30)
              ndof++;
          }
        }
      }
      ndof -= npar;

      r.chisquare.push_back(chisquare);
      double P = TMath::Prob(chisquare, ndof);
      if (P > P_cutoff_val) {
        break;
      }
    }

    delete gMinuit;
    return r;
  }
}
Revision:	1.14
Committed:	Fri Nov 13 04:01:17 2009 UTC (15 years, 5 months ago) by dnisson
Branch:	MAIN
CVS Tags:	HEAD
Changes since 1.13:	+0 -0 lines
State:	*FILE REMOVED*
Log Message:	Removed now-obsolete jetfit.cpp and jetfit.h files
#	User	Rev	Content
1	dnisson	1.1	/* jetfit.cpp - Package to fit multi-Gaussian distributions to histograms
2			* Author: David Nisson
3			*/
4
5			#include <cstdlib>
6			#include <cmath>
7			#include <ctime>
8			#include <cstdio>
9			#include <cstring>
10			#include <cctype>
11			#include <iostream>
12			#include <fstream>
13			#include <sstream>
14			#include <string>
15			#include <vector>
16			#include <set>
17			#include <map>
18			#include <limits>
19			#include <exception>
20
21			#include <TMinuit.h>
22			#include <TH1.h>
23			#include <TH2.h>
24			#include <TF2.h>
25			#include <TRandom3.h>
26			#include <Rtypes.h>
27			#include <TFormula.h>
28			#include <TSystem.h>
29	dnisson	1.12	#include <TMath.h>
30	dnisson	1.1
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			model_def *mdef;
44
45			model_def& curr_model_def() {
46			return *mdef;
47			}
48
49			void set_model_def(model_def *_mdef) {
50			mdef = _mdef;
51			}
52
53	dnisson	1.10	int model_def::get_ngauss() {
54	dnisson	1.1	return ngauss;
55			}
56
57	dnisson	1.10	void model_def::set_ngauss(int _ngauss) {
58	dnisson	1.1	ngauss = _ngauss;
59			}
60
61			// fit function
62	dnisson	1.10	double model_def::fit_fcn(double x, double y, double *xval) {
63			int npar_indiv = get_formula()->GetNpar();
64	dnisson	1.1	double val = 0.0;
65			for (int i = 0; i < ngauss; i++) {
66	dnisson	1.10	get_formula()->SetParameters(xval + i*npar_indiv);
67	dnisson	1.1	val += mdef->get_formula()->Eval(x, y);
68			}
69			return val;
70			}
71
72			// TF2-compatible fit function
73			double fit_fcn_TF2(double x, double pval) {
74	dnisson	1.10	double val = mdef->fit_fcn(x[0], x[1], pval);
75	dnisson	1.1	return val;
76			}
77
78	dnisson	1.9	// Integral of (model formula)^2 / chisquare sigma
79	dnisson	1.10	double model_def::formula_int(double xlo, double xhi, double ylo, double yhi,
80	dnisson	1.1	double *pval, double XbinSize, double YbinSize,
81			double *xval) {
82			double xstep = (xhi - xlo) / static_cast<double>(20);
83			double ystep = (yhi - ylo) / static_cast<double>(20);
84
85			double fsum = 0.0;
86			double pval_old[256];
87	dnisson	1.10	int npar = get_formula()->GetNpar();
88	dnisson	1.1	if (npar > 256) {
89			cerr << "Parameter overload" << endl;
90			return 0.0;
91			}
92	dnisson	1.10	get_formula()->GetParameters(pval_old);
93			get_formula()->SetParameters(pval);
94	dnisson	1.1
