Polarised proton fits

[Radonirinaunimi]

This PR implements the necessary changes to perform polarised fits. It deprecates the intial implementation in #1893.

Remaining Todos:

• add positivity bound plots to report
• add some tests (Some tests, such as test_fit, requires Polarised DIS and DY Commondata #1816 to be merged and then rebased here.)
• generate the files for regression tests (will be regenerated at the very end with CI)
• fix positivity dataset plots no longer needed as superseded by BC plots

[scarlehoff]

@Radonirinaunimi it would be good to rebase again since the changes to the convolution recently implemented will have an effect here (but they also simplify a lot what needs to be done, in that everything will be much more clear)

[Radonirinaunimi]

@Radonirinaunimi it would be good to rebase again since the changes to the convolution recently implemented will have an effect here (but they also simplify a lot what needs to be done, in that everything will be much more clear)

Yep, I am doing so at the moment. Hopefully this will not be painful 🙃

[Radonirinaunimi]

@Radonirinaunimi it would be good to rebase again since the changes to the convolution recently implemented will have an effect here (but they also simplify a lot what needs to be done, in that everything will be much more clear)

Yep, I am doing so at the moment. Hopefully this will not be painful 🙃

It was slightly tricky but I believe it works. I will clean up a bit while adding some docs and tests.

[scarlehoff]

I think this is good to go. Thank you for the hard work! I like how they are changes that are relatively major for how the fit works but are small nudges to the code ^__^

The only thing left for me to be happy would be having a fit + report with this latest version of the code (after comments are applied or ignored in case-by-case basis xd) before merging

[Radonirinaunimi]

The only thing left for me to be happy would be having a fit + report with this latest version of the code (after comments are applied or ignored in case-by-case basis xd) before merging

I already read your mind earlier and the evolutions are running atm xd Will generate some reports tomorrow while addressing your suggestions.

[Radonirinaunimi]

The reports are now failing because of some recent changes in the computation of stats that appear following lines:

nnpdf/validphys2/src/validphys/pdfplots.py

Lines 617 to 619 in aeb2a96

	unpol_cv = unpol_stats.central_value()
	unpol_stdv = unpol_stats.errorbarstd()
	ax.plot(xgrid, unpol_cv + unpol_stdv, color="red")

I am fixing this now.

[scarlehoff]

What changes?
But in any case, that function is the only non-trivial so if you can deal also with the TODO that would be perfect.

[scarlehoff]

That should only happen if you are getting a list and not an array, and that should definitely not happen because the data is made into an array upon entering the Stats class.

nnpdf/validphys2/src/validphys/core.py

Line 828 in aeb2a96

self.data = np.atleast_2d(data)

So there might be something wrong somewhere else?

[Radonirinaunimi]

With possibly some improvements in the sumrules module, namely (for my own recollection) modifying the _integral function to return the individual limit instead of directly their sums, I think this is now finished.

Here are some reports:

• comparing NLO and NNLO: https://vp.nnpdf.science/-E_ygtqjQl6JON209Ilp1w==/ The fits were done with not the very last commit but the report is (which is fine since the last commits only regard plotting)
• comparing published sets with NLO fits: https://vp.nnpdf.science/HKvrJTOtRIOQlqbNjPLKoQ==/
• comparing published sets with NNLO fits: https://vp.nnpdf.science/7YDRSpWLT42oh6JFRdcwpQ==/

PS: These will be the fits we'll send to Stagnitto et al.

[github-actions]

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

• Fit Name: NNBOT-489c196ac-2024-04-18
• Fit Report wrt master: https://vp.nnpdf.science/LUbVtjTAQgitiZEb2s-iHQ==
• Fit Report wrt latest stable reference: https://vp.nnpdf.science/mJqhP2O2QdmkR7bw2SV97A==
• Fit Data: https://data.nnpdf.science/fits/NNBOT-489c196ac-2024-04-18.tar.gz

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

[scarlehoff]

The fitbot corresponding to the last commit #2057 (comment)