Error band from fitting using TMultiGraph

[lamya86]

Hi all,
I have a function to fit multiGraph and I would like to show the uncertainties of the fit. I have tried to follow the example of ConfidenceIntervals.C but the band of the uncertainties is not shown (see attached plot). Any help or tips of how I can show the band of the uncertainties from fitting. Also, is there away to show the band within 1 or 2 sigma?
Thanks in advance,
below is a part of my code

mg_eff_1 = TMultiGraph()
mg_eff_1.Add( data_faceIDcut_inb10p6Allbin)
mg_eff_1.Add( data_faceIDcut_inb10p2Allbin)
mg_eff_1.Add( data_faceIDcut_outb10p6Allbin)
mg_eff_1.Add( oldclas6_data)

mg_eff_1.SetTitle('#scale[1.2]{#bf{#font[22]{Neutron Detection Eff}}}; #scale[1.2]{#bf{#font[22]{P_{mm} [GeV]}}} ; #scale[1.2]{#bf{#font[22]{NDE}}}')
mg_eff_1.GetHistogram().SetMaximum(1.0)
mg_eff_1.GetHistogram().SetMinimum(0.0)
mg_eff_1.GetXaxis().SetTitleOffset(1.3);
mg_eff_1.GetXaxis().SetTitleSize(0.05);
mg_eff_1.GetXaxis().SetLabelSize(0.05);
mg_eff_1.GetXaxis().SetLabelFont(22);

mg_eff_1.GetYaxis().SetTitleOffset(1.2);
mg_eff_1.GetYaxis().SetTitleSize(0.05);
mg_eff_1.GetYaxis().SetLabelSize(0.05);
mg_eff_1.GetYaxis().SetLabelFont(22);

label.DrawLatex(0.35, 0.40, caption_oldclas6_data)
label.DrawLatex(0.35, 0.36, caption_data_faceIDcut_inb10p6Allbin)

label.DrawLatex(0.35, 0.32, caption_data_faceIDcut_outb10p6Allbin)
label.DrawLatex(0.35, 0.28, caption_data_faceIDcut_inb10p2Allbin)
mg_eff_1.Draw("AP")
can.Print('{}'.format(mon_out_file_name))

# define fit function
fnde = TF1("fnde","[0]*(1-(1/(1+TMath::Exp(((x-[1])/[2])))))",1.0,7.8)
fnde.SetLineColor(2)
fnde.SetLineWidth(3)
fnde.SetParameter(0, 7.24281e-01)
fnde.SetParameter(1, 7.79712e-01)
fnde.SetParameter(2, 8.02135e-01)
mg_eff_1.Fit('fnde', 'RB')
mg_eff_1.Draw("AP");
ngr2 =36
# grint = TGraphErrors(data_faceIDcut_outb10p6Allbin.GetN())
interval = TGraphErrors(data_faceIDcut_outb10p6Allbin.GetN())
for i in range(0, ngr2):
    # print(i, data_faceIDcut_outb10p6Allbin.GetX()[i])
    interval.SetPoint(i, data_faceIDcut_outb10p6Allbin.GetX()[i], data_faceIDcut_outb10p6Allbin.GetY()[i])
    ROOT.TVirtualFitter.GetFitter().GetConfidenceIntervals(interval, 0.98)
  
interval.SetFillColor(kRed-7)

interval.SetFillStyle(3004);
interval.SetFillColor(kRed)

interval.Draw("e3 same")
mg_eff_1.Draw("AP same");

can.Print('{}'.format(mon_out_file_name))


can.Print('{}]'.format(mon_out_file_name)) 

[

]

[couet]

Your code cannot be run as it is. It is quite incomplete.
What do you get when you do the single plot:

interval.Draw("A E3");

?

[lamya86]

@couet

nterval.Draw("A E3");

Hi couet,
Thanks for the quick replay. still, the band error is not shown. Attached below is a complete code with text files that are needed to run my code.

fiteff.py (25.6 KB)
validate_nde_tcor_inb10p6_data.txt (2.1 KB)
validate_nde_tcor_inb10p2_data.txt (2.1 KB)
validate_nde_tcor_outb10p6_data.txt (2.1 KB)

[couet]

Try only:

interval.SetFillColor(kRed-7)

interval.SetFillStyle(3004);
interval.SetFillColor(kRed)

interval.Draw("A E3")

Do no draw mg_eff_1
What do you get ?

[lamya86]

here what I got. is it a possible to be too small !!

[couet]

Yes that’s it. Change the fill style your will see better. These hatches are not the best for a such thin polygon.

[lamya86]

Thanks @couet , I wondering if there is a way to show the band error within only the range of fitting. because one I changed the xvalue to create the band as:

clas12_x_bins = sorted([1.4625, 1.5375, 1.6125, 1.6875, 1.7625, 1.8375, 1.9125005, 
                        1.987501, 2.0750005, 2.175, 2.25, 2.375, 2.475, 2.625,
                        2.825, 3.025, 3.325, 3.725, 4.1625, 4.8, 5.6, 6.75, 7, 7.8])

# pmm.append(float(x_val[ii]))
# grint = TGraphErrors(data_faceIDcut_outb10p6Allbin.GetN())
interval = TGraphErrors(data_faceIDcut_outb10p6Allbin.GetN())
for i in range(0, 24):
    interval.SetPoint(i, clas12_x_bins[i], 0)
    # print('interval ', interval)
    ROOT.TVirtualFitter.GetFitter().GetConfidenceIntervals(interval, 0.99)

I end up as what you see in attached

[couet]

I do not understand how you get this. When I zoom a graph-errors drawn with option 3 the clipping is correct. Can you provide a small reproducer showing this effect? or save the graph in a root file.

[lamya86]

attached is a root file for the graph. I am not familiar with the right way to save a graph with error band and below is what I did to save it.

fout = TFile('fiteff.root','recreate')
mg_eff_1.Write("mg_eff_1")
interval.Write("interval")
fout.Close()

fiteff.root (10.6 KB)

[couet]

I see, it is not a problem of range but a problem of points ordering. The last point of the graph has the lower X value.
do:

 interval->Sort();
 interval->Draw("E3");

[lamya86]

@couet

see, it is not a problem of range but a problem of points ordering. The last point of the graph has the lower X value.

Thank you so much @couet but now I getting the error band below 1.5, how I can remove it? I do not quite understand why if I fit a certain range =[1.5-8], I see the error for fitting in the whole range.

[couet]

Si it means that points is an extra one. Unwatered. Instead of Sort try Remove

[lamya86]

Thank you so much @couet . Now it works. I have just one more question. How I can print the value of the uncertainty from fitting? I would like to show the band as I have drawn it by hand (black) instead of showing it in the curve.

[couet]

I think the output of the fit should give the information ?
May be @moneta can tell more.

[moneta]

Hi,
for drawing the function confidence bands after fitting, please look at this tutorial:
https://root.cern.ch/doc/master/ConfidenceIntervals_8C.html

Best,

Lorenzo

[lamya86]

Hi @moneta ,
yes I followed the example and I already got the band. My question is how I can print the output of the band. I would like to show the band of uncertainty as a black band in the figure (I just drew it by my hand) instead of showing it in the data points (as you see it with kMagenta color)

[moneta]

You can get the band values either as the error of the TGraphErrors grint in the given example, or you can also use directly the function FitResult::GetConfidenceIntervals.