Jaco ter Hoeve

Hi @scarlehoff , not sure I understand what you are suggesting, could you be more specific? Even though the training/validation ratio is specified per dataset, i.e. `frac: 0.75`, the actual...

> * A fit with trvl: 1.0 (diagonal or not, the result should be equivalent) Results are equivalent, as you can see in this DIS only fit without any split:...

> * A fit in which trvl: 0.75 but you first apply the cuts, then diagonalize training and validation (perhaps with genrep = false to make it easier). That agrees...

Very good, I see. Just to clarify that I am _applying_ the mask after the rotation. I just meant to say that I only _construct_ the mask before the rotation....

Hi @scarlehoff , I have performed the check you proposed. That is, comparing 1. First splitting, then diagonalise training and validation separately 2. Same split, don't diagonalise I put the...

Just for the record and to help us to keep track, these are the reports with the negative modes removed in the diagonal basis: - real data: https://vp.nnpdf.science/V-SA51LcS_Kx7wEx8p6vFg== - fake...

Indeed, I was also just checking this and you're right, it's called by experiment! This wasn't obvious to me before as I had tested it only on datasets within the...

@scarlehoff In case of MHOU, do you have an idea where in validphys the various experiments get collected such that they are treated as one? I'm seeing it works this...

It seems the grouping in case of MHOU is handled by this statement: https://github.com/NNPDF/nnpdf/blob/3d10bbc0a6ba93d33d7951a573fa6bbb8d91456a/validphys2/src/validphys/config.py#L1654

Hi @scarlehoff, I am working on the tests, and I'm seeing that `test_pseudodata.py` results in ``` FAILED validphys2/src/validphys/tests/test_pseudodata.py::test_recreate_fit_pseudodata - AttributeError: 'NoneType' object has no attribute 'is_polarized' FAILED validphys2/src/validphys/tests/test_pseudodata.py::test_recreate_pdf_pseudodata - AttributeError:...