Capturing training dataset stats when models are added?

[nlathia]

A trained machine learning model is inherently tied to the dataset that it was trained with

The modelstore library already has a def _model_data(self, **kwargs) -> dict function that enables users to pass in their training set when uploading a model, and some stats are automatically pulled out of it (example here). But this feature is not implemented across all model types.

• Should this be kept in modelstore?
• If yes, what summary stats would be useful to pull out here?