Allow default transformers to have specific arguments

Author: frances-h

Diff for: sdv/data_processing/data_processor.py

@@ -1,5 +1,6 @@
 """Single table data processing."""
 
+import inspect
 import json
 import logging
 import warnings
@@ -494,6 +495,15 @@ def create_regex_generator(self, column_name, sdtype, column_metadata, is_numeri
 
         return transformer
 
+    @staticmethod
+    def _get_transformer_kwargs(transformer):
+        args = inspect.getfullargspec(transformer.__init__).args[1:]
+        return {
+            key: getattr(transformer, key)
+            for key in args
+            if key != 'model_missing_values' and hasattr(transformer, key)
+        }
+
     def _get_transformer_instance(self, sdtype, column_metadata):
         transformer = self._transformers_by_sdtype[sdtype]
         if isinstance(transformer, AnonymizedFaker):
@@ -512,7 +522,8 @@ def _get_transformer_instance(self, sdtype, column_metadata):
 
         if kwargs and transformer is not None:
             transformer_class = transformer.__class__
-            return transformer_class(**kwargs)
+            default_transformer_kwargs = self._get_transformer_kwargs(transformer)
+            return transformer_class(**{**default_transformer_kwargs, **kwargs})
 
         return deepcopy(transformer)
 

Diff for: tests/unit/data_processing/test_data_processor.py

@@ -1033,6 +1033,30 @@ def test__get_transformer_instance_kwargs(self):
         assert isinstance(result, FloatFormatter)
         assert result.computer_representation == 'Int32'
 
+    def test__get_transformer_instance_passes_kwargs_from_default(self):
+        """Test the ``_get_transformer_instance`` uses the default transformers kwargs.
+
+        Test than when the default transformer has custom kwargs, they are also used
+        when creating a new instance of a transformer.
+        """
+        # Setup
+        dp = DataProcessor(SingleTableMetadata())
+        dp._transformers_by_sdtype['numerical'] = FloatFormatter(
+            missing_value_replacement='random',
+            missing_value_generation='from_column',
+            learn_rounding_scheme=False
+        )
+
+        # Run
+        result = dp._get_transformer_instance(
+            'numerical', {'computer_representation': 'Int32'})
+
+        # Assert
+        assert isinstance(result, FloatFormatter)
+        assert result.missing_value_replacement == 'random'
+        assert result.missing_value_generation == 'from_column'
+        assert result.learn_rounding_scheme is False
+
     @patch('sdv.data_processing.data_processor.LOGGER')
     @patch('sdv.data_processing.data_processor.rdt')
     def test__update_constraint_transformers(self, mock_rdt, mock_log):