not use lambda method

Author: R-Palazzo

sdv/sequential/par.py

@@ -38,6 +38,11 @@
 LOGGER = logging.getLogger(__name__)
 
 
+def _diff_and_bfill(series):
+    """Compute the diff of a pandas Series and backfill the first NaN."""
+    return series.diff().bfill()
+
+
 class PARSynthesizer(LossValuesMixin, MissingModuleMixin, BaseSynthesizer):
     """Synthesizer for sequential data.
 
@@ -322,7 +327,7 @@ def _transform_sequence_index(self, data):
         else:
             diff_series = sequence_index.groupby(self._sequence_key, group_keys=False)[
                 self._sequence_index
-            ].transform(lambda s: s.diff().bfill())
+            ].transform(_diff_and_bfill)
 
         sequence_index_sequence = diff_series.to_frame(name=self._sequence_index)
         if all(sequence_index_sequence[self._sequence_index].isna()):

tests/unit/sequential/test_par.py

@@ -14,11 +14,24 @@
 from sdv.metadata.metadata import Metadata
 from sdv.metadata.single_table import SingleTableMetadata
 from sdv.sampling import Condition
-from sdv.sequential.par import PARSynthesizer
+from sdv.sequential.par import PARSynthesizer, _diff_and_bfill
 from sdv.single_table.base import BaseSynthesizer
 from sdv.single_table.copulas import GaussianCopulaSynthesizer
 
 
+def test__diff_and_bfill():
+    """Test the ``_diff_and_bfill`` method."""
+    # Setup
+    data = pd.Series([10, 15, 20, 30])
+
+    # Run
+    result = _diff_and_bfill(data)
+
+    # Assert
+    expected = pd.Series([5.0, 5.0, 5.0, 10.0])
+    pd.testing.assert_series_equal(result, expected)
+
+
 class TestPARSynthesizer:
     def get_metadata(self, add_sequence_key=True, add_sequence_index=False):
         metadata = Metadata()