Updates to implentation of MiNNLO muR/muF polynomial variations (also to use them for unfolding and theory agnostic for OOA) #508
Conversation
… to use them for unfolding and theory agnostic)
|
https://gitlab.cern.ch/cms-wmass/wremnants-data/-/merge_requests/73 will need to be rebased and wremnants-data commit updated |
| @@ -405,6 +406,9 @@ def setup(args, inputFile, inputBaseName, inputLumiScale, fitvar, genvar=None, x | |||
| def assertSample(name, startsWith=["W", "Z"], excludeMatch=[]): | |||
| return any(name.startswith(init) for init in startsWith) and all(excl not in name for excl in excludeMatch) | |||
|
|
|||
| def assertSampleWithOOA(name, startsWith=["W", "Z"], excludeMatch=[]): | |||
| return any(name.startswith(init) for init in startsWith) and name.endswith("OOA") and all(excl not in name for excl in excludeMatch) | |||
|
|
|||
There was a problem hiding this comment.
It might probably be possible to use the function assertSample adding more arguments to require the presence of a string or that it appears at the end (its presence is probably sufficient in this case)
| @@ -588,7 +597,7 @@ def assertSample(name, startsWith=["W", "Z"], excludeMatch=[]): | |||
| for g in cardTool.procGroups["signal_samples"] for m in cardTool.datagroups.groups[g].members}, | |||
| ) | |||
|
|
|||
| if args.muRmuFPolVar and not isTheoryAgnosticPolVar: | |||
| if args.muRmuFPolVar and not isPoiAsNoi: | |||
There was a problem hiding this comment.
Is this change correct? Right now we are always using isPoiAsNoi = True (except maybe for one case with unfolding), so this condition would never be entered
| variation_hist = hist.Hist(*hist_in.axes, name = out_name, storage = hist.storage.Double()) | ||
| variation_hist = variation_hist[...,0:1,:] | ||
|
|
||
| for i in range(len(hist_in.axes[len(hist_in.axes)-2])): |
There was a problem hiding this comment.
Once you select the axis by ID, the number of bins for that axis can be obtained with hist.axes[id].size (or .extent if you want to include possible overflow bins when they exist, .size should always return the number of bins in acceptance regardless).
In general, is it always guaranteed that the axis you want will always be the second from last, rather than the n-th from the left for example ? Perhaps, it might be safer to select the axis by name, which is more guaranteed to stay as it is (and if the name is eventually modified outside then it won't be found here and the code will just raise errors)
|
What's the status of this PR? |
This is still a work in progress, but it contains updates to the MiNNLO muRmuFPolVar. We noticed that the variations we had before were underestimated, we were computing them incorrectly (we were varying only the denominator instead of only the numerator in the ratio sigma_i/sigma_UL). This PR also contains a first implementation of fully correlated variations in a smooth way, at the moment (but it might change, since this is not giving us what we want) the are derived from the combination of the individual uncorrelated variations, directly in the pt-eta plane and not from the Ais directly, but given how these variations are defined it is mathematically the same. The default is still to use the binned ones, but the can be enabled with the "--muRmuFFullyCorr" option is setupCombine (to be given with "--muRmuFPolVar", which changes only the uncorrelated component).