Object-Oriented refactor

[smangham]

The branch issue-oo-refactor contains a sketch of refactoring the code to be object-oriented, following the outline diagram here:

I've done two of the types of ModelComponent (Lorentzian and DampedRandomWalk), and built the Model class that would collect the components.

The diagram, and an example "How I think you'd use the OO code" Python script, are in docs/.

[andresgur]

@smangham could you have a look at the code when you get a chance and see if we can close this one? Cheers!

[smangham]

Right, I thought I'd try to automatically generate an inheritance diagram, except that meant setting up Sphinx, which I've done...
Except of course most of that diagram is association not inheritance so I'm going to try and do it manually.

Thoughts from that so far: Simulator.py:E13_sim calls simulate_lightcurve, which as far as I can tell doesn't exist? But the tests all pass...

My IDE's also flagging gpmodelling.py:_build_mean_model:

amplitude_guess = (maxy - np.min(y)) * np.sqrt(2 * np.pi)* sigma_guess

Looks like y here doesn't exist?

[andresgur]

Thanks a lot for looking into this.
E13_sim is ancient code I need to get rid of it... can you ignore it for now? There's a bunch of methods there you'll see are not used anywhere.

You are right that's a typo, it should be
self._lightcurve.y

Can fix it from my side or wait for your PR?

[smangham]

Yeah, I've ignored it, it works fine otherwise. I might wait for the PR, as I've edited a fair few lines adding type hints, even if I haven't changed the actually code.

I think the only critique I have is that there's a real profusion of 'Model' objects and types, which is a little confusing. You have the mean_models which are celerite.modeling.Models, celerite_models which are celerite.terms.Terms, and psd_models which are astropy.modelling.Models. Then each different type of Model is used for a different purpose and as far as I can tell isn't exchangeable? And this is complicated by the fact you have multiple models with the same names, e.g. psd_models.Lorentzian and celerite_models.Lorentzian which makes imports pretty confusing.

I'd probably standardise the model naming, and probably include their application in them. (Disclaimer: I'm rushing and might be misremembering) E.g. renaming Lorenztian in celerite_models.py to FittingLorentzian, to make it more distinct from psd_models.py, where you renamed it SimulatedLorentzian?