CMS 3D CMS Logo

/data/refman/pasoursint/CMSSW_5_3_0/src/CommonTools/Statistics/src/LinearFitErrorsIn2Coord.cc

Go to the documentation of this file.
00001 #include "CommonTools/Statistics/interface/LinearFitErrorsIn2Coord.h"
00002 #include "CommonTools/Statistics/interface/LinearFit.h"
00003 #include <cmath>
00004 
00005 float LinearFitErrorsIn2Coord::slope(
00006   const std::vector<float> & x, const std::vector<float> & y, int ndat, 
00007   const std::vector<float> & sigx, const std::vector<float> & sigy) const
00008 {
00009 
00010   // scale y and sigy, compute scaled errors
00011   float scale = sqrt(variance(x, ndat) / variance(y, ndat));
00012   std::vector<float> yScaled = y;
00013   std::vector<float> sigyScaled = sigy;
00014   std::vector<float> sig(ndat);
00015   for (int i = 0; i != ndat; i++) {
00016     yScaled[i] *= scale; 
00017     sigyScaled[i] *= scale;
00018     sig[i] = sqrt(sigx[i]*sigx[i] + sigyScaled[i]*sigyScaled[i]);
00019   }
00020 
00021   // usual linear fit
00022   LinearFit lf;
00023   float fs, fi, covss, covii, covsi;
00024   lf.fit(x, yScaled, ndat, sig, fs, fi, covss, covii, covsi);
00025 
00026   // unscale result
00027   fs /= scale;
00028   return fs;
00029 
00030 }
00031 
00032 
00033 float LinearFitErrorsIn2Coord::intercept(
00034   const std::vector<float> & x, const std::vector<float> & y, int ndat, 
00035   const std::vector<float> & sigx, const std::vector<float> & sigy) const
00036 {
00037   
00038   float fs = slope(x, y, ndat, sigx, sigy);
00039   float fi = 0;
00040   float sumWi = 0;
00041   for (int i = 0; i != ndat; i++) {
00042     float wi = 1./(sigy[i] + fs*fs*sigx[i]);
00043     fi += wi*(y[i] - fs*x[i]);
00044     sumWi += wi;
00045   }
00046 
00047   return fi / sumWi;
00048 
00049 }
00050 
00051 
00052 float 
00053 LinearFitErrorsIn2Coord::variance(const std::vector<float> & x, int ndat) const 
00054 {
00055   double m1 = 0., m2 = 0.;
00056   for (int i = 0; i != ndat; i++) {
00057     m1 += x[i];
00058     m2 += x[i]*x[i];
00059   }
00060   m1 /= ndat;
00061   m2 /= ndat;
00062 
00063   return float(m2 - m1*m1);
00064 
00065 }