Misc validphys speed ups

[scarlehoff]

Trying to speed up a few things I noticed the convolutions spent a good fraction of the time just moving things around inside the dataframe.

Now, this is necessary because the old fktables didn't follow any particular order... when loading fktables with pineappl instead this is not the case. Among other things because all fktables get casted to the same array:

nnpdf/validphys2/src/validphys/pineparser.py

Line 226 in bcfe00b

for x_point in missing_x_points:

with these changes a full vp-comparefit goes from 40 min to about 20 in my computer.

Note: I have not bothered to port this speed up to old fktables. In principle it could be done just the same by ordering them sensibly after they are loaded.

Note 2: we load the pineappl fktables into an intermediate dataframe that with these changes is no longer necessary, but that's trickier to remove since the fktable is assumed to be a dataframe at various points.

[github-actions]

Greetings from your nice fit 🤖 !
I have good news for you, I just finished my tasks:

• Fit Name: NNBOT-43bd99f9b-2025-05-20
• Fit Report wrt master: https://vp.nnpdf.science/4su3vxtPTJ2AajjC40GkfA==
• Fit Report wrt latest stable reference: https://vp.nnpdf.science/y2aITq03Sh-ooavGNYljHQ==
• Fit Data: https://data.nnpdf.science/fits/NNBOT-43bd99f9b-2025-05-20.tar.gz

Check the report carefully, and please buy me a ☕ , or better, a GPU 😉!

[scarlehoff]

Hi @tgiani, thanks for looking into this. I've added a bunch of comments and reordered a bit the code so that it is clearer what is done for old fktables and for new

[scarlehoff]

All commits from this branch are now in #2317 (I won't delete this branch for now in case we need to split it again, but let me close the PR for the time being)