95			for (int i = 0; i < 20; i++) {
96			for (int j = 0; j < 20; j++) {
97			double x = (static_cast<double>(i) + 0.5)*xstep + xlo;
98			double y = (static_cast<double>(j) + 0.5)*ystep + ylo;
99			double sig_cut = 1.0e-5;
100			double nu = XbinSize * YbinSize * fit_fcn(x, y, xval);
101	dnisson	1.10	double sigma = fabs(chisquare_error(0.0)) < sig_cut
102			? sig_cut : chisquare_error(0.0);
103	dnisson	1.1
104	dnisson	1.10	fsum += pow(xstep * ystep * get_formula()->Eval(x, y), 2.0)
105	dnisson	1.1	/ sigma / sigma;
106			}
107			}
108
109	dnisson	1.10	get_formula()->SetParameters(pval_old);
110	dnisson	1.1	return fsum;
111			}
112
113			// MINUIT objective function
114			void fcn(int &npar, double grad, double &fcnval, double xval, int iflag)
115			{
116			double chisquare = 0.0;
117			double XbinSize = (energy.Xhi - energy.Xlo)
118			/ static_cast<double>(energy.bins.size());
119			double YbinSize = (energy.Yhi - energy.Ylo)
120			/ static_cast<double>(energy.bins.at(0).size());
121			try {
122			// add errors in data points in histogram
123			for (int i = 0; i < energy.bins.size(); i++) {
124			for (int j = 0; j < energy.bins.at(i).size(); j++) {
125	dnisson	1.4	double x = (static_cast<double>(i) + 0.5) * (energy.Xhi - energy.Xlo)
126	dnisson	1.1	/ static_cast<double>(energy.bins.size()) + energy.Xlo;
127			double y = (static_cast<double>(j) + 0.5) * (energy.Yhi - energy.Ylo)
128			/ static_cast<double>(energy.bins.at(i).size()) + energy.Ylo;
129	dnisson	1.10	double nu = mdef->fit_fcn(x, y, xval) * XbinSize * YbinSize;
130	dnisson	1.1	double sig_cut = 1.0e-5;
131	dnisson	1.7	if (fabs(mdef->chisquare_error(energy.bins.at(i).at(j))) > sig_cut \|\| !ignorezero) {
132	dnisson	1.1	chisquare += pow(energy.bins.at(i).at(j) - nu, 2.0)
133	dnisson	1.7	/ pow(mdef->chisquare_error(energy.bins.at(i).at(j)), 2.0);
134	dnisson	1.3	}
135			else {
136	dnisson	1.1	chisquare += pow(energy.bins.at(i).at(j) - nu, 2.0)
137			/ pow(sig_cut, 2.0);
138	dnisson	1.3	}
139	dnisson	1.1	}
140			}
141	dnisson	1.7
142	dnisson	1.1	// add errors due to Gaussians outside histogram
143	dnisson	1.6	double eps = 0.01; // accuracy set for this function
144	dnisson	1.10	for (int i = 0; i < mdef->get_ngauss(); i++) {
145	dnisson	1.3	double pval = xval+i(mdef->get_formula()->GetNpar());
146	dnisson	1.1	double par_x = pval[mdef->get_indiv_max_x()];
147			double par_y = pval[mdef->get_indiv_max_y()];
148			double par_sig = pval[3];
149	dnisson	1.7	double cutoff_rad = sqrt(-log(eps * 2.0 * PI * par_sig * par_sig
150			/ pval[0]) * 2.0 * par_sig * par_sig);
151	dnisson	1.1
152			bool left = par_x < energy.Xlo + cutoff_rad;
153			bool right = par_x > energy.Xhi - cutoff_rad;
154			bool top = par_y > energy.Yhi - cutoff_rad;
155	dnisson	1.2	bool bottom = par_y < energy.Ylo + cutoff_rad;
156	dnisson	1.1
157			if (left) {
158			double xlo = par_x - cutoff_rad;
159			double xhi = energy.Xlo;
