kfactor module maintenance

[giacomomagni]

we should maintain the kfactor module, right now it isn't documented, variables names are not consistent (max_as should look somenting as order_to_update) , and there are some hardcoded pieces which most likely will not work anymore with the new commondata:

pineko/src/pineko/kfactor.py

Lines 401 to 403 in 0645803

	cfac_path = kfactor_folder / f"CF_QCD_{grid}.dat"
	if "ATLASDY2D8TEV" in grid:
	cfac_path = kfactor_folder / f"CF_QCDEWK_{grid}.dat"

[alecandido]

Btw: a huge rework, that could simplify it a lot, is now possible because of merge support in PineAPPL:
NNPDF/pineappl#247

This was not available at the time this module was written (same for scale variations), and because of this many files were generated and later merged.
Ok, PineAPPL subgrids are not NumPy arrays, so not all operations are available. But indexing, scalar multiplications and a combination of sum/concatenation (through merge) is. This should be enough to reimplement it as:

• collect weights and paths (from here to there times this)
• apply the operation in memory

[giacomomagni]

Yes indeed that we can also benefit.
But also other pieces can be simplified:

pineko/src/pineko/kfactor.py

Lines 115 to 121 in 0645803

	max_as = to_construct_order[0] - m
	alpha_val = alphas.alphasQ2(Q2)
	alpha_term = 1.0 / pow(alpha_val, max_as - (nec_order[0] - m))
	k_term = central_k_factor[bin_index] - 1.0
	if order_exists and (max_as - (nec_order[0] - m)) == 0:
	k_term = central_k_factor[bin_index]
	return k_term * alpha_term

for instace here m is subtracted and added back, so can simply be dropped by the function.

[giacomomagni]

@alecandido do you remember what happen in case of merging with an overlapping order ?
It's simply overwritten?

[alecandido]

Not really, the two subgrids will get merged.

The Grid.merge() algorithm is simple: if there is only one, pick that, if there are two, call Subgrid.merge().

Merging subgrids has a different meaning for each subgrid kind, but the overall idea semantic is an addition (e.g. for the n-tuple means concatenating the tuples' lists, and for Lagrange should be actual addition of the weights).