FITL1DTSLICES Namespace Reference

Functions
double	logPoison (double n, double nhat, double logP_min=-999999.9)
double	getLogQuality (const TH1D *data, const TH1D *model, int di, double data_bkg=0.0001, double model_bkg=0.0001)

Function Documentation

logPoison()

double FITL1DTSLICES::logPoison	(	double	n,
		double	nhat,
		double	logP_min = -999999.9 )

Definition at line 29 of file JFitL1dtSlices.cc.

                                                                       {
    if (nhat < 0.0 || n < 0.0) {
      FATAL("logPoisson: n (" << n <<") or nhat (" << nhat << ") is < 0.0" << std::endl);
    }
    if (n == 0.0 && nhat == 0.0) {
      return 0.0; // ok.
    }
    if (n == 0.0 && nhat > 0.0) {
      return -1*nhat; // log( lab^0 * e^-lab / 0! ) = log( e^-lab ) = -lab, ok.
    }
    if (n >= 0.0 && nhat == 0.0) {
      return 0.0; // should be -inf, we only consider bins where we expect non-zero entries: no expected entries we throw away.
    }
    Double_t poisson = TMath::Poisson(n, nhat);
    if (poisson == 0) return 0.0; // only when model and data are close enough do we care about their values, so if P=0, do nothing, since we are not in the region we care about.
    else              return TMath::Log(poisson);
  }

getLogQuality()

double FITL1DTSLICES::getLogQuality	(	const TH1D *	data,
		const TH1D *	model,
		int	di,
		double	data_bkg = 0.0001,
		double	model_bkg = 0.0001 )

Definition at line 47 of file JFitL1dtSlices.cc.

                                                                                                                         {
 
    double q = 0.0;
 
    // If you calculate over the whole histogram you lose bins at edges: only when dt=0 are all N bins considered, everywhere else 
    // either data or model bins get lost at the edge. So integrate over a smaller range:
    int i_low = data->GetNbinsX() / 4; // Start at a quarter of the hist
    int i_hgh = data->GetNbinsX() / 4 * 3; // End at three quarters
    for (int i = i_low; i <= i_hgh; ++i) {
      double n = data_bkg;
      double nhat = model_bkg;
      if (i >= 1 && i <= data->GetNbinsX()) {
        // Count in bin i for the data
        n = data->GetBinContent(i);
      }
      if (i+di >= 1 && i+di <= model->GetNbinsX()) {
        // Count in bin i+di for the model
        nhat = model->GetBinContent(i+di);
      }
      double logP = logPoison(n, nhat);
      q += logP; // Added 0 for when the background is zero in model and data, but then P(n|0) gives inf! So removed overwritten logPmin
    }
    return q;
  }