160			double ylo = energy.Ylo;
161			double yhi = energy.Yhi;
162	dnisson	1.10	chisquare += 2.0*mdef->formula_int(xlo, xhi, ylo, yhi, pval,
163	dnisson	1.1	XbinSize, YbinSize, xval);
164			}
165			if (right) {
166			double xlo = energy.Xhi;
167			double xhi = par_x + cutoff_rad;
168			double ylo = energy.Ylo;
169			double yhi = energy.Yhi;
170	dnisson	1.10	chisquare += 2.0*mdef->formula_int(xlo, xhi, ylo, yhi, pval,
171	dnisson	1.1	XbinSize, YbinSize, xval);
172			}
173			if (top) {
174			double xlo = left ? par_x - cutoff_rad : energy.Xlo;
175			double xhi = right ? par_x + cutoff_rad : energy.Xhi;
176			double ylo = energy.Yhi;
177			double yhi = par_y + cutoff_rad;
178	dnisson	1.10	chisquare += 2.0*mdef->formula_int(xlo, xhi, ylo, yhi, pval,
179	dnisson	1.1	XbinSize, YbinSize, xval);
180			}
181			if (bottom) {
182			double xlo = left ? par_x - cutoff_rad : energy.Xlo;
183			double xhi = right ? par_x + cutoff_rad : energy.Xhi;
184			double ylo = par_y - cutoff_rad;
185			double yhi = energy.Ylo;
186	dnisson	1.10	chisquare += 2.0*mdef->formula_int(xlo, xhi, ylo, yhi, pval,
187	dnisson	1.1	XbinSize, YbinSize, xval);
188			}
189			}
190	dnisson	1.7
191	dnisson	1.1	fcnval = chisquare;
192			}
193			catch (exception ex) {
194			cerr << "Exception in jetfit::fcn" << endl;
195			}
196			}
197
198			bool get_ignorezero() {
199			return ignorezero;
200			}
201
202			void set_ignorezero(bool _iz) {
203			ignorezero = _iz;
204			}
205
206			TFormula * model_def::get_formula() {
207			return indiv_formula;
208			}
209
210			void model_def::set_formula(TFormula *new_formula) {
211			indiv_formula = new_formula;
212			}
213
214			void model_def::get_indiv_par(int i, string &name, double &start, double &step,
215			double &lo, double &hi) {
216			try {
217			name = indiv_par_names.at(i);
218			start = indiv_par_starts.at(i);
219			step = indiv_par_start_steps.at(i);
220			lo = indiv_par_lo.at(i);
221			hi = indiv_par_hi.at(i);
222			} catch (exception ex) {
223			cerr << "exception in jetfit::model_def::get_indiv_par" << endl;
224			}
225			}
226
227			void model_def::set_indiv_par(int i, string name, double start, double step,
228			double lo, double hi) {
229			if (indiv_par_names.size() < i+1) {
230			indiv_par_names.resize(i+1);
231			indiv_par_starts.resize(i+1);
232			indiv_par_start_steps.resize(i+1);
233			indiv_par_lo.resize(i+1);
234			indiv_par_hi.resize(i+1);
235			}
236			try {
237			indiv_par_names.at(i) = name;
238			indiv_par_starts.at(i) = start;
239			indiv_par_start_steps.at(i) = step;
240			indiv_par_lo.at(i) = lo;
241			indiv_par_hi.at(i) = hi;
242			}
243			catch (exception ex) {
244			cerr << "exception in jetfit::model_def::set_indiv_par" << endl;
245			}
246			}
247
248			int model_def::get_indiv_max_E() {
249			return indiv_max_E;
250			}
251
252			void model_def::set_indiv_max_E(int _new_max_E) {
253			indiv_max_E = _new_max_E;
