Breaking the pyqg dependency

[jdenholm]

The utils.py currently depends on the pyqg repository, but Laure says she would like this equation discovery repository to be completely independent of pyqg .

• Is dropping the pyqg dependency definitely what we want to achieve?
• How can we refactor/redesign this repository such that it is standalone with no pyqg dependency?

[LaureZanna]

Yes, it would be great to removepyqg dependency so that users can feed in their data without having to install pyqg which is a numerical solver.

[asross]

My take on this is that we should support pyqg and require it for certain things (e.g. computing similarity metrics that are specifically designed for pyqg.Models), but we shouldn't require it for the entire library, since much of the code (e.g. the actual hybrid symbolic procedure) can operate just as easily on an xarray.Dataset.

To accomplish that, I think we'll need to do something like:

• try importing pyqg and store a boolean indicating whether the import succeeded
• replace existing type checks for pyqg.Model with something like is_pyqg_model that returns isinstance(pyqg.Model) if the import succeeded and False if the import failed

Does that make sense? I think that's a better approach than removing all references to pyqg altogether, since then it will become very difficult to run the library with pyqg models.

[LaureZanna]

Thanks @asross; i like the idea. I feel as long as one can use the hybrid symbolic procedure without pyqg that would be really great (including calculating the derivatives).