In case of TGraph (where by definition no errors are defined),
TGraph::Fit recompute the errors given the chisquare and NDF.
If you print the errors on the parameters or SetOptFit,
you will see the correct errors.
Result of fit
p0 = 0.644571 +- 0.016115
p1 = -1.03577 +- 0.181181
See your program modified below.
Rene Brun
{
gStyle->SetOptFit(111);
float Xarr[5];
float Yarr[5];
Xarr[0] = 0.062893; Yarr[0] = 0.584000;
Xarr[1] = 0.000000; Yarr[1] = 0.646000;
Xarr[2] = 0.062893; Yarr[2] = 0.572000;
Xarr[3] = 0.125786; Yarr[3] = 0.497000;
Xarr[4] = 0.125786; Yarr[4] = 0.533000;
c1 = new TCanvas("c1","proba");
TF1 *f1 = new TF1("f1","pol1");
f1->SetRange(0,0.15);
TGraph *tg = new TGraph(5, Xarr, Yarr);
tg->Fit("f1","R","AP");
tg->Draw("A*");
printf("Result of fit\n");
int npar = f1->GetNpar();
for (int i=0;i<npar;i++) {
printf(" %s = %g +-
%g\n",f1->GetParName(i),f1->GetParameter(i),f1->GetParError(i));
}
}
On Sun, 1
Aug 2004, Avi Malki wrote:
>
> Hi Rene,
>
> While trying to do linear fit on a graph using root 4.00/08 and red-hat 7.3
> I noticed I am getting large errors for the fit parameters. I found a
> reference to a similar problem (see below)
> when trying the same code as below, I got the same results of like Voropaev S.G.,
>
> 1 p0 6.44571e-01 8.45154e-01 2.18445e-04 -1.48898e-14
> 2 p1 -1.03577e+00 9.50208e+00 2.45598e-03 -8.82907e-16
>
> Was this problem indeed fixed ?, or am I doing something wrong ?
>
> =================================================================
> The e-mail
> {
>
> Voropaev S.G. wrote:
> >
> > Hello,
> > I have new problem with fitting unsorted data. I am trying to fit unsorted
> > array X,Y to line. I use object TGraph. Parameters of line is well, but error
> > is very strange.
> > Test macro with only 5 data point. (In real life number of data point is
> > larger, but problem still exist):
> >
> > Result of fitting: p0 = 0.64 +- 0.85 !, p1 = -1.04 +- 9.5 !!!
> >
> > Where is error? It is my error or not?
> >
> > I use ROOT 1.02/00 1 august 1997 ( Valery Fine version) on Windows NT
> >
> > With best regards Voropaev S.G.
>
> This problem is now fixed in the development version 1.03/02.
> Result of fitting: p0 = 0.644 +- 0.023, p1 = -1.036 +- 0.256
>
> The fitting algorithm in version 1.02 from TGraph and TGraphErrors objects
> assumed errors = 1 !
> I have now implemented an algorithm that really uses the error specified
> in TGraphErrors and compute an approximation of the errors for TGraph
> objects.
> Thanks for reporting
>
> Rene Brun
> */
>
> float Xarr[5];
> float Yarr[5];
> Xarr[0] = 0.062893; Yarr[0] = 0.584000;
> Xarr[1] = 0.000000; Yarr[1] = 0.646000;
> Xarr[2] = 0.062893; Yarr[2] = 0.572000;
> Xarr[3] = 0.125786; Yarr[3] = 0.497000;
> Xarr[4] = 0.125786; Yarr[4] = 0.533000;
> // Xarr[5] = 0.000000; Yarr[5] = 0.626000;
> // Xarr[6] = 0.125786; Yarr[6] = 0.529000;
> // Xarr[7] = 0.251572; Yarr[7] = 0.432000;
> // Xarr[8] = 0.125786; Yarr[8] = 0.534000;
> // Xarr[9] = 0.251572; Yarr[9] = 0.441000;
> // Xarr[10] = 0.188679; Yarr[10] = 0.480000;
> // Xarr[11] = 0.000000; Yarr[11] = 0.627000;
> // Xarr[12] = 0.377358; Yarr[12] = 0.338000;
> // Xarr[13] = 0.188679; Yarr[13] = 0.467000;
> // Xarr[14] = 0.251572; Yarr[14] = 0.405000;
> // Xarr[15] = 0.000000; Yarr[15] = 0.646000;
> // Xarr[16] = 0.503145; Yarr[16] = 0.252000;
> // Xarr[17] = 0.251572; Yarr[17] = 0.366000;
>
> c1 = new TCanvas("c1","proba");
>
> TF1 *f1 = new TF1("f1","pol1");
> f1->SetRange(0,0.15);
>
> TGraph *tg = new TGraph(5, Xarr, Yarr);
> tg->Fit("f1","R","AP");
> tg->Draw("A*");
> }
>
>
> -------------------------------------------------------------------
> Avi Malki
> Department of Nuclear Physics
> School of Physics and Astronomy
> Tel-Aviv University
> Israel
>
> University - Phone : 972-3-6405280 Fax : 972-3-6406369
> Home - Phone : 972-3-5081734
>
> E-Mail: malki@tauphy.tau.ac.il
> WWW : http://www-nuclear.tau.ac.il/~malki/
> -------------------------------------------------------------------
>
This archive was generated by hypermail 2b29 : Sun Jan 02 2005 - 05:50:09 MET