Config Class (Dataclass?)

[dougbrn]

It may be a good idea to introduce an ensemble-level config class. There are a small number of configurable aspects to TAPE, users being able to view/manage these via a config class would make these more approachable. I also think this would make saving an ensemble easier, as the config can be stored and reloaded, and would aid in reproducibility to easily know the configuration settings that were used in a given workflow.

Implementation-wise, I'm thinking something as simple as a dictionary, but could also be a dataframe.

Edit: More details from slack

I was thinking of it as a central place for some of the various flags and settings we have going in TAPE right now, a few examples:

• keep_empty_objects is a current “flag” (settable as an init kwarg) that lets users keep objects that have no sources on syncs
• The dirty flags could be nice to keep in a config file when we implement saving and loading ensembles If we end up implementing more sync modes/options, those would also be good to track
• Any client/port information would be good to store as well for the current session
• Even the current ColumnMapper could be stored on-hand

So mainly just giving settings that have the ability to influence how the ensemble is behaving a more central place for inspection/reproducibility.

[dougbrn]

Additionally, this could hold unit information if available. We're still trying to understand the scope of how TAPE should support units (is it just for reference, minimal usage in unit conversion functions, or full blown unit tracking in all TAPE operations?). The config class would offer the ability to store these by column name, giving us a way to inherit unit information and metadata as it's provided by hipscat, and the ability to save that metadata to file when ensemble saving is implemented so that information is not lost between sessions.