Integration of correction-lib within CorrectedJets/METFactory
#1134
Conversation
| class CorrectedJetsFactory(object): | ||
| def __init__(self, name_map, jec_stack): | ||
| def __init__(self, name_map, jec_stack, jec_names, jec_year): |
There was a problem hiding this comment.
Please do not alter the signature of this function (especially with new non-defaulted arguments). We want to maintain backwards compatibility for the time being.
Could you try implementing this such that the code detects correction-lib inputs coming in from the JEC stack, and then follow a different code path in the case of correctionlib inputs?
There was a problem hiding this comment.
@lgray thank you for pointing out this. Would it work if jec_stack is a list or a dict if one wants to use correction-lib? IMU, the JECStack class (as well as the auxiliary ones) is not needed when using correction-lib, as the latter already has functionalities to get the right numbers from the formulas that are part of the JME corrections/
There was a problem hiding this comment.
The JEC stack also defines the mapping from names in your data to the arguments of the functions to evaluate. Sure, you can have the jec_stack be something much simpler (like a list) if correctionlib does a bunch of the bookkeeping for you.
Though I think you'll need to do some arranging yourself for the JES uncertainties.
| @@ -171,6 +191,29 @@ def build(self, jets, lazy_cache): | |||
| out_dict = dict(in_dict) | |||
|
|
|||
| # take care of nominal JEC (no JER if available) | |||
| ## General setup to use correction-lib | |||
| clib_json = self.jec_year | |||
| json_path = f"/cvmfs/cms.cern.ch/rsync/cms-nanoAOD/jsonpog-integration/POG/JME/{clib_json}/jet_jerc.json.gz" | |||
There was a problem hiding this comment.
we do not put hard dependencies on cvmfs paths in coffea.
There was a problem hiding this comment.
This should be configurable.
I'd also be careful of building new objects every time you call uncertainties, this can become very slow.
There was a problem hiding this comment.
Also, the json file name would change depending on if it's AK8 or AK4. So, I would also recommend against hard-coding this and instead keeping this in the jec_stack.
| self.jec_names = jec_names | ||
| self.jec_year = jec_year | ||
| ## the last element of jec_names has to be a boolean, specifying if the use wants to save the different correction levels | ||
| self.separated = self.jec_names.pop() |
There was a problem hiding this comment.
This could be a useful debug feature, but from the JME point of view, providing just the name of the JEC tag for the required step should suffice. I don't foresee a need for analyses to store the result of each step, but I'm open to hearing about any exceptions.
There was a problem hiding this comment.
@garvitaa I would suggest keeping it to provide users more flexibility, without the need for them to put the hands on the coffee scripts
|
|
||
| cumCorr = None | ||
| if len(corrections_list) > 0: | ||
| ones = numpy.ones_like(corrections_list[-1], dtype=numpy.float32) |
| has_jer = False | ||
| if self.jec_stack.jer is not None and self.jec_stack.jersf is not None: | ||
| has_jer = True | ||
| elif self.tool == "clib": | ||
| has_jer = True |
There was a problem hiding this comment.
Add a sanity check if the input contains jer or not
| elif self.tool == "clib": | ||
| ## needed to attach to jets the JECs | ||
| jer_out_parms = out.layout.parameters | ||
| jer_out_parms["corrected"] = True |
There was a problem hiding this comment.
Consider to add a method here:
https://github.com/CoffeaTeam/coffea/blob/master/src/coffea/nanoevents/methods/nanoaod.py#L499
| out_dict["jet_energy_resolution"], numpy.float32 | ||
| ), | ||
| awkward.values_astype( | ||
| out_dict["jet_resolution_rand_gauss"], numpy.float32 |
There was a problem hiding this comment.
Try to use the correction-lib random values
|
Following the in-person discussion with @nsmith- , I will implement his suggestions (that I left here and there as comments to my commits), as well as work to make jec_stack a data class when using |
|
Hello, may I ask for the status of this pull request? I am looking for a method where I could ask for compounded JEC from a given list, something like [L1FastJet, L2Relative, L3Absolute, L2L3Residual], and then also get JEC and JER uncertainties in a clean manner. This pull request seems to do that, but perhaps I am wrong? |
|
Hi @green-cabbage , that's correct, feel free to use it and provide any type of feedback! |
|
@anpicci Noted! On the surface, this seems to work with coffea 0.7, which is convenient for me. |
This PR is intended to track the progress on integrating
correction-libwithin the coffea JME tools.The very first contribution is the calculation of the JECs using inputs from the
correction-libtool.FYI @lgray @nsmith-