Fix: use pd.Series(...).value_counts instead of pd.value_counts(...) (#213)

Signed-off-by: Arthur Lacote <arthur.lacote@probabl.ai>

Author: cakedev0

sklbench/datasets/loaders.py

@@ -42,8 +42,8 @@ def load_openml_data(
 ) -> Tuple[Dict, Dict]:
     x, y = load_openml(openml_id, raw_data_cache)
     data_desc = dict()
-    unique_labels = dict(pd.value_counts(y))
-    if len(unique_labels) < 32 and all(map(lambda x: x > 4, unique_labels.values())):
+    unique_labels = pd.Series(y).value_counts()
+    if len(unique_labels) < 32 and (unique_labels > 4).all():
         data_desc["n_classes"] = len(unique_labels)
     return {"x": x, "y": y}, data_desc
 

sklbench/utils/special_params.py

@@ -214,7 +214,7 @@ def assign_case_special_values_on_run(
         and estimator.endswith("Classifier")
     ):
         y_train = convert_to_numpy(data[1])
-        value_counts = pd.value_counts(y_train).sort_index()
+        value_counts = pd.Series(y_train).value_counts().sort_index()
         if len(value_counts) != 2:
             logger.info(
                 f"Number of classes ({len(value_counts)}) != 2 "