GaussianCopula get_learned_distributions crashes if nothing was learned

sdv/single_table/copulagan.py

@@ -246,9 +246,11 @@ def get_learned_distributions(self):
                 "Distributions have not been learned yet. Please fit your model first using 'fit'."
             )
 
-        field_transformers = self._gaussian_normalizer_hyper_transformer.field_transformers
-
         learned_distributions = {}
+        if not hasattr(self._gaussian_normalizer_hyper_transformer, 'field_transformers'):
+            return learned_distributions
+
+        field_transformers = self._gaussian_normalizer_hyper_transformer.field_transformers
         for column_name, transformer in field_transformers.items():
             if isinstance(transformer, rdt.transformers.GaussianNormalizer):
                 learned_params = deepcopy(transformer._univariate.to_dict())

sdv/single_table/copulas.py

@@ -216,6 +216,9 @@ def get_learned_distributions(self):
                 "Distributions have not been learned yet. Please fit your model first using 'fit'."
             )
 
+        if not hasattr(self._model, 'to_dict') or not self._model.to_dict():
+            return {}
+
         parameters = self._model.to_dict()
         columns = parameters['columns']
         univariates = deepcopy(parameters['univariates'])

tests/unit/single_table/test_copulagan.py

@@ -369,6 +369,29 @@ def test_get_learned_distributions(self):
             },
         }
 
+    def test_get_learned_distributions_nothing_learned(self):
+        """Test that ``get_learned_distributions`` returns an empty dict when nothing is learned."""
+        # Setup
+        metadata = Metadata().load_from_dict({
+            'tables': {
+                'table1': {
+                    'columns': {
+                        'col_1': {'sdtype': 'id'},
+                        'col_2': {'sdtype': 'credit_card_number'},
+                    },
+                }
+            }
+        })
+        data = pd.DataFrame({'col_1': range(100), 'col_2': range(100)})
+        synthesizer = CopulaGANSynthesizer(metadata, default_distribution='beta')
+        synthesizer.fit(data)
+
+        # Run
+        result = synthesizer.get_learned_distributions()
+
+        # Assert
+        assert result == {}
+
     def test_get_learned_distributions_raises_an_error(self):
         """Test that ``get_learned_distributions`` raises an error."""
         # Setup

tests/unit/single_table/test_copulas.py

@@ -568,6 +568,29 @@ def test_get_learned_distributions(self):
             'one': {'distribution': 'uniform', 'learned_parameters': {'loc': 1.0, 'scale': 0.0}},
         }
 
+    def test_get_learned_distributions_nothing_learned(self):
+        """Test that ``get_learned_distributions`` returns an empty dict when nothing is learned."""
+        # Setup
+        metadata = Metadata().load_from_dict({
+            'tables': {
+                'table1': {
+                    'columns': {
+                        'col_1': {'sdtype': 'id'},
+                        'col_2': {'sdtype': 'credit_card_number'},
+                    },
+                }
+            }
+        })
+        data = pd.DataFrame({'col_1': range(100), 'col_2': range(100)})
+        synthesizer = GaussianCopulaSynthesizer(metadata, default_distribution='beta')
+        synthesizer.fit(data)
+
+        # Run
+        result = synthesizer.get_learned_distributions()
+
+        # Assert
+        assert result == {}
+
     def test_get_learned_distributions_raises_an_error(self):
         """Test that ``get_learned_distributions`` returns a dict.