How to plot with a linear and log scale on two different axes on same plot?

[feickert]

I am trying to figure out how to plot something on a linear axis and then plot another quantity on a log axis on the same plot and then have the two y-axes be on different scales as well. I’m currently attempting to do this with drawing the first histogram on a TPad and then drawing the second histogram on a transparent TPad above it. However, I am not sure how to get a log scale for the second histogram (the log scale is a property of the TCanvas I think, not a TPad) and then how to use TGaxis properly with the log scale and second histogram.

Any suggestions are very welcome.

I’m attached an example RootPy script that mocks up a demo of something close to what I’m trying to do (linear_log_axes_test.py (1.7 KB)) and I’m running it on lxplus with

$ lsetup root
$ root-config --version
6.10/04

The script contents and output are below as well

import ROOT
ROOT.gROOT.SetBatch(True)  # Supress output to screen


def main():
    # Create the canvas
    can = ROOT.TCanvas('c', '', 800, 600)
    x_range = [10, 70]

    # Use a base pad and then a transparent pad
    base_pad = ROOT.TPad('base_pad', '', 0, 0, 1, 1)
    clear_pad = ROOT.TPad('clear_pad', '', 0, 0, 1, 1)
    clear_pad.SetFillColor(0)
    clear_pad.SetFillStyle(4000)
    clear_pad.SetFrameFillStyle(0)

    base_pad.Draw()
    base_pad.cd()

    # Turn off the tick marks on the right hand side y-axis
    base_pad.SetTicky(0)

    h1 = ROOT.TH1F('h1', '', 5, x_range[0], x_range[1])
    for itr in range(0, 1001):
        h1.Fill(ROOT.gRandom.Gaus(40, 5))
    h1.Scale(1 / h1.Integral())
    h1.SetLineColor(ROOT.kBlack)
    h1.GetYaxis().SetRangeUser(0, 1.4)
    h1.GetYaxis().SetTitle('Hist 1 axis label')
    h1.GetXaxis().SetTitle('x axis label')

    h1.Draw()

    # distiguish visually efficiency of 1
    eff_of_one_line = ROOT.TLine(x_range[0], 1.0, x_range[1], 1.0)
    eff_of_one_line.SetLineStyle(2)
    eff_of_one_line.Draw('SAME')

    clear_pad.Draw()
    clear_pad.cd()
    clear_pad.SetTicks(0, 0)

    # The pad should be on a log scale, but how?
    can.SetLogy(1)

    h2 = ROOT.TH1F('h2', '', 5, x_range[0], x_range[1])
    for bin in range(1, 6):
        h2.SetBinContent(bin, 0.2 * h1.GetBinContent(bin))
    h2.Scale(1 / h2.Integral())
    h2.SetLineColor(ROOT.kBlue)
    h2.Draw('A')

    log_axis_y_min = 0
    log_axis_y_max = 1
    log_axis = ROOT.TGaxis(70, 0, 70, 1, log_axis_y_min,
                           log_axis_y_max, 510, '+L')
    log_axis.SetTitle('hist 2 axis label')
    log_axis.Draw()

    can.SaveAs('linear_log_axes_test.pdf')


if __name__ == '__main__':
    main()

[couet]

The log attribute belongs to the pad. So you can set log scale on a pad.

[feickert]

@couet correct. I have internal data in my real plot which is causing me to get a

Error in <THistPainter::PaintInit>: Cannot set Y axis to log scale

at the moment when I try to do the

clear_pad.SetLogy(1)

so I guess I need to debug that before I can go any farther with dealing with the TGaxis

[couet]

You do not need TGAxis. You can plot a Pad with log scale on top of a Pad in linear scale.
Then the option X+ and/or Y+ will allow you to draw the axis on top or the right when you
draw the histogram is log scale.

[feickert]

@couet That works great! Thanks!

The other problems I have are due to the ATLAS Style enforcement messing up the canvas. If I just use vanilla ROOT then I get what I wanted. Here is an example of what all I was trying to do: linear_log_axes_test.py (1.9 KB)

import ROOT
ROOT.gROOT.SetBatch(True)  # Supress output to screen


def main():

    ROOT.gStyle.SetOptStat(0)
    # Create the canvas
    can = ROOT.TCanvas('can', '', 800, 600)
    x_range = [10, 70]

    # Use a base pad (the canvas) and then a transparent pad
    clear_pad = ROOT.TPad('clear_pad', '', 0, 0, 1, 1)
    clear_pad.SetFillColor(0)
    clear_pad.SetFillStyle(4000)
    clear_pad.SetFrameFillStyle(0)

    # Turn off the tick marks on the right hand side y-axis
    can.SetTicky(0)

    h1 = ROOT.TH1F('h1', '', 35, x_range[0], x_range[1])
    h1.GetXaxis().SetTitle('x axis label')
    h1.GetYaxis().SetTitle('Left y axis label')
    h1.GetYaxis().SetRangeUser(1e-16, 1.4)
    h1.SetLineColor(ROOT.kBlack)
    h1.SetMarkerColor(ROOT.kBlack)
    h1.SetMarkerStyle(ROOT.kFullCircle)

    for i in range(h1.GetNbinsX() + 1):
        if i % 2 == 0:
            h1.SetBinContent(i, 0.8)
        else:
            h1.SetBinContent(i, 0.7)
        h1.SetBinError(i, 0.1)

    h1.Draw('')

    # distiguish visually efficiency of 1
    eff_of_one_line = ROOT.TLine(x_range[0], 1.0, x_range[1], 1.0)
    eff_of_one_line.SetLineStyle(2)
    eff_of_one_line.Draw('SAME')

    # Draw h2 information
    clear_pad.Draw()
    clear_pad.cd()
    clear_pad.SetTicks(0, 0)
    # h2 should be on a log scale
    clear_pad.SetLogy(1)

    h2 = ROOT.TH1F('h2', '', 35, x_range[0], x_range[1])
    h2.GetYaxis().SetTitle('Right y axis label')
    h2.GetYaxis().SetTitleOffset(1.3)
    h2_y_min = 1e-2
    h2_y_max = 20
    h2.GetYaxis().SetRangeUser(h2_y_min, h2_y_max)
    h2.SetLineColor(ROOT.kBlue)
    h2.SetMarkerColor(ROOT.kBlue)
    h2.SetMarkerStyle(ROOT.kOpenCircle)

    for i in range(h2.GetNbinsX() + 1):
        if i % 2 == 0:
            h2.SetBinContent(i, 0.15)
        else:
            h2.SetBinContent(i, 0.1)
        h2.SetBinError(i, 0.1)

    h2.Draw('Y+')

    can.SaveAs('linear_log_axes_test.pdf')


if __name__ == '__main__':
    main()

[couet]

When i execute the script you posted I get the plot you posted also, which seems to me is correct … what should do to see the invalid plot ?

[feickert]

@couet You need to also do

ROOT.gROOT.LoadMacro('./your-path-to-goes-here/AtlasStyle.C')
ROOT.SetAtlasStyle()

at the top. Though I also ended up taking out the legend and watermarks that were getting pushed around and causing issues, so that alone might not cause issues.

[couet]

Sorry but I do not have AtlasStyle.C

[feickert]

@couet That’s okay. This isn’t really a ROOT issue anymore. I can look into the bug on my own later and follow up with the necessary people in ATLAS. Thanks for all your help!

[couet]

Ok, that’s good. Do not hesitate to come back in case of problem.