ValueError when sampling PII columns

[npatki]

Environment Details

• SDV version: 1.1.0

Description

For PII columns only, it should be ok if the input pandas.dtype is not the same as the output pandas.dtype. For example, the input data may be all 0'd out (since it's sensitive). But I expect the synthetic data data should have strings, based on the sdtype that I have selected in the metadata (and ultimately the Faker that is used).

Steps to reproduce

import pandas as pd
from sdv.metadata import SingleTableMetadata
from sdv.single_table import GaussianCopulaSynthesizer

data = pd.DataFrame(data={
    'id': [0, 1, 2, 3, 4],
    'address': [0, 0, 0, 0, 0],
    'numerical': [0.234234, 0.123213, 0.123123, 0.123123, 0.1345435]
})

metadata = SingleTableMetadata.load_from_dict({
    'primary_key': 'id',
    'columns': {
        'id': { 'sdtype': 'id' },
        'address': { 'sdtype': 'address' },
        'numerical': { 'sdtype': 'numerical' }
    }
})

synth = GaussianCopulaSynthesizer(metadata)
synth.fit(data)
synth.sample(10)

Output:

ValueError: Error: Sampling terminated. Partial results are stored in a temporary file: .sample.csv.temp. This file will be overridden the next time you sample. Please rename the file if you wish to save these results.
invalid literal for int() with base 10: 'PSC 4734, Box 5790\nAPO AA 75326'

Additional Context

For PII and ID columns only, we can do a try-catch whenever we try to cast the data back to the original dtype. If we cannot do the casting, then just return the data without casting. Log an INFO message when this happens.

logger.INFO The real data in 'column_name' was stored as 'int' but the synthetic data could not be cast back
to this type. If this is a problem, please check your input data and metadata settings.