Sigma Mu computation is not generic for JSON sterilised binned data

[jackaraz]

This bug report is created as a TODO for the next iteration of the software

JSON sterilised data does not come with a Bkg tag. Hence the if condition below, which is looking for Bkg within samples, will never be invoked. @Ga0l said that the Bkg tag is added for simplified JSON. This does not generally exist; this is not future proof since it's not generic for any future simplified statistical model generator. A simple workaround can be implemented without the need to check sample names. Since JSON files are independent of the patches, one can go through the entire sample without merging them and extract the necessary information from a non-merged workspace.

The function in question can be found below:

smodels/smodels/tools/pyhfInterface.py

Lines 551 to 573 in 53ffc53

	def getSigmaMu(self, workspace):
	"""given a workspace, compute a rough estimate of sigma_mu,
	the uncertainty of mu_hat"""
	obss, bgs, bgVars, nsig = {}, {}, {}, {}
	channels = workspace.channels
	for chdata in workspace["channels"]:
	if not chdata["name"] in channels:
	continue
	bg = 0.0
	var = 0.0
	for sample in chdata["samples"]:
	if sample["name"] == "Bkg":
	tbg = sample["data"][0]
	bg += tbg
	hi = sample["modifiers"][0]["data"]["hi_data"][0]
	lo = sample["modifiers"][0]["data"]["lo_data"][0]
	delta = max((hi - bg, bg - lo))
	var += delta**2
	if sample["name"] == "bsm":
	ns = sample["data"][0]
	nsig[chdata["name"]] = ns
	bgs[chdata["name"]] = bg
	bgVars[chdata["name"]] = var

Thanks, @Ga0l and @WolfgangWaltenberger, for the explanations.