fix: Treatment of weighted TH1 histograms in comparisons

[danwern]

The function _check_counting_histogram currently prevents plotting comparisons between weighted TH1* histograms.
The issue seems to go back to the different treatment of TH1 in uproot and the ROOT pythonization in the newest version:

• According to the documentation, uproot sets kind=Kind.COUNT for any TH1: https://uproot.readthedocs.io/en/latest/uproot.behaviors.TH1.Histogram.html#kind
• In the newest version of root, however, the choice is made to treat weighted histograms as Kind.MEAN and only unweighted ones as Kind.COUNT: https://github.com/root-project/root/blob/63a25f7e7b4228cf342e3542105c1f6ba6721321/bindings/pyroot/pythonizations/python/ROOT/_pythonization/_uhi.py#L486

I did not find clear instructions in the UHI documentation whether either of those approaches is "incorrect". If both should be supported, I suppose it would be good to support ratios of kind.MEAN in mplhep.

[andrzejnovak]

Hi @danwern thanks for the report! This is being introduced as part of our current merge with plothist. To make this easier for us to test, could you add a snippet/mwe for us to test against?

[danwern]

I hope this mwe helps:

import ROOT
from mplhep import get_comparison

h1 = ROOT.TH1F("h1", "h1", 1, 1.0, 2.0)
h2 = ROOT.TH1F("h2", "h2", 1, 1.0, 2.0)

for val,weight in ([1,1], [2,2]):
    h1.Fill(val, weight)
    h2.Fill(val, weight*10)

values, high_uncertainty, low_uncertainty = get_comparison(
    h1, h2, comparison="ratio"
)

Output

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/path/to/mplhep/comparison.py", line 484, in get_comparison
    _check_counting_histogram([h1_plottable, h2_plottable])
  File "/path/to/mplhep/utils.py", line 820, in _check_counting_histogram
    raise ValueError(msg)
ValueError: The histogram must be a counting histogram, but the input histogram has kind Kind.MEAN.

I also compared the approach with and without weights in this simple test file:

simple test file for ratios of root TH1

import ROOT
import numpy as np
from mplhep import get_comparison

import pytest

@pytest.fixture
def setup_root_histograms():
    """
    Fixture factory to create histograms with custom fills.
    """

    def _create(h1_fill=None, h1_weight=None, h2_fill=None, h2_weight=None):
        h1 = ROOT.TH1F("h1", "h1", 1, 1.0, 2.0)
        h2 = ROOT.TH1F("h2", "h2", 1, 1.0, 2.0)
        if h1_fill is not None:
            if h1_weight is not None:
                for value, weight in zip(h1_fill, h1_weight):
                    h1.Fill(value, weight)
            else:
                for value in h1_fill:
                    h1.Fill(value)
        if h2_fill is not None:
            if h2_weight is not None:
                for value, weight in zip(h2_fill, h2_weight):
                    h2.Fill(value, weight)
            else:
                for value in h2_fill:
                    h2.Fill(value)
        return h1, h2

    return _create

def test_ratio_root(setup_root_histograms):
    """
    Test ratio with simple values.
    """
    ROOT = pytest.importorskip("ROOT")
    # Unweighted
    h1, h2 = setup_root_histograms(h1_fill=[1] * 2, h2_fill=[1] * 20)

    values, high_uncertainty, low_uncertainty = get_comparison(
        h1, h2, comparison="ratio"
    )
    assert pytest.approx(values) == np.array([0.1])
    assert pytest.approx(high_uncertainty) == np.array([0.07416198])
    assert pytest.approx(low_uncertainty) == high_uncertainty

    # Weighted
    h1, h2 = setup_root_histograms(h1_fill=[1] * 5, h1_weight=[1] * 5, h2_fill=[1] * 5, h2_weight=[10]*5)

    values, high_uncertainty, low_uncertainty = get_comparison(
        h1, h2, comparison="ratio"
    )
    assert pytest.approx(values) == np.array([0.1])

[0ctagon]

Thank you for the report @danwern! It seems to be a problem with the ROOT pythonization implementation root-project/root#20265

[0ctagon]

The fix was merged in ROOT root-project/root#20485, thank you for raising this issue.