The closure tests depend on the prepare_nnpdf_rng function here
|
def prepare_nnpdf_rng(filterseed:int, rngalgo:int, seed:int): |
which uses the c++ rng. As far as I remember this was kept so that we could use older closure tests while analyzing other parts of the code. This doen't make a lot of sense any longer. Not only are we keeping a dependency on the old code, risking having the same problem when people produce new batches of closure tests, but we end up with required keys that are not documented, e.g. here
https://docs.nnpdf.science/tutorials/closuretest.html
We should get rid of that function and use the same rng as for nfit.
The closure tests depend on the prepare_nnpdf_rng function here
nnpdf/validphys2/src/validphys/filters.py
Line 80 in 94b80c3
which uses the c++ rng. As far as I remember this was kept so that we could use older closure tests while analyzing other parts of the code. This doen't make a lot of sense any longer. Not only are we keeping a dependency on the old code, risking having the same problem when people produce new batches of closure tests, but we end up with required keys that are not documented, e.g. here
https://docs.nnpdf.science/tutorials/closuretest.html
We should get rid of that function and use the same rng as for nfit.