Start filter implementation in python

[scarrazza]

Issue by scarrazza
Thursday Nov 23, 2017 at 21:41 GMT
Originally opened as https://github.com/NNPDF/libnnpdf/issues/47

---

• define format
• produce prototype

[scarrazza]

Comment by Zaharid
Friday Nov 24, 2017 at 10:06 GMT

---

The high level operation looks like:

• We have one YAML file specifying the default cut rules. This is equivalent to the current kincuts.cc but with rules that can in fact be read by humans.

• The cut config file contains a set of parameters controlling the cuts, such as the min Q² cut for DIS, and a set of rules. Examples for valid rules look like those in https://github.com/NNPDF/nnpdfcpp/issues/26.

• The cut config file is not supposed to be edited frequently. To first approximation only for bugfixes, or fundamental changes (like we figure out some kinematic limit we include is unreliable).

• The parameters in the cut config file can be overridden by the fit config file. The themselves cannot.

• The resolved input for the rules, as well as the resolved output (i.e. the masks) are both stored in the fit result folder for the purposes of reproducibility. Tools (i.e. vp2) know how to process both the default and the per-fit output.

• In principle a data-theory comparison plot with cuts requires only the theoryID and the dataset specification. This should be enough to resolve the cuts, given the default cut file.

[scarrazza]

From the discussion today we have decided to start this project asap so we can integrate the theory covariance matrix generation without jumping between different languages and programs.

The idea is to start from a simple translation, without revolutionizing cuts in a first stage, then proceed with the improvements.

I would suggest to call this program prefit instead of filter.

[Zaharid]

FWIW I like setup-fit. Anyhow, this should allow calling filter (the thing that does the cuts as opposed to the setup code) as a validphys provider, for which it should be at least in its own library. From there we can change how it works soon-ish. I actually thing that writing a parser like I have in mind would not be very hard once the scafolding is in place.

[voisey]

This has been dealt with, right?

[siranipour]

Yeah, think we can close