Return all members when using MC PDFs #1522

scarlehoff · 2022-02-15T15:29:54Z

As the title suggest.

After the two previous PR (#1515 and #1517) where Stats where used everywhere, the only two things that are problematic are:

The sumrules (I guess nobody cared whether the central value was computed because by definition it would fall inside the envelope anyway?)
The N3PDF interface (which was a hack to be compatible with what vp was doing, so now the hack has to be undone).

Nothing is evidently broken, but many thing might be broken in not obvious ways. The first commit is just so that the regression for the sumrules passes so I can update it in the next commit (where all members will be computed also there for both Hessian and MC PDFs).

The last commit will be a simplification of the N3PDF interface to validphys. And that should be all.

After all this series of PR nothing should be accessing libNNPDF results any longer other than the commondata datapoints which are immediately wrapped into a Stats class:

nnpdf/validphys2/src/validphys/results.py

Line 81 in 0321658

self._central_value = dataobj.get_cv()

but loading them in python is the next step: #1511 :)

(ofc, this last one is not necessary to destroy C++ but since we were already here we might as well do it, if you guys prefer to keep doing MC pdfs as we were doing up to know please let me know asap so I don't build on top of this PR)

github-actions · 2022-02-16T19:33:06Z

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

Fit Name: NNBOT-a5ce2abff-2022-02-16
Fit Report: https://vp.nnpdf.science/dRpTyDA4QVib3q273C5F7Q==
Fit Data: https://data.nnpdf.science/fits/NNBOT-a5ce2abff-2022-02-16.tar.gz

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

scarlehoff · 2022-02-16T19:39:20Z

almost there

scarlehoff · 2022-02-17T09:09:05Z

This should be ready for review. We can talk this afternoon about who is going to review this because I guess all three PR should be done by the same person and hopefully at the same time.

I've run all the examples in the validphys folder with the current commit. I can gzip them all and send them to the reviewer if they wish.

The fit bot can be run when we get the extra space.

Zaharid

Will take me a while to go over the bits.

n3fit/src/n3fit/vpinterface.py

Co-authored-by: Zaharid <zk261@cam.ac.uk>

scarlehoff · 2022-02-25T09:07:21Z

TODO:

Access the arrays directly in correlations.py comment
Check whether copy_grid in XplottingGrid can be made to only accept Stats without breaking anything comment
Either remove mask_replicas completely or make it part of the Stats class comment

As discussed, moving here the minor comments of the other PRs so that they can be merged

Zaharid · 2022-02-25T12:30:56Z

@scarlehoff do you want to merge the other PR as they are and keep working here, or close the others and only merge this one?

scarlehoff · 2022-02-25T12:39:08Z

@scarlehoff do you want to merge the other PR as they are and keep working here, or close the others and only merge this one?

Maybe merging them one by one makes doing the "undo" easier in case something is wrong obviously with any of them. But I don't think it makes a difference tbh

Zaharid · 2022-02-25T12:44:25Z

TBH I find looking at the diff here a bit confusing, so if we could merge the others and rebase possible further changes here, I'd be happier.

scarlehoff · 2022-02-25T12:46:15Z

if you approve the others I'll merge them

scarlehoff · 2022-03-02T11:48:10Z

I've implemented now the changes that were requested on the other two PRs.

n3fit/src/n3fit/vpinterface.py

RoyStegeman

This seems alright. The diffs are a real mess though..

n3fit/src/n3fit/vpinterface.py

n3fit/src/n3fit/io/writer.py

validphys2/src/validphys/sumrules.py

validphys2/src/validphys/core.py

validphys2/src/validphys/correlations.py

validphys2/src/validphys/sumrules.py

Co-authored-by: Roy Stegeman <roystegeman@live.nl>

scarlehoff · 2022-03-03T17:18:18Z

This seems alright. The diffs are a real mess though..

You are going to hate me so much when you're made to review #1529...

scarlehoff · 2022-03-03T20:12:54Z

I think I addressed everything.

RoyStegeman

Thanks. This looks good!

scarlehoff · 2022-03-08T08:41:48Z

@Zaharid are you happy merging this? (it's currently blocked by you)

Zaharid · 2022-03-09T13:00:00Z

validphys2/src/validphys/results.py

@@ -547,8 +547,7 @@ def phi_data(abs_chi2_data):
    1410.8849
    """
    alldata, central, npoints = abs_chi2_data
-    cv = float(alldata.central_value())
-    return (np.sqrt((cv - central) / npoints), npoints)
+    return (np.sqrt((alldata.error_members().mean() - central) / npoints), npoints)


Maybe we should make this and abs_chi2_data Stats as well, but that is another PR.

validphys2/src/validphys/sumrules.py

Zaharid

If Roy is happy I am happy.

