supportboth s2s and s2p

spanish_passage_retrieval.py

@@ -37,7 +37,7 @@ class PRES(datasets.GeneratorBasedBuilder):
             name=name,
             description=f'{name.title()} of the Spanish Passage Retrieval dataset.',
         )
-        for name in ['corpus', 'queries', 'qrels']
+        for name in ['corpus.sentences', 'corpus.documents', 'queries', 'qrels.s2s', 'qrels.s2p']
     ]
 
     BUILDER_CONFIG_CLASS = PRESConfig
@@ -91,9 +91,14 @@ class PRES(datasets.GeneratorBasedBuilder):
                 with open(os.path.join(extracted_path, 'relevance_passages.json')) as f:
                     rel_passages = json.load(f)
 
-            corpus = []
+                with open(os.path.join(extracted_path, 'relevance_documents.tsv')) as f:
+                    rel_docs = [(item, item[2]) for item in list(csv.reader(f))[1:]]
+
+            corpus_sentences = []
+            corpus_documents = []
             queries = dict()
-            qrels = dict()
+            qrels_s2s = dict()
+            qrels_s2p = dict()
             topic_to_queries = dict()
             for topic in topics['topics']:
                 topic_to_queries[topic['number']] = []
@@ -101,24 +106,38 @@ class PRES(datasets.GeneratorBasedBuilder):
                     qid = query['number']
                     queries[qid] = query['text']
                     topic_to_queries[topic['number']].append(qid)
-                    qrels[qid] = []
+                    qrels_s2s[qid] = []
+                    qrels_s2p[qid] = []
 
             for annotated_topic in rel_passages['topics']:
                 topic = annotated_topic['number']
                 for annotation in annotated_topic['annotations']:
                     did = f'doc_{annotation["docNo"]}_{annotation["start"]}_{annotation["end"]}'
-                    corpus.append({'_id': did, 'text': annotation['text']})
+                    corpus_sentences.append({'_id': did, 'text': annotation['text']})
                     for qid in topic_to_queries[topic]:
-                        qrels[qid].append(did)
+                        qrels_s2s[qid].append(did)
+
+            for doc in docs['documents']:
+                did = f'doc_{doc["docId"]}'
+                corpus_documents.append({'_id': did, 'text': doc['text']})
+
+            for topic_id, doc_id in rel_docs:
+                for q_num in topic_to_queries[topic_id]:
+                    qrels_s2p[f'PR_ES_{q_num}'].append(f'doc_{doc_id}')
 
             self._data = {
-                'corpus': corpus,
+                'corpus.sentences': corpus_sentences,
+                'corpus.documents': corpus_documents,
                 'queries': queries,
-                'qrels': qrels,
+                'qrels.s2s': qrels_s2s,
+                'qrels.s2p': qrels_s2p
             }
 
-        if self.config.name == 'corpus':
-            for line in self._data['corpus']:
+        if self._config.name not in self._data:
+            raise ValueError(f'Unknown config name: {self.config.name}')
+
+        if self.config.name.startswith('corpus'):
+            for line in self._data[self.config.name]:
                 yield line['_id'], line
         elif self.config.name == 'queries':
             for qid, query in self._data['queries'].items():
@@ -126,11 +145,9 @@ class PRES(datasets.GeneratorBasedBuilder):
                     "_id": qid,
                     "text": query,
                 }
-        elif self.config.name == 'qrels':
-            for qid, dids in self._data['qrels'].items():
+        elif self.config.name.startswith('qrels'):
+            for qid, dids in self._data[self.config.name].items():
                 yield qid, {
                     "_id": qid,
                     "text": ' '.join(dids),
-                }
-        else:
-            raise ValueError(f'Unknown config name: {self.config.name}')
+                }