254			}
255
256			int model_def::get_indiv_max_x() {
257			return indiv_max_x;
258			}
259
260			void model_def::set_indiv_max_x(int new_max_x) {
261			indiv_max_x = new_max_x;
262			}
263
264			int model_def::get_indiv_max_y() {
265			return indiv_max_y;
266			}
267
268			void model_def::set_indiv_max_y(int new_max_y) {
269			indiv_max_y = new_max_y;
270			}
271
272			unsigned model_def::get_n_special_par_sets() {
273			return special_par_starts.size();
274			}
275
276			void model_def::get_special_par(int g, int i, double &pstart,
277			double &perr, double &plo, double &phi) {
278			try {
279			pstart = special_par_starts.at(g).at(i);
280			perr = special_par_start_steps.at(g).at(i);
281			plo = special_par_lo.at(g).at(i);
282			phi = special_par_hi.at(g).at(i);
283			} catch (exception ex) {
284			cerr << "Exception in model_def::get_special_par" << endl;
285			}
286			}
287
288			void model_def::set_special_par(int g, int i, double pstart,
289			double perr, double plo, double phi) {
290			if (g+1 > special_par_starts.size()) {
291			special_par_starts.resize(g+1);
292			special_par_start_steps.resize(g+1);
293			special_par_lo.resize(g+1);
294			special_par_hi.resize(g+1);
295			}
296
297			try {
298			if (i+1 > special_par_starts.at(g).size()) {
299			special_par_starts.at(g).resize(i+1);
300			special_par_start_steps.at(g).resize(i+1);
301			special_par_lo.at(g).resize(i+1);
302			special_par_hi.at(g).resize(i+1);
303			}
304
305			special_par_starts.at(g).at(i) = pstart;
306			special_par_start_steps.at(g).at(i) = perr;
307			special_par_lo.at(g).at(i) = plo;
308			special_par_hi.at(g).at(i) = phi;
309			}
310			catch (exception ex) {
311			cerr << "exception in jetfit::model_def::set_special_par" << endl;
312			}
313			}
314
315	dnisson	1.5	void fill_histo_with_vec(TH2 *hist, vector< vector<double> > &vec) {
316			for (int i = 0; i < vec.size(); i++) {
317			for (int j = 0; j < vec[i].size(); j++) {
318	dnisson	1.6	hist->SetBinContent(i+1, j+1, vec[i][j]);
319	dnisson	1.5	}
320			}
321			}
322
323	dnisson	1.10	void model_def::par_init(TH2 hist, TMinuit gMinuit, vector<TString> &pars,
324	dnisson	1.1	double pval, double perr, double plo, double phi,
325	dnisson	1.13	int npar, FitResults r) {
326	dnisson	1.10	int npar1 = npar - get_formula()->GetNpar();
327	dnisson	1.1	bool init_spec_pars = false;
328	dnisson	1.10	if (ngauss <= get_n_special_par_sets()) {
329	dnisson	1.1	init_spec_pars = true;
330			for (int k = 0; k < ngauss; k++) {
331	dnisson	1.10	int ipar0 = k*get_formula()->GetNpar(); // index of indiv par 0
332			for (int l = 0; l < get_formula()->GetNpar(); l++) {
333	dnisson	1.1	int ipar = ipar0 + l;
334			if (ipar < npar)
335	dnisson	1.10	get_special_par(k, l, pval[ipar], perr[ipar], plo[ipar],
336	dnisson	1.1	phi[ipar]);
337			else
338			cerr << "WARNING: Attempt to set parameter out of index range!"
