Inequality constraint raises RuntimeWarning (invalid value encountered in log)

[npatki]

Environment Details

• SDV version: 1.0.0b0 (Beta version)
• Python version: 3.8
• Operating System: Linux (Colab Notebook)

Error Description

When I apply an Inequality constraint to the demo data, I see a RuntimeWarning when fitting my data.

/usr/local/lib/python3.8/dist-packages/sdv/constraints/tabular.py:451: RuntimeWarning:

invalid value encountered in log

This doesn't seem to affect the data quality so I'm ignoring it. But we should check to see if it's indicative of any unintended behavior.

Steps to reproduce

from sdv.datasets.demo import download_demo
from sdv.single_table import GaussianCopulaSynthesizer

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

checkin_lessthan_checkout = {
    'constraint_class': 'Inequality',
    'constraint_parameters': {
        'low_column_name': 'checkin_date',
        'high_column_name': 'checkout_date'
    }
}

synthesizer = GaussianCopulaSynthesizer(metadata)
synthesizer.add_constraints([
    checkin_lessthan_checkout
])
synthesizer.fit(real_data)

Context

The following line is causing the error in constraints/tabular.py:

table_data[self._diff_column_name] = np.log(diff_column + 1)

It seems like np.log throws this warning if any value is 0 or negative. But this is odd: All values should be 1 or more because:

• We are ensuring that the constraint is True in the real data (so each value in diff_column must be >=0 )
• We are adding 1 to each item, so each value must be >= 1