ADR suggestion: ParameterAnalysis

[henrikjacobsenfys]

Name is also work in progress.

General

When users have fitted their raw data (using Analysis or the not-yet-existing Job (or Project?)) they may want to fit the resulting Parameters to figure out trends and to obtain parameters that describe the Q or temperature dependence of their parameters. For now, we will only be concerned with the Q dependence. There are two main cases to distinguish: Parameters describing diffusion, and Parameters that don't describe diffusion. I'm unsure if they should be treated by different classes.

For diffusion, users will want to fit the area and width of a particular Lorentzian to the corresponding area and width calculated by a DiffusionModel. Some users will want to only fit the width.

For other cases, most users will want to fit a single Parameter to a function described by a ComponentCollection or just a ModelComponent (most likely a Polynomial). Some advanced users may want to fit multiple Parameters to a combined model.

I propose to create a new class, ParameterAnalysis for this.

Defining the data

It should be easy for users to pass their Parameters. The standard method will be to pass the Analysis (or Job in the future) object and the name of either the Lorentzian if fitting a DiffusionModel or the name of the Parameter (or a list of names for multiple Parameters).
It will also be allowed to pass a scipp.Dataset, which is obtained from Analysis.parameters_to_dataset(), or a scipp.DataArray with no name arg - the scipp.DataArray is then obtained from the scipp.Dataset by selecting the name manually.

Defining the fit function

There are many use cases, and so it may be a good idea to split the functionality into different subclasses.
The fit function can be a DiffusionModel, in which case users also need to be able to set a fit_area and fit_width boolean. Internally, it will use the DiffusionModel.calculate_width() and DiffusionModel.calculate_area() methods.

For a single Parameter, the fit function should be either a ComponentCollection or ModelComponent, in which case the .evaluate() method is used.

For multiple Parameters, the fit function will be a list (each element corresponds to a Parameter) of either ComponentCollections or, ModelComponents, or a mix. Either way, the evaluate method is used.

All the fitting will be done with either Fitter or MultiFitter.

Methods

The data and fit functions should be properties with setters and getters.

Users should be allowed to change fit settings.

Methods are needed to plot the results.

Future extensions

Care needs to be taken when temperature is also involved - people will want to sometimes fit as function of temperature instead of Q and sometimes fit the same model to all temperatures (analogous to how people want to fit the same model to all Q instead of just one Q). I will cross this bridge when I get to it.