search

NNPDF / nnpdf

An open-source machine learning framework for global analyses of parton distributions.
https://docs.nnpdf.science/
GNU General Public License v3.0
30 stars 6 forks source link

Additional sources of theory uncertainty #285

Closed nhartland closed 8 months ago

nhartland commented 6 years ago

We have a plan for how new theory covariance matrices should be generated and communicated in Issue #186. However in that issue the solution to point (3) remains unspecified. This issue is to discuss how it is to be solved. To recap:

Point 3 : How do we specify which theory errors we want to add to which matrices. We want to deprecate the existing system for fudge-factors and C-factor errors (sys10) and have them included in the theory matrices by vp-setupfit. How the construction of the matrices is specified is undecided.

It should: Express which theory errors (MHOU, C-factor, fudge) should be applied to which matrices (fitting, sampling). This should be driven by the fit runcard.

At the code p.c today it was decided that we need to be able to handle, very generally, both correlated and uncorrelated uncertainties (possible also multiplicative/additive) from external sources (i.e a plain text file). This requires a new scheme for the specification of these uncertainties.

One possibility is a CommonData type file format with CORR UNCORR and named systematics for a completely general treatment (CommonTheory?).

The strategy would be that validphys reads these files and builds a theory covariance matrix from them. This can then be used to augment the theory matrix from scale variations.

The structure of such a file should be discussed here, along with the control/configuration apparatus required in order to manage all the different possible combinations of theory errors.

Zaharid commented 6 years ago

This could be used to integrate #269. We could use some systypes-like mechanism to encode the defaults for Zpt and jets, with some mechanism to select the "right" configuration based on the other settings, in such a way that we can have backwards compatibility.

Zaharid commented 6 years ago

@scarrazza How should we proceed with this (and more to the point, who)?

scarrazza commented 6 years ago

I would suggest to first write down a file layout proposal for a CommonTheory, and for sure this issue will require much more discussion. Lets take some days to think about this idea and then find an agreement.

scarlehoff commented 8 months ago

This ship has sailed (and probably sinked just out of the harbour)