DOC: demonstrate expression serialization

[mmikhasenko]

The first attempt to serialize symbolic model (save-model.ipynb) and reuse in a separate notebook (load-model.ipynb)

Currently, the model is saved as a pickle file that includes:

dict_forms = {
    "intensity_expr": unfolded_intensity_expr,
    "variables": {k: v.doit() for (k, v) in model.variables.items()},
    "parameter_defaults": model.parameter_defaults,
    "sigma3": sigma3via12(),
}

It would be useful to be able to retrieve the symbolic model with minimal dependencies.
Here are some things to discuss:

• what is a good balance between done/substituted expressions
• Is it possible to dump customary nodes, like Kibble, BW, ... - minimal library for lineshapes(?)
• Foresee the use with polarization

[redeboer]

Thanks for the PR!
I merged the notebooks and moved them to the appendix. The notebook is a bit like an experimental TR, but I think it's fine to leave it in the documentation repository (if only because it's more convenient with the model formulation).

• what is a good balance between done/substituted expressions

Probably best addressed through ComPWA/ampform-dpd#44

• Is it possible to dump customary nodes, like Kibble, BW, ... - minimal library for lineshapes(?)

I think pickle should already be able to handle it, see e.g. ComPWA/ampform#139. You will still need to call doit() after importing though.

• Foresee the use with polarization

It is indeed complicated of a long-term solution for this: "use" depends on what the requirements of the target application are. This also points to a larger issue: so far, this repository was more like an application (an analysis), not a general-purpose library. We want to generalise the code now, but we have to decide what general functionality this library should provide. See also #224.

[mmikhasenko]

@redeboer what is missing here for merging?

[redeboer]

We could merge it (it works), but the solution presented is copy-paste snippets, because there is no function that does the serialization for you. But we can also let this PR be limited to showing how to serialize so that it is at least in the documentation.

[mmikhasenko]

The current github page docs do not have the section on the model serialization. Is it a matter of rerunning the tox job in CI?