339			<< endl;
340			}
341			}
342			}
343			else {
344			double XbinSize = (energy.Xhi - energy.Xlo)
345			/ static_cast<double>(energy.bins.size());
346			double YbinSize = (energy.Xhi - energy.Xlo)
347			/ static_cast<double>(energy.bins.at(0).size());
348			vector< vector<double> > sub_energy(energy.bins.size());
349			double max_sub, max_x = 0.0, max_y = 0.0;
350			double pval_other[256];
351	dnisson	1.3	max_sub = -(numeric_limits<double>::max());
352	dnisson	1.1	for (int i = 0; i < energy.bins.size(); i++) {
353			sub_energy[i].resize(energy.bins[i].size());
354			for (int j = 0; j < energy.bins[i].size(); j++) {
355			double x = static_cast<double>(i)*(energy.Xhi - energy.Xlo)
356			/ static_cast<double>(energy.bins.size()) + energy.Xlo;
357			double y = static_cast<double>(j)*(energy.Yhi - energy.Ylo)
358			/ static_cast<double>(energy.bins[i].size()) + energy.Ylo;
359			if (ngauss > 1) {
360	dnisson	1.6	// subtract integral of fit function
361	dnisson	1.1	try {
362			int npval_other = r.pval.at(r.pval.size()-1).size();
363			if (npval_other > 256) {
364			cerr << "Parameter overload" << endl;
365			return;
366			}
367			for (int k = 0; k < npval_other; k++) {
368			pval_other[k] = r.pval[r.pval.size()-1][k];
369			}
370			ngauss--;
371			double nu = fit_fcn(x, y, pval_other) * XbinSize * YbinSize;
372			ngauss++;
373	dnisson	1.6	sub_energy[i][j] = energy.bins[i][j] - nu;
374	dnisson	1.1	}
375			catch (exception ex) {
376			cerr << "Exception in par_init" << endl;
377			}
378			}
379			else {
380			sub_energy[i][j] = energy.bins[i][j];
381			}
382			}
383			}
384	dnisson	1.5	TH2D *hist_sub = new TH2D("hist_sub", "Subtracted histo",
385			sub_energy.size(), energy.Xlo,
386			energy.Xhi,
387			sub_energy[0].size(), energy.Ylo,
388			energy.Yhi);
389			fill_histo_with_vec(hist_sub, sub_energy);
390	dnisson	1.8	max_sub = 0.0;
391	dnisson	1.9	max_x = 0.0;
392			max_y = 0.0;
393	dnisson	1.8	for (int i = 1; i <= hist_sub->GetXaxis()->GetNbins(); i++) {
394			for (int j = 1; j <= hist_sub->GetYaxis()->GetNbins(); j++) {
395	dnisson	1.9	double nu = energy.bins[i-1][j-1] - sub_energy[i-1][j-1];
396			if (hist_sub->GetBinContent(i, j)
397	dnisson	1.10	- 3.0*pow(chisquare_error(nu), 2.0) > max_sub) {
398	dnisson	1.8	max_sub = hist_sub->GetBinContent(i, j);
399	dnisson	1.9	max_x = static_cast<double>(i-1)*XbinSize
400			+ hist_sub->GetXaxis()->GetXmin();
401			max_y = static_cast<double>(j-1)*YbinSize
402			+ hist_sub->GetYaxis()->GetXmin();
403	dnisson	1.8	}
404			}
405	dnisson	1.9	}
406	dnisson	1.1
407			for (int i = npar1; i < npar; i++) {
408	dnisson	1.10	if (get_indiv_max_E() == i - npar1) {
409	dnisson	1.9	pval[i] = max_sub / XbinSize / YbinSize;
410	dnisson	1.4	perr[i] = mdef->chisquare_error(pval[i])*0.5;
411	dnisson	1.1	plo[i] = 0.0;
412			phi[i] = 1.0e6;
413			}
414	dnisson	1.10	else if (get_indiv_max_x() == i - npar1) {
415	dnisson	1.1	pval[i] = max_x;
416			perr[i] = (energy.Xhi - energy.Xlo) / static_cast<double>(hist->GetXaxis()->GetNbins());
