Reimplement ATLAS Z0 7 TeV Low mass

[comane]

This pull request introduces a filtering module for the ATLAS Z0 7 TeV low mass dataset.

Benchmark with master

(master) https://vp.nnpdf.science/6PBYe_IRTgiWJXpprEGH6Q==

(this branch) https://vp.nnpdf.science/wemXLoM2Sn-87lfEBreTCA==

Compatibility Checks

Covariance matrix check:

In [4]: from validphys.api import API
   ...: import numpy as np
   ...: 
   ...: inp1 = {"dataset_input": {"dataset": f"ATLAS_Z0_7TEV_LOMASS_M"}, "theoryid": 40_000_000, "use_cuts": "internal", "t0pdfset": "NNPDF40_nnlo_as_01
   ...: 180", "use_t0": True}
   ...: inp2 = {"dataset_input": {"dataset": f"ATLAS_Z0_7TEV_LOMASS_M", "variant": "legacy"}, "theoryid": 40_000_000, "use_cuts": "internal", "t0pdfset"
   ...: : "NNPDF40_nnlo_as_01180", "use_t0": True}
   ...: 
   ...: covmat1 = API.covmat_from_systematics(**inp1)
   ...: covmat2 = API.covmat_from_systematics(**inp2)
   ...: 
   ...: t0_covmat1 = API.t0_covmat_from_systematics(**inp1)
   ...: t0_covmat2 = API.t0_covmat_from_systematics(**inp2)
   ...: 
   ...: print(np.all(np.isclose(covmat1, covmat2)))
   ...: print(np.all(np.isclose(t0_covmat1, t0_covmat2)))
True
True

[comane]

Hi @scarlehoff @enocera , tests are failing I think because stat unc are forced to be ADD and UNCORR.
However, for this dataset the stat unc are provided as MULT by hepdata.

I now set the stat unc to ADD as it makes no difference at the level of the covmat.

[scarlehoff]

stat unc are forced to be ADD and UNCORR.

This is correct. I wonder whether having them mult will break something, like the replica generation or closure tests, where the systematics are perhaps being taken explicitly separated.

[scarlehoff]

Hi @comane thanks for this. Could you also set the kinematic_override to be the identity and use one of the process types in the library of process option

This most likely will break something and you might need to change the choice of kinematic variables (from pt to pt2 or smt) and it will also define what to do for the rest of the DY data (the same will be necessary in your datasets @jacoterh @achiefa).

The reason for this is that we want to get rid of these "kinematic overrides" and instead use the right variables directly in the datasets.

(I think you don't need to implement new x-Q mappings or processes because we already had some DY datasets in the data for the pheno project)

[scarlehoff]

Thanks.

The things missing here are:

• Change the kinematic variable from k_i to their actual names
• The kinematics_override should be removed or set to the identity
• The process_type must be change to one of the processes in

  nnpdf/validphys2/src/validphys/process_options.py

  Line 354 in 01eb983

  accepted_variables=(_Vars.y, _Vars.eta, _Vars.m_W2, _Vars.m_Z2, _Vars.sqrts),

• Version comment, bump the version and remove the "Port of old commondata"

[jacoterh]

Hi @comane , I left some very minor comments, looks good otherwise! Does the new implementation still reproduce the old commondata one as you wrote in the beginning of the PR?

[jacoterh]

Very minor, but I think the other implementations put down an actual url, so https://www.hepdata.net/record/ins1288706 in this case

[jacoterh]

The version can be set to 1

[comane]

Hi @jacoterh and @scarlehoff thanks for reviewing this.
This is the report for the last commit (everything stays the same)

https://vp.nnpdf.science/PtYiybc1Rbmd9iXlN3adEQ==