FixedIncrements constraint cannot be applied conjunction with Inequality constraint

[frances-h]

Environment Details

Please indicate the following details about the environment in which you found the bug:

• SDV version:
• Python version:
• Operating System:

Error Description

Overlapping other constraints with the FixedIncrements constraint may cause errors during sampling. This was discovered when trying to test the ChainedInequality constraint.

The issue is that the FixedIncrements constraint will directly modify the constraint column, which may make downstream constraints invalid. Data validation is only done on the original data and not on the intermediate input data to the constraint, so this is not being caught during fit. In the below example, null/-inf values are being created by the Inequality constraint during preprocess. During sample, trying to force null/inf values back to ints is what is causing the error.

Applying Inequality before FixedIncrements works as expected.

Steps to reproduce

import pandas as pd
import numpy as np

from sdv.metadata import Metadata
from sdv.single_table import GaussianCopulaSynthesizer

data = pd.DataFrame(data={
    'id': [0, 1, 2, 3, 4],
    'low_value': [1, 3, 5, 4, 3],
    'high_value': [5, 5, 10, 15, 5],
})

metadata = Metadata.load_from_dict({
    'tables': {
        'table': {
            'primary_key': 'id',
            'columns': {
                'id': { 'sdtype': 'id' },
                'low_value': { 'sdtype': 'numerical' },
                'high_value': { 'sdtype': 'numerical' },
            }
        }
    }
})

fixed_increments = {
    'constraint_class': 'FixedIncrements',
    'constraint_parameters': {
        'column_name': 'high_value',
        'increment_value': 5
    }
}

inequality = {
    'constraint_class': 'Inequality',
    'constraint_parameters': {
        'high_column_name': 'high_value',
        'low_column_name': 'low_value'
    }
}

synthesizer = GaussianCopulaSynthesizer(metadata)
synthesizer.add_constraints([fixed_increments, inequality])
synthesizer.sample(num_rows=10)

I would expect the sampling here to work. Although there is an overlapping column, the Inequality constraint can fall back to using reject sampling. Instead, I see that there is an error during sampling:

IntCastingNaNError: Error: Sampling terminated. No results were saved due to unspecified "output_file_path".