Python 3

[nsmith-]

As of CMSSW 11_2, ROOT v6.22 is compiled against both python2 and python3. Are there plans to start making Combine python3-compatible?

[nsmith-]

It wasn't too much trouble to get the C++ side compiling in 11_2 (see #648 ) so at least python3 users can now use the combine-specific ROOT classes while building their workspace and datacards (my main motivation)

[nucleosynthesis]

I remember that Jonas (https://github.com/guitargeek/) had a fork at one point in which the python was migrated to py3 (https://hypernews.cern.ch/HyperNews/CMS/get/higgs-combination/1322.html). I suspect that we could automate quite a lot of the main tools (t2w, python models ...) and then spend some time making sure the various python tools/scripts in test are updated.

[guitargeek]

Hi! The fork doesn't exist anymore, because I put what was special about it in my repository with build scripts for Arch Linux:
https://github.com/guitargeek/PKGBUILDs/tree/master/cms-higgs-combine

In the PKGBUILD file you can see what I did for the python 3 migration:

• format code with autopep8 -i -r . such that python-modernize doesn't hang because of tab/space mixing
• code formatting with python-modernize -w -n .

Maybe that's helpful!

By the way, since February I'm not in CMS anymore but in the ROOT team, working as the main RooFit developer. If you have any suggestions on how we could improve RooFit upstream to benefit CMS, please don't hesitate to get in touch with me!

I am already using some Higgs combine fit examples to benchmark and optimize RooFit. So for example, you could let me know about some fits that are much slower than they need to be and we could find ways to make them faster. Another thing that might be interesting could be to bring the speed optimizations you made for some pdfs in combine also to RooFit, or integrate some shapes that are widely used in CMS but are missing upstream.

[nsmith-]

Pasting here open PRs that touch python code:

• Anomalous couplings histograms merged #644 (likely new physics model only)
• Anomalous top-Higgs coupling model #366 (new physics model only, not required)

[kcormi]

Closing as resolved by #759 . @nsmith- is it worth opening a separate issue for the dictionary ordering (which I think was never addressed?). I guess it shouldn't have any effect, but perhaps there are cases where being able to reproduce the ordering is helpful?

[nsmith-]

Dictionary ordering may in principle affect results via numerical precision and instability, but hopefully only in rare instances. Indeed #759 (comment) is not directly addressed at the moment. If we are OK not having consistent behavior between combine v8/10x/py2 and combine v9/py3, then at least going forward things should stay rather consistent due to stable dictionary ordering in py3.7+.