Address FutureWarning re fillna silent no_silent_downcasting

Author: sorenmacbeth

src/pytorch_tabular/categorical_encoders.py

@@ -4,7 +4,7 @@
 # Modified https://github.com/tcassou/mlencoders/blob/master/mlencoders/base_encoder.py to suit NN encoding
 """Category Encoders."""
 
-from pandas import DataFrame, Series, unique
+from pandas import DataFrame, Series, option_context, unique
 
 try:
     import cPickle as pickle
@@ -65,10 +65,12 @@ def transform(self, X):
         category_cols = X_encoded.select_dtypes(include="category").columns
         X_encoded[category_cols] = X_encoded[category_cols].astype("object")
         for col, mapping in self._mapping.items():
-            X_encoded[col] = X_encoded[col].fillna(NAN_CATEGORY).map(mapping["value"])
+            with option_context("future.no_silent_downcasting", True):
+                X_encoded[col] = X_encoded[col].fillna(NAN_CATEGORY).infer_objects(copy=False).map(mapping["value"])
 
             if self.handle_unseen == "impute":
-                X_encoded[col] = X_encoded[col].fillna(self._imputed)
+                with option_context("future.no_silent_downcasting", True):
+                    X_encoded[col] = X_encoded[col].fillna(self._imputed).infer_objects(copy=False)
             elif self.handle_unseen == "error":
                 if np.unique(X_encoded[col]).shape[0] > mapping.shape[0]:
                     raise ValueError(f"Unseen categories found in `{col}` column.")
@@ -157,7 +159,12 @@ def fit(self, X, y=None):
                 not X[self.cols].isnull().any().any()
             ), "`handle_missing` = `error` and missing values found in columns to encode."
         for col in self.cols:
-            map = Series(unique(X[col].fillna(NAN_CATEGORY)), name=col).reset_index().rename(columns={"index": "value"})
+            with option_context("future.no_silent_downcasting", True):
+                map = (
+                    Series(unique(X[col].fillna(NAN_CATEGORY).infer_objects(copy=False)), name=col)
+                    .reset_index()
+                    .rename(columns={"index": "value"})
+                )
             map["value"] += 1
             self._mapping[col] = map.set_index(col)
 

src/pytorch_tabular/tabular_datamodule.py

@@ -303,11 +303,15 @@ def _update_config(self, config) -> InferredConfig:
         else:
             raise ValueError(f"{config.task} is an unsupported task.")
         if self.train is not None:
-            category_cols = self.train[config.categorical_cols].select_dtypes(include="category").columns
-            self.train[category_cols] = self.train[category_cols].astype("object")
-            categorical_cardinality = [
-                int(x) + 1 for x in list(self.train[config.categorical_cols].fillna("NA").nunique().values)
-            ]
+            with pd.option_context("future.no_silent_downcasting", True):
+                category_cols = self.train[config.categorical_cols].select_dtypes(include="category").columns
+                self.train[category_cols] = self.train[category_cols].astype("object")
+                categorical_cardinality = [
+                    int(x) + 1
+                    for x in list(
+                        self.train[config.categorical_cols].fillna("NA").infer_objects(copy=False).nunique().values
+                    )
+                ]
         else:
             category_cols = self.train_dataset.data[config.categorical_cols].select_dtypes(include="category").columns
             self.train_dataset.data[category_cols] = self.train_dataset.data[category_cols].astype("object")