Sampled data with foreign key that points to more than one primary key results in InvalidDataError

[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

Given metadata where a foreign key in a child table points to two primary keys in distinct tables, the resulting sampled data fails metadata validation. This should be considered invalid metadata (a foreign key should not be able to point to two separate primary keys), but I can successfully fit and and sample. However, when validating the sampled data with the metadata, I get an InvalidDataError because the sampled child contains entirely unknown references to one of the parents.

Suggested fix

To prevent this issue, we should update metadata validation to disallow multiple relationships between a single foreign key and multiple primary keys.

Steps to reproduce

from sdv.datasets.demo import download_demo
from sdv.metadata import Metadata
from sdv.multi_table import HSASynthesizer


data, _ = download_demo('multi_table', 'Student_loan_v1')
metadata = Metadata.detect_from_dataframes(data)
# The column 'name' in the table 'enlist' now points to 3 primary keys across different tables
metadata.validate_data(data)

synthesizer = HSASynthesizer(metadata)
synthesizer.fit(data)
sampled = synthesizer.sample()

metadata.validate_data(sampled)