Added 'test_collection'

test.py

@@ -1,4 +1,8 @@
 from datasets import load_dataset
+
+collection = load_dataset('trec-cast-2019-multi-turn.py')
+collection.items()
+
 qrels = load_dataset('trec-cast-2019-multi-turn.py', 'qrels')
 qrels.items()
 

trec-cast-2019-multi-turn.py

@@ -20,11 +20,22 @@ import csv
 
 # Find for instance the citation on arxiv or on the dataset repo/website
 _CITATION = """\
+@misc{dalton2020trec,
+      title={TREC CAsT 2019: The Conversational Assistance Track Overview}, 
+      author={Jeffrey Dalton and Chenyan Xiong and Jamie Callan},
+      year={2020},
+      eprint={2003.13624},
+      archivePrefix={arXiv},
+      primaryClass={cs.IR}
+}
 """
 
 # You can copy an official description
 _DESCRIPTION = """\
-
+The Conversational Assistance Track (CAsT) is a new track for TREC 2019 to facilitate Conversational Information 
+Seeking (CIS) research and to create a large-scale reusable test collection for conversational search systems. 
+The document corpus is 38,426,252 passages from the TREC Complex Answer Retrieval (CAR) and Microsoft MAchine 
+Reading COmprehension (MARCO) datasets.
 """
 
 _HOMEPAGE = "http://www.treccast.ai"
@@ -37,12 +48,16 @@ _URL = "https://huggingface.co/datasets/uva-irlab/trec-cast-2019-multi-turn/reso
 _URLs = {
     'topics': _URL+"cast2019_test_annotated.tsv",
     'qrels': _URL+"2019qrels.txt",
+    'test_collection': {
+        'msmarco': 'https://msmarco.blob.core.windows.net/msmarcoranking/collection.tar.gz',
+        'car': "http://trec-car.cs.unh.edu/datareleases/v2.0/paragraphCorpus.v2.0.tar.xz",
+    }
 }
 
 
 class TrecCast2019MultiTurn(datasets.GeneratorBasedBuilder):
     """
-    Voskarides et al. have preprocessed CANARD in different ways depending on their experiment.
+
     """
 
     VERSION = datasets.Version("1.0.0")
@@ -59,12 +74,19 @@ class TrecCast2019MultiTurn(datasets.GeneratorBasedBuilder):
     # data = datasets.load_dataset('my_dataset', 'first_domain')
     # data = datasets.load_dataset('my_dataset', 'second_domain')
     BUILDER_CONFIGS = [
-        datasets.BuilderConfig(name="qrels", version=VERSION, description=""),
-        datasets.BuilderConfig(name="topics", version=VERSION, description=""),
+        datasets.BuilderConfig(name="qrels",
+                               version=VERSION,
+                               description=""),
+        datasets.BuilderConfig(name="topics",
+                               version=VERSION,
+                               description="The topics contain the queries, query IDs and their history."),
+        datasets.BuilderConfig(name="test_collection",
+                               version=VERSION,
+                               description="The test collection will provide the passages of TREC CAR and MSMARCO"),
     ]
 
     # It's not mandatory to have a default configuration. Just use one if it make sense.
-    DEFAULT_CONFIG_NAME = None
+    DEFAULT_CONFIG_NAME = "test_collection"
 
     def _info(self):
         # This is the name of the configuration selected in BUILDER_CONFIGS above
@@ -82,6 +104,10 @@ class TrecCast2019MultiTurn(datasets.GeneratorBasedBuilder):
                     'rank': datasets.Value("string"),
                 })),
             })
+        elif self.config.name == 'test_collection':
+            features = datasets.Features({
+                "docid": datasets.Value("string"),
+            })
         return datasets.DatasetInfo(
             # This is the description that will appear on the datasets page.
             description=_DESCRIPTION,
@@ -138,7 +164,7 @@ class TrecCast2019MultiTurn(datasets.GeneratorBasedBuilder):
             for qid in qrels.keys():
                 yield qid, {'qid': qid, 'qrels': qrels[qid]}
 
-        if split == 'topics':
+        elif split == 'topics':
             topics_file = csv.reader(open(file), delimiter="\t")
             topics = defaultdict(list)
             for row in topics_file:
@@ -152,3 +178,6 @@ class TrecCast2019MultiTurn(datasets.GeneratorBasedBuilder):
                     query = queries[idx]
                     qid = f"{conversation_id}_{str(idx+1)}"
                     yield qid, ({'query': query, 'history': queries[:idx], 'qid': qid})
+
+        else:
+            raise NotImplementedError(f"'{split}' is not yet implemented")