reactivate use_t0_sampling

[comane]

This PR reintroduces the option of generating MC pseudo-data using the t0 covariance matrix

Note: this feature was removed here: #1626

Report:
Compares an nnpdf fit done with t0-covmat sampling against a baseline one

https://vp.nnpdf.science/DMbQZIFqTwecA_N6L5XycQ==/

Iterated fits
https://vp.nnpdf.science/DJ4VtvTZQ5O8fZsl3pSDGQ==

• Fix tests failing because of new use_t0_sampling default

[RoyStegeman]

Please document it and include it in one of the example runcards

[comane]

Please document it and include it in one of the example runcards

Hi @RoyStegeman, where would you add this in the documentation?

[RoyStegeman]

Have a look at the docs and see where you think it should go or fits best. Probably in the n3fit runcard section (which is already incomplete) is enough, but maybe I missed something

[scarlehoff]

So, as far as I can see this will become the default from now on.

[RoyStegeman]

So, as far as I can see this will become the default from now on.

Why? Did I miss a discussion about this? Stefano was always in favor of sampling exp so does he agree with doing this now?

[comane]

So, as far as I can see this will become the default from now on.

Why? Did I miss a discussion about this? Stefano was always in favor of sampling exp so does he agree with doing this now?

During the code meeting today, SF was in favour of having t0 as sampling covmat.

[scarlehoff]

Stefano was always in favor of sampling exp so does he agree with doing this now?

fwiw, he was not completely convinced about the toy model but since even in the extreme toy model case the difference was sub per-mille he said it was academic at that point (while making things easier for the diagonal covmat)

[RoyStegeman]

Okay good. It also makes things easier for TCM so I'm happy with this

[RoyStegeman]

When do you plan to fix and merge this?

[scarlehoff]

Remember that for the fit regression tests you can use the label redo-regression

(also, please run the fitbot and update the reference)

[github-actions]

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

• Fit Name: NNBOT-b382bbb69-2025-03-24
• Fit Report wrt master: https://vp.nnpdf.science/xNF3yakpQHqtoCvV9KrUvQ==
• Fit Report wrt latest stable reference: https://vp.nnpdf.science/GdohBuapStqERtNPuN_W3A==
• Fit Data: https://data.nnpdf.science/fits/NNBOT-b382bbb69-2025-03-24.tar.gz

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

[scarlehoff]

A byproduct of using t0 for the sampling is that now generating pseudodata requires a predictions with some theory, which in turns require cuts.

Some of the tests should be modified to set use_t0_sampling: False (and some might need to be updated manually, some of the regression tests in validphys/n3fit might fall into this category).

[scarlehoff]

I'm also worried that the fitbot here produced exactly the same result as in this other branch https://vp.nnpdf.science/WUyHqXh5QsSxY24iIxQHxw== ?

[RoyStegeman]

What is the status on this? Since it was agreed that this is what should be done going forwards there is no need to keep it out of master for so long

[scarlehoff]

lgtm

Just added two requests for tests.

[scarlehoff]

When the new fitbot finishes, please update the fitbot reference and I think we can merge?

[comane]

When the new fitbot finishes, please update the fitbot reference and I think we can merge?

Yes, I think that we can merge. What do you mean by updating the fitbot reference?

[github-actions]

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

• Fit Name: NNBOT-5684b6fe5-2025-05-12
• Fit Report wrt master: https://vp.nnpdf.science/BPTinb_0TQ6bDVm8UktLlw==
• Fit Report wrt latest stable reference: https://vp.nnpdf.science/nk1cR29yQfiOOKJa96K4_A==
• Fit Data: https://data.nnpdf.science/fits/NNBOT-5684b6fe5-2025-05-12.tar.gz

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

[scarlehoff]

Thanks, after this + the diagonal covmat I think we can start with the 4.1.X tags.