DecayLanguage compatibility

[simonthor]

Solves scikit-hep/decaylanguage#86. See this issue for further information.

The code works and all tests pass on Linux.
Tasks left to do:

• Add a way of installing the required packages for the fulldecay part (e.g., make it possible to run pip install phasespace[fulldecay])
• Rename fulldecay to something else?
• Add example use cases and documentation. Where should this be put?

[eduardo-rodrigues]

Great to see this PR! Can you explicitly link to DecayLanguage's task scikit-hep/decaylanguage#86 in the description header since this is addressing it?

[jonas-eschle]

Very nice and thanks a lot! I agree with Eduardos comments, additional thoughts

• name, what about "fromdec" or "fromdecay"?
• the docstyle in phasespace is the google one, not the numpy (see other docs)

[jonas-eschle]

For the docs, I would suggest to create a jupyter notebook with a good example, this can then be converted into a static html page (such as done in zfit for example).

I suggest to create a notebook (in docs is fine) that explains and introduces the feature

[eduardo-rodrigues]

I think this is ready for a first version 👍.

[jonas-eschle]

Yep, I agree! Just figuring out how to successfully also run the notebook, there is some weird issue with graphviz

[eduardo-rodrigues]

Yep, I agree! Just figuring out how to successfully also run the notebook, there is some weird issue with graphviz

Ah, if it is for conda then you need python-graphviz, see https://github.com/conda-forge/decaylanguage-feedstock/blob/master/recipe/meta.yaml#L26.

[jonas-eschle]

So I didn't figure out the CI failures yet fully, but it is not related to the PR directly but rather that some software packages play weird together. I am still checking it

[eduardo-rodrigues]

So I didn't figure out the CI failures yet fully, but it is not related to the PR directly but rather that some software packages play weird together. I am still checking it

Sorry to hear about the pain. For Python 3.6 I had a speedy look just now and see
E ImportError: This version of TensorFlow Probability requires TensorFlow version >= 2.7; Detected an installation of version 2.6.2. Please upgrade TensorFlow to proceed.
So that's either a trivial update or a dependency nightmare. In any case, given the scope of this package it might not be a problem to any user if you drop 3.6 support at this stage ;-).

[jonas-eschle]

Yeah, Python 3.6 makes sense, thanks! That is a more trivial one (also, we should soon remove support for it I think), the others are just weird... and I can't reproduce it locally. Still trying, and maybe there is also gonna be a magical fix soon somewhere :D

[eduardo-rodrigues]

This is really annoying. Looking online for the error I got to tensorflow/tensorflow#51592. Does it help you?

[jonas-eschle]

Yeaish, that's what I suspect, so I hope it may gets fixed. But then I cannot reproduce that locally, which is rather weird. Version mismatches should not really happen, or if, then it's maybe just because a new version was released newly and now packages react.

I'll give it a week or so more, and if that doesn't resolve it (TF2.7 was just resently released), I'll dig more into it.

Thanks though for the help!

[jonas-eschle]

Since we dropped Python 3.6, this is now surely a go! Many thanks @simonthor and @eduardo-rodrigues for getting this in!

[simonthor]

Thank you both as well for the guidance throughout this process!