Unable to turn off rounding scheme for a column (need a warning)

[npatki]

Environment Details

• SDV version: 1.17.10 (latest)
• Python version: 3.10
• Operating System: Linux (Colab Notebook)

Error Description

I am explicitly trying to turn off rounding for a column, but if I set up FloatFormatter with learn_rounding_scheme=False, I see that this doesn't have any effect.

Steps to reproduce

from sdv.datasets.demo import download_demo
from sdv.single_table import GaussianCopulaSynthesizer
from rdt.transformers.numerical import FloatFormatter

data, metadata = download_demo(
    modality='single_table',
    dataset_name='fake_hotel_guests'
)

synthesizer = GaussianCopulaSynthesizer(metadata)
synthesizer.auto_assign_transformers(data)

synthesizer.update_transformers({
    'amenities_fee': FloatFormatter(learn_rounding_scheme=False) # I'm trying to turn off rounding for this column
})

synthesizer.fit(data)
synthesizer.sample(5)

Expected Results

The code itself is actually working as intended -- because the rounding is being enforced at a synthesizer level, it's not possible to override it by just updating a transformer. In this case, I expect a warning should be thrown after update_transformers.

Warning: Unable to turn off rounding scheme for column 'amenities_fee', because the overall synthesizer is
enforcing rounding. We recommend setting the synthesizer's 'enforce_rounding' parameter to False.

Note that this warning should only appear if:

• The synthesizer's enforce_rounding is set to True (this is also the default) AND
• The user is providing a transformer with learn_rounding_scheme=False (they are trying to turn it off)

Additional Context

Note that the behavior is fine the other way around -- i.e. if the synthesizer's enforce_rounding=False, I am still able to turn on rounding for individual columns with using the FloatFormatter. So in this particular case, I could do this instead:

# at synthesizer level, do not enforce rounding for any of the columns
synthesizer = GaussianCopulaSynthesizer(metadata, enforce_rounding=False) 

# now, you can selectively turn rounding back ON for any columns you need
synthesizer.update_transformers({
    'room_rate': FloatFormatter(learn_rounding_scheme=True)
})

synthesizer.fit(data)
synthesizer.sample(5)