debugging

docbank.py

@@ -133,6 +133,9 @@ class Docbank(datasets.GeneratorBasedBuilder):
 
     # DEFAULT_CONFIG_NAME = "small"  # It's not mandatory to have a default configuration. Just use one if it make sense.
     TOKENIZER = LayoutXLMTokenizerFast.from_pretrained("microsoft/layoutxlm-base")
+    LABELS = ['abstract', 'author', 'caption', 'date', 'equation', 'figure', 'footer', 'list', 'paragraph', 'reference', 'section', 'table', 'title']
+    ID2LABEL = {k: v for k, v in enumerate(LABELS)}
+    LABEL2ID = {v: k for k, v in enumerate(LABELS)}
 
     def _info(self):
         # TODO: This method specifies the datasets.DatasetInfo object which contains informations and typings for the dataset
@@ -230,7 +233,7 @@ class Docbank(datasets.GeneratorBasedBuilder):
                         normalized_bbox = [int(x) for x in row[1:5]]
                         tokens.append(row[0])
                         bboxes.append(normalized_bbox)
-                        labels.append(row[9])
+                        labels.append(self.LABEL2ID(row[9]))
                         # tokenized_input = self.TOKENIZER(
                         #     row[0],
                         #     add_special_tokens=False,