Always create a sample size of 1 no matter how small the scale.

sdv/sampling/hierarchical_sampler.py

@@ -1,5 +1,6 @@
 """Hierarchical Samplers."""
 import logging
+import warnings
 
 import pandas as pd
 
@@ -273,13 +274,24 @@ def _sample(self, scale=1.0):
         # DFS to sample roots and then their children
         non_root_parents = set(self.metadata._get_parent_map().keys())
         root_parents = set(self.metadata.tables.keys()) - non_root_parents
+        send_min_sample_warning = False
         for table in root_parents:
             num_rows = round(self._table_sizes[table] * scale)
+            if num_rows <= 0:
+                send_min_sample_warning = True
+                num_rows = 1
             synthesizer = self._table_synthesizers[table]
             LOGGER.info(f'Sampling {num_rows} rows from table {table}')
             sampled_data[table] = self._sample_rows(synthesizer, num_rows)
             self._sample_children(table_name=table, sampled_data=sampled_data, scale=scale)
 
+        if send_min_sample_warning:
+            warn_msg = (
+                "The 'scale' parameter is too small. Some tables may have 1 row."
+                ' For better quality data, please choose a larger scale.'
+            )
+            warnings.warn(warn_msg)
+
         added_relationships = set()
         for relationship in self.metadata.relationships:
             parent_name = relationship['parent_table_name']

tests/integration/multi_table/test_hma.py

@@ -1839,3 +1839,27 @@ def test_disjointed_tables():
     # Assert
     for table in real_data:
         assert list(real_data[table].columns) == list(disjoin_synthetic_data[table].columns)
+
+
+def test_small_sample():
+    """Test that the sample function still works with a small scale"""
+    # Setup
+    data, metadata = download_demo(
+        modality='multi_table',
+        dataset_name='fake_hotels'
+    )
+    synthesizer = HMASynthesizer(metadata)
+    synthesizer.fit(data)
+
+    # Run and Assert
+    warn_msg = re.escape(
+        "The 'scale' parameter is too small. Some tables may have 1 row."
+        ' For better quality data, please choose a larger scale.'
+    )
+    with pytest.warns(Warning, match=warn_msg):
+        synthetic_data = synthesizer.sample(scale=0.01)
+
+    assert (len(synthetic_data['hotels']) == 1)
+    assert (len(synthetic_data['guests']) >= len(data['guests']) * .01)
+    assert synthetic_data['hotels'].columns.tolist() == data['hotels'].columns.tolist()
+    assert synthetic_data['guests'].columns.tolist() == data['guests'].columns.tolist()

tests/unit/sampling/test_hierarchical_sampler.py

@@ -667,3 +667,33 @@ def test___enforce_table_size_clipping(self):
 
         # Assert
         assert data['parent']['__child__fk__num_rows'].to_list() == [2, 2, 4]
+
+    def test___enforce_table_size_too_small_sample(self):
+        """Test it enforces the sampled data to have the same size as the real data.
+
+        If the sample scale is too small ensure that the function doesn't error out.
+        """
+        # Setup
+        instance = MagicMock()
+        data = {
+            'parent': pd.DataFrame({
+                'fk': ['a', 'b', 'c'],
+                '__child__fk__num_rows': [1, 2, 3]
+            })
+        }
+        instance.metadata._get_foreign_keys.return_value = ['fk']
+        instance._min_child_rows = {'__child__fk__num_rows': 1}
+        instance._max_child_rows = {'__child__fk__num_rows': 3}
+        instance._table_sizes = {'child': 4}
+
+        # Run
+        BaseHierarchicalSampler._enforce_table_size(
+            instance,
+            'child',
+            'parent',
+            .001,
+            data
+        )
+
+        # Assert
+        assert data['parent']['__child__fk__num_rows'].to_list() == [0, 0, 0]