HMASynthesizer errors out when fitting a dataset that has a table which holds primary key and foreign keys only

[pvk-developer]

Environment Details

• SDV version: 1.0.0.dev
• Python version: 3.8
• Operating System: MacOS

Error Description

The HMASynthesizer fails to fit a dataset if it has a table that is contains only a primary key and foreign keys. This error occurs because all the columns have been dropped for the given table therefore it can't fit the model.

File ~/Projects/sdv-dev/SDV/sdv/multi_table/hma.py:188, in HMASynthesizer._model_table(self, table_name, tables)
    184 self._clear_nans(table)
    185 LOGGER.info('Fitting %s for table %s; shape: %s', self._synthesizer.__name__,
    186             table_name, table.shape)
--> 188 self._table_synthesizers[table_name].fit_processed_data(table)
    190 for name, values in keys.items():
    191     table[name] = values

File ~/Projects/sdv-dev/SDV/sdv/single_table/base.py:421, in BaseSynthesizer.fit_processed_data(self, processed_data)
    414 def fit_processed_data(self, processed_data):
    415     """Fit this model to the transformed data.
    416 
    417     Args:
    418         processed_data (pandas.DataFrame):
    419             The transformed data used to fit the model to.
    420     """
--> 421     self._fit(processed_data)
    422     self._fitted = True
    423     self._fitted_date = datetime.datetime.today().strftime('%Y-%m-%d')

File ~/Projects/sdv-dev/SDV/sdv/single_table/copulas.py:132, in GaussianCopulaSynthesizer._fit(self, processed_data)
    130 with warnings.catch_warnings():
    131     warnings.filterwarnings('ignore', module='scipy')
--> 132     self._model.fit(processed_data)

....

ValueError: need at least one array to concatenate

This is what the Metadata structure looks like:

As seen in the image the error occurs on the gmember table.

Steps to reproduce

from sdv.multi_table import HMASynthesizer
from sdv.datasets.demo import download_demo, get_available_demos

data, metadata = download_demo('multi_table', 'Biodegradability_v1')

hmas = HMASynthesizer(metadata)
hmas.fit(data)