Open
Description
ArviZ should store metadata at the variable level in order to take advantage of it and automate things better.
Context
We should use this issue to discuss what metadata is relevant and once we have a clear idea of what would be interesting to support, start thinking about implementations to make that a reality. For now let's refrain from discussing implementation details as much as possible.
This is relevant to inferencedata, if we want data and analysis to be decoupled we need to make sure if metadata is used, it is added from any language in the same way. Should we move this to arviz_governance or something of the sort? cc @sethaxen
Ideas
- Variable type
- continuous
- discrete
- circular
- time?
- categorical
- boolean
- imaginary/complex
- variable constraints
- bounds, which could be:
- constant
- vectorized (in multidimensional variables only)
- dependent on other variables?
- bounds, which could be:
should transforms also be considered metadata?