fix

sdgx/data_models/inspectors/numeric.py

@@ -171,7 +171,9 @@ def fit(self, raw_data: pd.DataFrame, *args, **kwargs):
 
         # Iterate all columns and determain the final data type
         for col in raw_data.columns:
-            if raw_data[col].dtype in ["int64", "float64"]:
+            if (pd.api.types.is_integer_dtype(raw_data[col].dtype)
+                    or pd.api.types.is_float_dtype(raw_data[col].dtype)):
+                # series type may be 32/64bit.
                 # float or int
                 if self._is_int_column(raw_data[col]):
                     self.int_columns.add(col)

sdgx/data_models/metadata.py

@@ -96,6 +96,15 @@ def check_categorical_threshold(self, num_categories):
             else False
         )
 
+    def get_column_encoder(self, column_name):
+        encoder_type = None
+        if (
+                self.categorical_encoder
+                and column_name in self.categorical_encoder
+        ):
+            encoder_type = self.categorical_encoder[column_name]
+        return encoder_type
+
     @property
     def tag_fields(self) -> Iterable[str]:
         """

sdgx/models/components/optimize/sdv_ctgan/data_transformer.py

@@ -90,7 +90,7 @@ def _fit_discrete(self, data, encoder_type: CategoricalEncoderType = None):
         num_categories = len(encoder.dummies)
         activate_fn = "softmax"
 
-        checked = self.metadata.check_categorical_threshold(num_categories)
+        checked = self.metadata.check_categorical_threshold(num_categories) if self.metadata else False
         if encoder_type == "onehot" or not checked:
             pass
         elif encoder_type == "label":
@@ -127,12 +127,7 @@ def fit(self, data_loader: DataLoader, discrete_columns=()):
             if column_name in discrete_columns:
                 #  or column_name in self.metadata.label_columns
                 logger.debug(f"Fitting discrete column {column_name}...")
-                encoder_type = None
-                if (
-                    self.metadata.categorical_encoder
-                    and column_name in self.metadata.categorical_encoder
-                ):
-                    encoder_type = self.metadata.categorical_encoder[column_name]
+                encoder_type = self.metadata.get_column_encoder(column_name) if self.metadata else None
                 column_transform_info = self._fit_discrete(data_loader[[column_name]], encoder_type)
             else:
                 logger.debug(f"Fitting continuous column {column_name}...")

tests/data_models/test_metadata.py

@@ -1,7 +1,6 @@
 from __future__ import annotations
 
 from pathlib import Path
-
 import pytest
 
 from sdgx.data_connectors.csv_connector import CsvConnector

tests/test_ctgan_synthesizer.py

@@ -41,7 +41,7 @@ def demo_single_table_data_pos_neg():
 
 @pytest.fixture
 def demo_single_table_data_pos_neg_metadata(demo_single_table_data_pos_neg):
-    yield Metadata.from_dataframe(demo_single_table_data_pos_neg)
+    yield Metadata.from_dataframe(demo_single_table_data_pos_neg.copy(), check=True)
 
 
 @pytest.fixture
@@ -82,7 +82,6 @@ def test_ctgan_synthesizer_with_pos_neg(
     demo_single_table_data_pos_neg,
 ):
     original_data = demo_single_table_data_pos_neg
-    metadata = demo_single_table_data_pos_neg_metadata
 
     # Train the CTGAN model
     ctgan_synthesizer.fit(demo_single_table_data_pos_neg_metadata)