enforce float32 dtype for target encoded features.

ml_garden/core/steps/encode.py

@@ -171,6 +171,8 @@ def _apply_encoding(
 
         encoded_data = self._restore_column_order(df, encoded_data)
         encoded_data = self._restore_numeric_dtypes(encoded_data, original_numeric_dtypes)
+        encoded_data = self._restore_target_encoded_dtypes(encoded_data, encoder)
+
         encoded_data = self._convert_float64_to_float32(encoded_data)
 
         feature_encoder_map = self._create_feature_encoder_map(encoder)
@@ -368,6 +370,22 @@ def _restore_numeric_dtypes(
                     )
         return encoded_data
 
+    def _restore_target_encoded_dtypes(self, encoded_data, encoder):
+        """Convert the columns handled by any TargetEncoder in the given encoder to float32."""
+        for name, transformer, cols in encoder.transformers:
+            # Direct TargetEncoder
+            if type(transformer).__name__ == "TargetEncoder":
+                for col in cols:
+                    encoded_data[col] = encoded_data[col].astype("float32")
+
+            # Nested transformers (if inside pipelines or additional ColumnTransformers)
+            elif hasattr(transformer, "transformers"):
+                for nested_name, nested_transformer, nested_cols in transformer.transformers:
+                    if type(nested_transformer).__name__ == "TargetEncoder":
+                        for col in nested_cols:
+                            encoded_data[col] = encoded_data[col].astype("float32")
+        return encoded_data
+
     def _convert_float64_to_float32(self, encoded_data: pd.DataFrame) -> pd.DataFrame:
         """Convert float64 columns to float32."""
         float64_columns = encoded_data.select_dtypes(include=["float64"]).columns