Don't modify my metadata object

[npatki]

Environment Details

• SDV version: 0.14.0 (and previous)

Description

Right now, if I pass in a metadata.Table object into a model and fit it, the process will modify my metadata (add extra fields to it).

This can lead to unexpected behavior due to the extra fields that were added, especially if I want to re-use the metadata object in a different model.

Can we make a deep copy of the metadata.Table object instead?

Steps to reproduce

See the code below for unexpected usage

from sdv.tabular import GaussianCopula
from sdv.demo import load_tabular_demo

metadata, data = load_tabular_demo('student_placements', metadata=True)

# create a model that estimates everything as a gamma distribution
model = GaussianCopula(table_metadata=metadata,
                       categorical_transformer='label_encoding',
                       default_distribution='gamma')
model.fit(data)
print(model.get_distributions()) # each distribution correctly is a Gamma

# create a new model that estimates everything as a beta
model2 = GaussianCopula(table_metadata=metadata,
                        categorical_transformer='label_encoding',
                        default_distribution='beta')

model2.fit(data)
print(model2.get_distributions()) # Everything is still gamma. This is unexpected!

This is happening because, unbeknownst to me, my metadata is modified to include a 'gamma' distribution field for each and every column. I am unknowingly passing it into model2, which is favoring the metadata over the default_distribution parameter.