pyMaCh3 on pyip via CD

[KSkwarczynski]

Now that we have pyMaCh3 it would be good as part of CD push it on pyip to have it always up to date.
It should be added here
https://github.com/mach3-software/MaCh3/blob/develop/.github/workflows/CDImage.yml

Quoting Patrick:

It does look like you need an account. It also unfortunately looks like an individual has to have the account at the moment. they’re developing organisation level accounts but it’s a closed beta

We can wait for the time being

[github-actions]

Stale issue message

[jonas-eschle]

It does look like you need an account. It also unfortunately looks like an individual has to have the account at the moment. they’re developing organisation level accounts but it’s a closed beta

We can wait for the time being

What about creating an account ("MaCh3") that is shared and transfer once orgas are widely available? It took us at Scikit-HEP about.... 2 years or something (I think they're checking manually and have been completely overwhelmed with requests (before, we've just shared a "Scikit-HEP" named account). Having mach3 pip-installable would strongly simplify things, +1 to add this to CD

[pjdunne]

It does look like you need an account. It also unfortunately looks like an individual has to have the account at the moment. they’re developing organisation level accounts but it’s a closed beta

We can wait for the time being

What about creating an account ("MaCh3") that is shared and transfer once orgas are widely available? It took us at Scikit-HEP about.... 2 years or something (I think they're checking manually and have been completely overwhelmed with requests (before, we've just shared a "Scikit-HEP" named account). Having mach3 pip-installable would strongly simplify things, +1 to add this to CD

Thank you that's very helpful feedback. The MaCh3 python is still in a somewhat 'beta' testing phase where we're not ready to deploy it more widely but we'll definitely consider that approach when it gets to roll out.

If you don't mind me asking @jonas-eschle it would be useful to hear if you have a use case for MaCh3 that you'd like us to help with.

[jonas-eschle]

That makes sense... it could stil be a nice option to make also sure that the pipeline works, the namespace is reserved, beta users could test it etc. But I understand, it's also extra effort, and it can just be built locally (but building is sometimes a pain).

Sure, basically I am new to the field and experiment and more on the technical side of things, coming from the large LHC experiments with a strong fitting and Python background. So I was mostly interested to checkout the tools used here.
What I was mostly interested in is to see the adaptation to other tools within Python (something we strongly pursued in Scikit-HEP or zfit etc). Namely the inference stage, i.e. the idea is to build a likelihood with pymach3 (which makes sense to have this domain specific) but to actually perform the inference (be it frequentist with minimizers, sampling for bayesian) using different libraries provided in the Python ecosystem. And of course to do plots etc.

If the likelihood has a nice interface, that is, a callabel which takes parameters, that's in the end all that is needed.

An overarching goal would also be to study likelihood serialization of profiles and more. Having a Python interface to the likelihood would hence greatly simplify things.

Btw, if you're interested, I am also glad to learn more about how fitting is done in the neutrino community, all the tricks you know and do, and I am open for a chat, (feel free to write an e-mail)

[EdAtkin]

Hi Jonas, thanks for getting in touch! So I'm chiming in because I've recently been playing around with using an external python library to do some ML-based stuff using MaCh3 so I updated the python interface a little and made a simple wrapper class in python to be able to grab different parts from MaCh3. If you're interested we could have a bit more of a chat about this 😄

Having gone through this experience though I would say the python build we have at the moment needs an overhaul which I will try to present at a MaCh3 meeting over the coming months. I think in the end the python bindings from the core classes will exist in the core but each experiment would then include these and make their own python library. But as I say I'm happy to chat about this more and I'll try to present it more broadly at a MaCh3 meeting in the near future.

[jonas-eschle]

That sounds good, up for a chat! Maybe we can take this offline? Can you drop me an e-mail?