Use directly validphys objects in n3fit

[scarlehoff]

Once #1680 is done the following step is to remove the unnecessary layer in n3fit that basically runs over the vp objects and makes them into dictionaries. Making ModelTrainer approximately 2000 lines smaller

[Zaharid]

We did a bit of this in Cambridge for simunet (but for new kinds of objects rather than the existing ones). But not sure if it is worth trying to port it (and I am not going to have time myself).

A few ideas from there:

• We'd need a few conveninece methods, such as Commondata.with_central_value, so that we can easily put the pseudodata in there (that @comane already added), but also (loaded) dataset wrapper classes with something along the lines of

    def split(self, trmask):
        """Split the set into two based on the boolean mask ``trmask``."""
        vlmask = ~trmask
        tr = self.__class__(
            commondata=self.commondata.with_cuts(trmask),
            prediction=self.prediction.with_cuts(trmask),

        )
        vl = self.__class__(
            commondata=self.commondata.with_cuts(vlmask),
            prediction=self.prediction.with_cuts(vlmask),
            
        )
        return tr, vl

(where we'd also need self.prediction).

Having these kinds of things would allow to refactor things like the large dictionary interfaces at a later stage.

[Zaharid]

@manuel-morales-a also looked bit into this.

[scarlehoff]

I think the only thing left from the vp point of view is interfacing this

nnpdf/validphys2/src/validphys/n3fit_data_utils.py

Line 16 in 97e0e13

class FittableDataSet:

with the commondata (with a variation of what you posted, basically have .tr_data(), vl_data() and central_data() or whatever.

(And create the grouping to have access to the relevant covmat)