Using Preferences

[sethaxen]

This is a meta-issue for all Julia component packages. Python ArviZ uses a matplotlib-like rcParams object to manage global configurable options. For projects with many analyses, this allows the user to set a preference once (e.g. default CI type or probability) and use it everywhere without needing to always include it in the keywords.

I think the right way to do this in Julia would be to use Preferences.jl and define a LocalPreferences.toml giving the preferences. unlike Python ArviZ, I think the user would need to set any changed preferences in their own LocalPreferences.toml before launching Julia. The question then is whether we should define an ArviZPreferences.jl package that only exists to set preferences for all of our packages (and would need to be a dependency for all our packages( or whether we should have each package define its own LocalPreferences.toml.

The advantage of the first option is a single place where all preferences are set, and we can easily print out the file in the docs. The disadvantage is that if we want to add a new preferences option for a given package, we need to first modify this dependency.

The advantage of the second option is that the preferences are defined where they are needed. The disadvantage is that the preferences are clustered by package name. Since ArviZ is a meta-package, this requires users to know in which package which parameter is defined. Not really a problem if we have a way to print out in the docs the preferences for all component packages.