Co-authored-by: Zaharid <zk261@cam.ac.uk>

scarlehoff force-pushed the mc_pdf_to_include_all branch from 21e4fea to 0321658 Compare February 15, 2022 15:39

scarlehoff added the run-fit-bot Starts fit bot from a PR. label Feb 15, 2022

scarlehoff added 4 commits February 16, 2022 15:43

make MC PDFs return all members (transitional commit for testing)

34340df

update regression test for sumrules and remove unnecesary if condition

907ae94

use directly central member in MCStats -and reupdate regressions-

178b8b5

make N3PDF closer to vp PDF (and add a N3LHAPDFSet)

c5425ff

scarlehoff force-pushed the mc_pdf_to_include_all branch from 80b8d4c to c5425ff Compare February 16, 2022 14:43

scarlehoff added run-fit-bot Starts fit bot from a PR. and removed run-fit-bot Starts fit bot from a PR. labels Feb 16, 2022

phi needs actually the mean of the error members

b0b8a0f

scarlehoff added run-fit-bot Starts fit bot from a PR. and removed run-fit-bot Starts fit bot from a PR. labels Feb 17, 2022

scarlehoff marked this pull request as ready for review February 17, 2022 09:01

scarlehoff removed the run-fit-bot Starts fit bot from a PR. label Feb 17, 2022

Zaharid requested changes Feb 17, 2022

View reviewed changes

n3fit/src/n3fit/vpinterface.py Show resolved Hide resolved

n3fit/src/n3fit/vpinterface.py Show resolved Hide resolved

n3fit/src/n3fit/vpinterface.py Outdated Show resolved Hide resolved

Update n3fit/src/n3fit/vpinterface.py

bca31e5

Co-authored-by: Zaharid <zk261@cam.ac.uk>

scarlehoff added the destroyingc++ label Feb 24, 2022

scarlehoff mentioned this pull request Feb 24, 2022

Utilize stats everywhere where it can be used #1515

Merged

Base automatically changed from utilize_error_members_instead_of_rawdata to master March 1, 2022 11:24

scarlehoff added 2 commits March 2, 2022 11:21

use the raw data for the correlations

01bd364

enforce grid_values being stats

7e28ee2

remove the special case of mask_replicas

7081d4d

Zaharid requested a review from RoyStegeman March 2, 2022 15:49

RoyStegeman reviewed Mar 3, 2022

View reviewed changes

n3fit/src/n3fit/vpinterface.py Outdated Show resolved Hide resolved

RoyStegeman reviewed Mar 3, 2022

View reviewed changes

scarlehoff commented Mar 3, 2022

View reviewed changes

validphys2/src/validphys/sumrules.py Outdated Show resolved Hide resolved

Apply suggestions from code review

08d347e

Co-authored-by: Roy Stegeman <roystegeman@live.nl>

scarlehoff added 2 commits March 3, 2022 20:53

remove grid_values_index

9199df9

add a warning for calls with different Q

2f0bc16

RoyStegeman approved these changes Mar 7, 2022

View reviewed changes

Zaharid reviewed Mar 9, 2022

View reviewed changes

validphys2/src/validphys/sumrules.py Outdated Show resolved Hide resolved

Zaharid reviewed Mar 9, 2022

View reviewed changes

Zaharid approved these changes Mar 9, 2022

View reviewed changes

Update validphys2/src/validphys/sumrules.py

bf14730

Co-authored-by: Zaharid <zk261@cam.ac.uk>

scarlehoff merged commit 53a145f into master Mar 9, 2022

scarlehoff deleted the mc_pdf_to_include_all branch March 9, 2022 14:11

Zaharid added the enhancement New feature or request label May 12, 2022

RoyStegeman mentioned this pull request Apr 5, 2024

fix std_error function in MCStats class #2036

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Return all members when using MC PDFs #1522

Return all members when using MC PDFs #1522

scarlehoff commented Feb 15, 2022

github-actions bot commented Feb 16, 2022

scarlehoff commented Feb 16, 2022

scarlehoff commented Feb 17, 2022 •

edited

Loading

Zaharid left a comment

scarlehoff commented Feb 25, 2022 •

edited

Loading

Zaharid commented Feb 25, 2022

scarlehoff commented Feb 25, 2022

Zaharid commented Feb 25, 2022

scarlehoff commented Feb 25, 2022

scarlehoff commented Mar 2, 2022

RoyStegeman left a comment

scarlehoff commented Mar 3, 2022

scarlehoff commented Mar 3, 2022

RoyStegeman left a comment

scarlehoff commented Mar 8, 2022

Zaharid Mar 9, 2022

Zaharid left a comment

Return all members when using MC PDFs #1522

Return all members when using MC PDFs #1522

Conversation

scarlehoff commented Feb 15, 2022

github-actions bot commented Feb 16, 2022

scarlehoff commented Feb 16, 2022

scarlehoff commented Feb 17, 2022 • edited Loading

Zaharid left a comment

Choose a reason for hiding this comment

scarlehoff commented Feb 25, 2022 • edited Loading

Zaharid commented Feb 25, 2022

scarlehoff commented Feb 25, 2022

Zaharid commented Feb 25, 2022

scarlehoff commented Feb 25, 2022

scarlehoff commented Mar 2, 2022

RoyStegeman left a comment

Choose a reason for hiding this comment

scarlehoff commented Mar 3, 2022

scarlehoff commented Mar 3, 2022

RoyStegeman left a comment

Choose a reason for hiding this comment

scarlehoff commented Mar 8, 2022

Zaharid Mar 9, 2022

Choose a reason for hiding this comment

Zaharid left a comment

Choose a reason for hiding this comment

scarlehoff commented Feb 17, 2022 •

edited

Loading

scarlehoff commented Feb 25, 2022 •

edited

Loading