417			plo[i] = 0.0;
418			phi[i] = 0.0;
419			}
420	dnisson	1.10	else if (get_indiv_max_y() == i - npar1) {
421	dnisson	1.1	pval[i] = max_y;
422			perr[i] = (energy.Yhi - energy.Ylo) / static_cast<double>(hist->GetYaxis()->GetNbins());
423			plo[i] = 0.0;
424			phi[i] = 0.0;
425			}
426			else {
427			string dummy;
428	dnisson	1.10	get_indiv_par(i-npar1, dummy, pval[i], perr[i], plo[i], phi[i]);
429	dnisson	1.1	}
430			}
431			}
432	dnisson	1.10	int n_pars_to_set = init_spec_pars ? ngauss*get_formula()->GetNpar()
433	dnisson	1.1	: npar - npar1;
434			int init_par_to_set = init_spec_pars ? 0 : npar1;
435			for (int i = 0; i < n_pars_to_set; i++) {
436			double dummy;
437			string s;
438			int ipar = i + init_par_to_set;
439	dnisson	1.10	get_indiv_par(i % get_formula()->GetNpar(), s,
440	dnisson	1.1	dummy, dummy, dummy, dummy);
441			ostringstream par_oss;
442	dnisson	1.10	par_oss << s << ipar / (get_formula()->GetNpar()) << flush;
443	dnisson	1.1	try {
444			pars.at(ipar) = TString(par_oss.str());
445			}
446			catch (exception ex) {
447			cerr << "exception 2 in par_init" << endl;
448			}
449			int error_flag;
450			try {
451			gMinuit->mnparm(ipar, pars.at(ipar),
452			pval[ipar], perr[ipar], plo[ipar], phi[ipar], error_flag);
453			} catch (exception ex) {
454			cerr << "exception 3 in par_init" << endl;
455			}
456			}
457			}
458
459	dnisson	1.13	FitResults fit_histo(TH2 *hist, vector<trouble> &t_vec,
460	dnisson	1.10	void (cc_minuit)(TMinuit , TH2 *, int),
461			int start_ngauss,
462			int rebinX, int rebinY,
463			double P_cutoff_val) {
464	dnisson	1.1	TMinuit *gMinuit = new TMinuit();
465			int npar_indiv = mdef->get_formula()->GetNpar();
466			int istat, erflg;
467	dnisson	1.13	FitResults r;
468	dnisson	1.1
469			gMinuit->SetFCN(fcn);
470			gMinuit->mninit(5, 6, 7);
471
472			// set error def and machine accuracy
473			double cs_err_def = 1.0;
474			double fcn_eps = 0.2;
475			gMinuit->mnexcm("SET ERR", &cs_err_def, 1, erflg);
476			gMinuit->mnexcm("SET EPS", &fcn_eps, 1, erflg);
477
478			// rebin histogram
479			TH2 *hist_rebinned;
480			if (rebinX > 1 && rebinY > 1)
481			hist_rebinned = hist->Rebin2D(rebinX, rebinY);
482			else
483			hist_rebinned = hist;
484
485			r.hist_rebinned = hist_rebinned;
486			// load histogram into energies vector
487			energy.bins.resize(hist_rebinned->GetXaxis()->GetNbins());
488			for (int i = 0; i < energy.bins.size(); i++) {
489			energy.bins[i].resize(hist_rebinned->GetYaxis()->GetNbins());
490			for (int j = 0; j < energy.bins[i].size(); j++) {
491			energy.bins[i][j] = hist_rebinned->GetBinContent(i+1, j+1);
492			}
493			}
494
495			energy.Xlo = hist->GetXaxis()->GetXmin();
496			energy.Xhi = hist->GetXaxis()->GetXmax();
497			energy.Ylo = hist->GetYaxis()->GetXmin();
498			energy.Yhi = hist->GetYaxis()->GetXmax();
499
500			if (start_ngauss < 1) {
501			start_ngauss = 1;
502			}
503
504	dnisson	1.11	for (mdef->set_ngauss(start_ngauss);
505			mdef->get_ngauss() <=
506			(MAX_GAUSS > start_ngauss ? MAX_GAUSS : start_ngauss);
