Fixed Bug for multi_eurlex

lextreme.py

@@ -52,6 +52,7 @@ class LextremeConfig(datasets.BuilderConfig):
             hf_hub_name,
             language="multilingual",
             config_name="default",
+            label_level=None,
             label_classes=None,
             task_type=None,
             **kwargs,
@@ -67,6 +68,7 @@ class LextremeConfig(datasets.BuilderConfig):
           hf_hub_name: `string`, huggingface dataset identifier
           hf_hub_name: `string`, the language of the dataset (either multilingual or a single language ISO code)
           config_name: `string`, huggingface dataset config name
+          label_level: `string`, the label level (only necessary for multieurlex)
           label_classes: `list[string]`, the list of classes if the label is
             categorical. If not provided, then the label will be of type
             `datasets.Value('float32')`.
@@ -82,6 +84,7 @@ class LextremeConfig(datasets.BuilderConfig):
         self.hf_hub_name = hf_hub_name
         self.language = language
         self.config_name = config_name
+        self.label_level = label_level
         self.label_classes = label_classes
         self.task_type = task_type
 
@@ -3041,6 +3044,7 @@ _MULTI_EURLEX_LEVEL_1 = {
         }
         """
     ,
+    "label_level": "level_1",
     "label_classes": [
         "100149",
         "100160",
@@ -3067,6 +3071,7 @@ _MULTI_EURLEX_LEVEL_1 = {
 }
 _MULTI_EURLEX_LEVEL_2 = {
     **_MULTI_EURLEX_LEVEL_1,
+    "label_level": "level_2",
     "label_classes": [
         "100215",
         "100211",
@@ -3199,6 +3204,7 @@ _MULTI_EURLEX_LEVEL_2 = {
 }
 _MULTI_EURLEX_LEVEL_3 = {
     **_MULTI_EURLEX_LEVEL_1,
+    "label_level": "level_3",
     "label_classes": [
         "1754",
         "5881",
@@ -4064,7 +4070,10 @@ class LEXTREME(datasets.GeneratorBasedBuilder):
     def _generate_examples(self, split):
         """This function returns the examples in the raw (text) form."""
         # we can just do this, since all our datasets are available on the huggingface hub
-        dataset = datasets.load_dataset(self.config.hf_hub_name, self.config.config_name, split=split)
+        dataset = datasets.load_dataset(self.config.hf_hub_name,
+                                        self.config.config_name,
+                                        label_level=self.config.label_level,
+                                        split=split)
         for id, item in enumerate(dataset):
             # In case we have a class label and not just a string: convert it back to the string
             label_col = self.config.label_col