BTag Tools and "RESHAPE" option

[aebid]

Hi experts, I have been trying to use the BTagScaleFactor class to calculate the SF from a CSV file and am running into an issue. From what I understand, there are 4 options, "LOOSE", "MEDIUM", "TIGHT", and "RESHAPE".

According to this line in the eval() function,

coffea/coffea/btag_tools/btagscalefactor.py

Lines 197 to 199 in 65b137f

	discr : numpy.ndarray or awkward.Array, optional
	The jet tagging discriminant value (default None), optional for all scale factors except
	the reshaping scale factor

the discriminator should be an optional use for all cases but necessary for the "RESHAPE" case. But the discriminator is then ignored for cases not "RESHAPE"

coffea/coffea/btag_tools/btagscalefactor.py

Lines 209 to 220 in 65b137f

	if systematic not in self._corrections:
	raise ValueError("Unrecognized systematic: %s" % systematic)
	if self.workingpoint == BTagScaleFactor.RESHAPE:
	if discr is None:
	raise ValueError("RESHAPE scale factor requires a discriminant array")
	return self._corrections[systematic](
	flavor, eta, pt, discr, ignore_missing=ignore_missing
	)
	else:
	return self._corrections[systematic](
	flavor, eta, pt, ignore_missing=ignore_missing
	)

Is this intended or am I confused about something? Looking at my CSV file from the BTagCalibration twiki, the other options should still use the discriminator. Even in the dense_map_lookup dictionary the discriminator edges are ignored.

coffea/coffea/btag_tools/btagscalefactor.py

Lines 167 to 180 in 65b137f

	if self.workingpoint == BTagScaleFactor.RESHAPE:
	self._corrections[syst] = dense_mapped_lookup(
	(self._flavor, edges_eta, edges_pt, edges_discr),
	mapping,
	numpy.array(corr["formula"]),
	3,
	)
	else:
	self._corrections[syst] = dense_mapped_lookup(
	(self._flavor, edges_eta, edges_pt),
	mapping[..., 0],
	numpy.array(corr["formula"]),
	2,
	)

Thanks!

[nsmith-]

In the past the correction algorithm did not depend on the discriminator value for cases other than reshape. It may have changed, but I would be surprised, since with a working point the discriminator effectively becomes a boolean (either greater or less than the WP).
That said, can you use the correctionlib json files for your use case? I was under the impression the csv is being phased out.