507			mdef->set_ngauss(mdef->get_ngauss()+1)) {
508
509			int ngauss = mdef->get_ngauss();
510	dnisson	1.1	t_vec.resize(t_vec.size() + 1);
511	dnisson	1.11	int npar = npar_indiv*mdef->get_ngauss();
512	dnisson	1.1	double pval[256], perr[256], plo[256], phi[256];
513			if (npar > 256) {
514			cerr << "Parameter overload" << endl;
515			return r;
516			}
517			vector<TString> pars(npar);
518
519			// initialize parameters
520	dnisson	1.11	mdef->par_init(hist, gMinuit, pars, pval, perr, plo, phi, npar, r);
521	dnisson	1.5
522	dnisson	1.1	// minimize
523			double chisquare, edm, errdef;
524			int nvpar, nparx;
525			trouble t;
526			t.occ = T_NULL;
527			t.istat = 3;
528	dnisson	1.4	// fix sigmas of the Gaussians
529	dnisson	1.1	for (int i = 0; i < ngauss; i++) {
530			double parno[2];
531	dnisson	1.4	parno[0] = i*npar_indiv + 4;
532			gMinuit->mnexcm("FIX", parno, 1, erflg);
533	dnisson	1.1	}
534			gMinuit->mnexcm("SIMPLEX", 0, 0, erflg);
535			gMinuit->mnexcm("MIGRAD", 0, 0, erflg);
536			gMinuit->mnstat(chisquare, edm, errdef, nvpar, nparx, istat);
537			if (istat == 3) {
538			/* we're not concerned about MINOS errors right now
539			gMinuit->mnexcm("MINOS", 0, 0, erflg);
540			gMinuit->mnstat(chisquare, edm, errdef, nvpar, nparx, istat);
541			if (istat != 3) {
542			t.occ = T_MINOS;
543			t.istat = istat;
544			}
545			*/
546			}
547			else {
548			t.occ = T_MIGRAD;
549			t.istat = istat;
550			}
551
552			try {
553			if (t_vec.size() < ngauss) {
554			t_vec.resize(ngauss);
555			}
556			t_vec.at(ngauss-1) = t;
557			}
558			catch (exception ex) {
559			cerr << "exception in fit_histo" << endl;
560			}
561
562			// put parameters in map
563			for (int i = 0; i < npar && i < 256; i++) {
564			int iuint; // internal parameter number
565			gMinuit->mnpout(i, pars[i], pval[i], perr[i], plo[i], phi[i], iuint);
566			}
567			vector<double> pval_copy(npar);
568			vector<double> perr_copy(npar);
569			vector<double> plo_copy(npar);
570			vector<double> phi_copy(npar);
571			for (int i = 0; i < npar && i < 256; i++) {
572			pval_copy[i] = pval[i];
573			perr_copy[i] = perr[i];
574			plo_copy[i] = plo[i];
575			phi_copy[i] = phi[i];
576			}
577			r.pars.push_back(pars);
578			r.pval.push_back(pval_copy);
579			r.perr.push_back(perr_copy);
580			r.plo.push_back(plo_copy);
581			r.phi.push_back(phi_copy);
582
583			// execute user minuit code
584			if (cc_minuit != 0)
585			(*cc_minuit)(gMinuit, hist_rebinned, ngauss);
586
587			int ndof = 0;
588			if (!ignorezero)
589			ndof = energy.bins.size() * energy.bins[0].size();
590			else {
591			for (int i = 0; i < energy.bins.size(); i++) {
592			for (int j = 0; j < energy.bins[i].size(); j++) {
593			if (energy.bins[i][j] > 1.0e-30)
594			ndof++;
595			}
596			}
597			}
598			ndof -= npar;
599
600	dnisson	1.2	r.chisquare.push_back(chisquare);
601	dnisson	1.12	double P = TMath::Prob(chisquare, ndof);
602	dnisson	1.1	if (P > P_cutoff_val) {
603			break;
604			}
605			}
606
607			delete gMinuit;
608			return r;
609			}
610			}