Skip to content

Changing functions which are still using NNPDF pseudodata

Emanuele Roberto Nocera requested to merge removing_cpp_pseudoreplicas into master

Created by: scarlehoff

I'm currently removing the libNNPDF dependencies and decided to start with RandomGenerator which seems the lowest hanging fruit as we now have make_replica which the fit now uses.

However some actions are still using NNPDF pseudodata (which is very bad because that means the pseudodata could perfectly be 100% different from the python version) either calling pseudodata o MakeReplica

These are:

  • n3fit_data_utils.py
  • chi2grids.py::computed_pseudorreplicas_chi2* @Zaharid
  • mc_gen.py::one_art_data_residuals
  • mc_gen.py::art_rep_generation
  • results.py::closure_pseudodata_replicas @Zaharid - [ ] filter.py

Sadly these functions have been mainly developed/touched for people who already left the collaboration so please @Zaharid @siranipour if you could have a close look or give pointers (maybe some functions can be totally removed?) they would be much appreciated.

Also, some of these doesn't seem to be used anywhere (computed_pseudorreplicas_chi2 for instance, I just tried removing it with no consequences for the test...) s

*that function clearly states #TODO: Everythning about this function is horrible. We need to rewrite and I would agree but I think making it use the python pseudodata is more pressing.

Edit: The RandomGenerator cannot be completely taken out from vp since until there is a python-only closure test. I've instead moved the import inside the appropriate function.

Merge request reports