How to create a histogram profile in a root file using uproot?

[rmelchi]

I am new to uproot, but I have noticed that it is able to recognize histogram types in files created by Root. I was wondering if it is possible to create an histogram profile with uproot and store it in a root file.

Thanks

[jpivarski]

Ideally, you'd want to make a profile-style histogram with hist (maybe the WeightedMean Storage type) and assign that into an uproot.WritableDirectory. Unfortunately, it doesn't look like Uproot recognizes that hist type directly:

uproot5/src/uproot/writing/identify.py

Lines 248 to 251 in 130c55b

	if obj.kind == "MEAN":
	raise NotImplementedError(
	"PlottableHistogram with kind='MEAN' (i.e. profile plots) not supported yet"
	)

If Uproot had recognized this histogram type, it would use uproot.writing.identify.to_TProfile to convert it into ROOT format. That function is available, though it has a lot of arguments (for all of the data that would come from the hist object):

uproot5/src/uproot/writing/identify.py

Lines 1552 to 1593 in 130c55b

	def to_TProfile(
	fName,
	fTitle,
	data,
	fEntries,
	fTsumw,
	fTsumw2,
	fTsumwx,
	fTsumwx2,
	fTsumwy,
	fTsumwy2,
	fSumw2,
	fBinEntries,
	fBinSumw2,
	fXaxis,
	fYaxis=None,
	fZaxis=None,
	fYmin=0.0,
	fYmax=0.0,
	fErrorMode=0,
	fNcells=None,
	fBarOffset=0,
	fBarWidth=1000,
	fMaximum=-1111.0,
	fMinimum=-1111.0,
	fNormFactor=0.0,
	fContour=None,
	fOption="",
	fFunctions=None,
	fBufferSize=0,
	fBuffer=None,
	fBinStatErrOpt=0,
	fStatOverflows=2,
	fLineColor=602,
	fLineStyle=1,
	fLineWidth=1,
	fFillColor=0,
	fFillStyle=1001,
	fMarkerColor=1,
	fMarkerStyle=1,
	fMarkerSize=1.0,
	):

The problem to be solved is knowing how to convert from the profile-like histogram you have into the arguments for this function. Once that's solved, the output of this function can be assigned directly into the directory:

with uproot.recreate("some_file.root") as file:
    file["some_profile"] = uproot.writing.identify.to_TProfile(...)

Conversion in the other direction exists: if Uproot extracts a TProfile object from a ROOT file and you call to_boost on it (hist is a subclass of boost-histogram), the data members like fBinEntries and fBinSumw2 get converted into the hist/boost-histogram internal data.

uproot5/src/uproot/behaviors/TProfile.py

Lines 296 to 332 in 130c55b

	def to_boost(self, metadata=boost_metadata, axis_metadata=boost_axis_metadata):
	boost_histogram = uproot.extras.boost_histogram()
	effective_counts = self.counts(flow=True)
	values, errors = self._values_errors(True, self.member("fErrorMode"))
	variances = numpy.square(errors)
	sum_of_bin_weights = numpy.asarray(self.member("fBinEntries"))
	storage = boost_histogram.storage.WeightedMean()
	xaxis = uproot.behaviors.TH1._boost_axis(self.member("fXaxis"), axis_metadata)
	out = boost_histogram.Histogram(xaxis, storage=storage)
	for k, v in metadata.items():
	setattr(out, k, self.member(v))
	if isinstance(xaxis, boost_histogram.axis.StrCategory):
	effective_counts = effective_counts[1:]
	values = values[1:]
	variances = variances[1:]
	sum_of_bin_weights = sum_of_bin_weights[1:]
	view = out.view(flow=True)
	# https://github.com/root-project/root/blob/ffc7c588ac91aca30e75d356ea971129ee6a836a/hist/hist/src/TProfileHelper.h#L668-L671
	with numpy.errstate(divide="ignore", invalid="ignore"):
	sum_of_bin_weights_squared = (sum_of_bin_weights**2) / effective_counts
	# TODO: Drop this when boost-histogram has a way to set using the constructor.
	# New version should look something like this:
	# view[...] = np.stack(sum_of_bin_weights, sum_of_bin_weights_squared, values, variances)
	# Current / classic version:
	view["sum_of_weights"] = sum_of_bin_weights
	view["sum_of_weights_squared"] = sum_of_bin_weights_squared
	view["value"] = values
	view["_sum_of_weighted_deltas_squared"] = variances * (
	sum_of_bin_weights - sum_of_bin_weights_squared / sum_of_bin_weights
	)

You'd want to reverse that transformation.

If you do solve this inverse transformation, we'd be in a good position to add it to uproot/writing/identify.py, so that there won't be a NotImplementedError anymore...