Fix invalid entity start/end indices for invalid examples

kbp37.py

@@ -217,15 +217,18 @@ class KBP37(datasets.GeneratorBasedBuilder):
                         subj_end = tokens.index("</e1>")
                         tokens.pop(subj_end)
 
-                    yield int(id_), {
-                        "id": id_,
-                        "token": tokens,
-                        "subj_start": subj_start,
-                        "subj_end": subj_end,
-                        "obj_start": obj_start,
-                        "obj_end": obj_end,
-                        "relation": relation,
-                    }
+                    # some examples, like train/1276 are invalid (empty head/tail), and
+                    # yield non-sensical examples without this check
+                    if subj_end > subj_start and obj_end > obj_start:
+                        yield int(id_), {
+                            "id": id_,
+                            "token": tokens,
+                            "subj_start": subj_start,
+                            "subj_end": subj_end,
+                            "obj_start": obj_start,
+                            "obj_end": obj_end,
+                            "relation": relation,
+                        }
                 else:
                     yield int(id_), {
                         "id": id_,