Fix feature type detector

ThomasMeissnerDS · Aug 3, 2024 · 68a1e1f · 68a1e1f
1 parent 5d29d05
commit 68a1e1f
Show file tree

Hide file tree

Showing 4 changed files with 8 additions and 19 deletions.
diff --git a/bluecast/general_utils/general_utils.py b/bluecast/general_utils/general_utils.py
@@ -182,14 +182,6 @@ def save_out_of_fold_data(
 
         y_true = y_true.astype(int)
 
-        print("DEBUUUUUUUUUUUUUUUUG")
-        print(y_true.shape)
-        print(y_true)
-
-        print("DEBUUUUUUUUUUUUUUUUG AGAIN")
-        print(y_hat.shape)
-        print(y_hat)
-
         oof_data_copy["predicted_class"] = y_classes
         oof_data_copy["target_class_predicted_probas"] = [
             1 - preds if cls == 0 else preds for preds, cls in zip(y_hat, y_classes)
@@ -211,14 +203,6 @@ def save_out_of_fold_data(
 
         y_true = y_true.astype(int)
 
-        print("DEBUUUUUUUUUUUUUUUUG MULTICLASS")
-        print(y_true.shape)
-        print(y_true)
-
-        print("DEBUUUUUUUUUUUUUUUUG MULTICLASS AGAIN")
-        print(y_hat.shape)
-        print(y_hat)
-
         oof_data_copy["predicted_class"] = y_classes
         oof_data_copy["target_class_predicted_probas"] = np.asarray(
             [pred[target_cls] for pred, target_cls in zip(y_hat, y_true)]

diff --git a/bluecast/ml_modelling/xgboost.py b/bluecast/ml_modelling/xgboost.py
@@ -748,7 +748,7 @@ def objective(trial):
                     shuffle=self.conf_training.shuffle_during_training,
                 )
                 folds = []
-                for train_index, test_index in skf.split(x_train, y_train):
+                for train_index, test_index in skf.split(x_train, y_train.astype(int)):
                     folds.append((train_index.tolist(), test_index.tolist()))
 
                 result = xgb.cv(

diff --git a/bluecast/preprocessing/feature_types.py b/bluecast/preprocessing/feature_types.py
@@ -229,6 +229,11 @@ def fit_transform_feature_types(self, df: pd.DataFrame) -> pd.DataFrame:
         bool_cols, no_bool_cols = self.identify_bool_columns(df_clean)
         self.identify_date_time_columns(df_clean, no_bool_cols)
         df_clean = self.cast_rest_columns_to_object(df_clean, bool_cols)
+        for key in self.detected_col_types:
+            if self.detected_col_types[key] == "datetime[ns]":
+                df_clean[key] = pd.to_datetime(df[key], yearfirst=True)
+            else:
+                df_clean[key] = df_clean[key].astype(self.detected_col_types[key])
         return df_clean
 
     def transform_feature_types(
@@ -246,7 +251,7 @@ def transform_feature_types(
         for key in self.detected_col_types:
             if ignore_cols and key not in ignore_cols and key in df_clean.columns:
                 if self.detected_col_types[key] == "datetime[ns]":
-                    df_clean[key] = pd.to_datetime(df[key], yearfirst=True)
+                    df_clean[key] = pd.to_datetime(df_clean[key], yearfirst=True)
                 else:
                     df_clean[key] = df_clean[key].astype(self.detected_col_types[key])
         return df_clean
diff --git a/bluecast/tests/test_feature_type_detector.py b/bluecast/tests/test_feature_type_detector.py
@@ -59,4 +59,4 @@ def test_feature_type_detector():
     assert transformed_df["categorical_col2"].dtype == np.object_
 
     # Ensure the transformed dataframe is not the same as the original dataframe (transformed datetime columns)
-    assert transformed_df.equals(df)
+    assert not transformed_df.equals(df)