Using FixedCombinations constraint with an integer constraint column causes sampling to fail

[srinify]

Environment Details

• SDV version: 1.15.0 (Latest)

Problem Description

When using FixedCombinations, if the underlying dtype of a constraint column is int, then an error is thrown during sampling.

Error Description

Error seems to start with this line of code: https://github.com/sdv-dev/SDV/blob/9301b964504ec53df977f1db5eab28b5b2e2c352/sdv/data_processing/data_processor.py

Steps to reproduce

Dataset: stock_missingcol.csv

Code:

from datetime import datetime

import pandas as pd
from sdv.evaluation.single_table import evaluate_quality, run_diagnostic
from sdv.metadata import SingleTableMetadata
from sdv.single_table import CTGANSynthesizer

constrains_column = ['Quantity', 'Total Price']
real_data = pd.read_csv('stock_missingcol.csv')
real_data.head()
metadata = SingleTableMetadata()
metadata.detect_from_dataframe(real_data)
metadata.update_columns(column_names=constrains_column, sdtype='categorical')

synthesizer = CTGANSynthesizer(metadata,epochs=2,verbose=True)
my_constraint = {
    'constraint_class': 'FixedCombinations',
    'constraint_parameters': {
        'column_names': constrains_column
    }
}

synthesizer.add_constraints(constraints=[
    my_constraint
])

synthesizer.fit(real_data)
synthetic_data = synthesizer.sample(num_rows=1000)

Internal Colab Notebook: https://colab.research.google.com/drive/1XUL42Wa13NQ2t0qewCoyCDYHHxmRO6ku?authuser=1#scrollTo=UGo3e-QJFk0a

Workaround

I was able to avoid this error by casting all int columns to float, then re-fitting, then sampling, then finally casting back to int in the sampled data.

# In real data, cast from int -> float
real_data['Total Price'] = real_data['Total Price'].astype(float)

# In synthetic data, cast from float -> int
synthetic_data['Total Price'] = synthetic_data['Total Price'].astype(int)