Making n3fit use python pseudodata

[siranipour]

No description provided.

[wilsonmr]

Nice!

We also want to be able to make closure data so I think the make_replica function will want to be able to take some different central values and generate data from there. I half wonder if one could add a method to load the filtered commondata to the CommonData class and then just explicit node loaded_commondata_with_cuts to either load the closure data or the regular commondata.. It's a smaller change but further ingrains the closure data system I know @Zaharid hates. The next thing is at some point we would also want the closure central values (or level 1 data) to be generated with the new code..

[wilsonmr]

Actually wait, I think because the commondataparser takes a CommonDataSpec which, when use_fitcommondata=True, sets the data path to the filtered data, this might already do the right thing!

EDIT: opened an issue where I will test this #1269

[siranipour]

Do we still want this seed logic:

nnpdf/validphys2/src/validphys/n3fit_data.py

Line 192 in 2d49938

base_mcseed = int(hashlib.sha256(str(data).encode()).hexdigest(), 16) % 10 ** 8

[Zaharid]

@siranipour probably we can do something simpler.

Do we still want this seed logic:

nnpdf/validphys2/src/validphys/n3fit_data.py

Line 192 in 2d49938

base_mcseed = int(hashlib.sha256(str(data).encode()).hexdigest(), 16) % 10 ** 8

[scarlehoff]

The generate_data_replica function should be removed right?

Also, any changes to the seed logic I would prefer it if they are done in separate PRs since they will also need to change all regression tests + fit bot, etc

[siranipour]

The generate_data_replica function should be removed right?

Yeah it can, I had just left it for now in case we wanted to use the old behaviour at any point

[siranipour]

The generate_data_replica function should be removed right?

Also, any changes to the seed logic I would prefer it if they are done in separate PRs since they will also need to change all regression tests + fit bot, etc

I'm beginning to think the changes you mention here need to be done in this PR since the pseudodata generation does not replicate precisely the C++ pseudodata. I believe this is why the linux tests are failing currently

[scarlehoff]

Oh. Didn't realise that. In that case yes, do the changes here. Leave just the bot for another PR (or for the very last commit) so that we have the report of this PR vs master.

[siranipour]

@scarlehoff is there any neat way of generating the quickcard_{replica}.json regression files (e.g in validphys it generates them automagically when you first run pytest) or do I need to create them manually?

[scarlehoff]

Nop, you have to do it manually.

[github-actions]

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

• Fit Name: NNBOT-d3034a7cc-2021-06-11
• Fit Report: https://vp.nnpdf.science/BqVRv_jmSw2f3jGZfS58Bw==
• Fit Data: https://data.nnpdf.science/fits/NNBOT-d3034a7cc-2021-06-11.tar.gz

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

[siranipour]

@scarlehoff the bot has been fixed in 0b21f84 so this can be merged if you're happy