Add support for table columns with Json type

[Uzzije]

Problem Description

Currently attempting to train a single or multi-table structure throws an error if there exists a column with JSON value.

Expected behavior

SDV model.fit() should be able to properly parse tables where columns have Json types

[npatki]

Hi @Uzzije,

Thanks for filing the feature request! Are you able to give an example of what the JSON looks like and what the ideal modeling result would be?

For example, if your JSON looks something like this:

{
  "key_1": "value_1",
  "key_2": 2,
  "key_3": { "key_4": "value_4" }
}

Would you expect each key to be modeled as an additional column? What about nested values?

[Uzzije]

Hi Npatki, Yes, Ideally keys would be model to represent the same level of nested ness as training data. The expectation is that the json data for each row are consistently structured though. So there will not be inconsistency in the nested nature of the data per row in table.