`IndexingError` with `FixedCombinations` constraint during sample

[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

Sampling can sometimes crash with a FixedCombinations constraint due to a pandas IndexingError. This is triggered when we filter out sampled rows that are invalid for the constraint after reverse transforming (link).

The index of the is_valid Series is meant to match the sampled data index. For FixedCombinations however, we reset the index on the Series, which triggers the error in sample.

Steps to reproduce

In the below example, FixedNullCombinations drops a few invalid rows after reverse transforming. FixedCombinations then causes the error with its is_valid Series because the indices no longer align.

data, metadata = download_demo('single_table', 'expedia_hotel_logs')

constraints = [
    FixedCombinations(column_names=['hotel_continent', 'hotel_country'], table_name='expedia_hotel_logs'),
    FixedNullCombinations(column_names=['user_location_city', 'orig_destination_distance'], table_name='expedia_hotel_logs'),
]
synthesizer = GaussianCopulaSynthesizer(metadata)

synthesizer.add_constraints(constraint_list)
synthesizer.fit(data)
sample = synthesizer.sample(num_rows=len(data))