First commit
Browse files- .gitattributes +1 -0
- create_examples_dataset.py +44 -0
- data/msmarco.json +3 -0
.gitattributes
CHANGED
@@ -51,3 +51,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
51 |
*.jpg filter=lfs diff=lfs merge=lfs -text
|
52 |
*.jpeg filter=lfs diff=lfs merge=lfs -text
|
53 |
*.webp filter=lfs diff=lfs merge=lfs -text
|
|
|
|
51 |
*.jpg filter=lfs diff=lfs merge=lfs -text
|
52 |
*.jpeg filter=lfs diff=lfs merge=lfs -text
|
53 |
*.webp filter=lfs diff=lfs merge=lfs -text
|
54 |
+
*.json filter=lfs diff=lfs merge=lfs -text
|
create_examples_dataset.py
ADDED
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import ftfy
|
2 |
+
import json
|
3 |
+
import random
|
4 |
+
random.seed(1)
|
5 |
+
import argparse
|
6 |
+
import ir_datasets
|
7 |
+
from tqdm.auto import tqdm
|
8 |
+
|
9 |
+
def load_examples(base_dataset):
|
10 |
+
dataset = ir_datasets.load(base_dataset)
|
11 |
+
|
12 |
+
documents = {}
|
13 |
+
for doc in tqdm(
|
14 |
+
dataset.docs_iter(), total=dataset.docs_count(), desc="Loading documents"
|
15 |
+
):
|
16 |
+
text = (
|
17 |
+
f"{doc.title} {doc.text}"
|
18 |
+
if "title" in dataset.docs_cls()._fields
|
19 |
+
else doc.text
|
20 |
+
)
|
21 |
+
documents[doc.doc_id] = ftfy.fix_text(text)
|
22 |
+
|
23 |
+
queries = {}
|
24 |
+
for query in dataset.queries_iter():
|
25 |
+
queries[query.query_id] = ftfy.fix_text(query.text)
|
26 |
+
|
27 |
+
return [
|
28 |
+
(queries[qrel.query_id], documents[qrel.doc_id])
|
29 |
+
for qrel in dataset.qrels_iter()
|
30 |
+
]
|
31 |
+
|
32 |
+
if __name__ == "__main__":
|
33 |
+
parser = argparse.ArgumentParser()
|
34 |
+
parser.add_argument('--num_examples', default=500_000)
|
35 |
+
parser.add_argument('--dataset', default='beir/msmarco/train')
|
36 |
+
parser.add_argument('--output', required=True)
|
37 |
+
args = parser.parse_args()
|
38 |
+
|
39 |
+
examples = load_examples(args.dataset)
|
40 |
+
random.shuffle(examples)
|
41 |
+
|
42 |
+
with open(args.output, 'w') as f:
|
43 |
+
for (query, document) in tqdm(examples[:args.num_examples], total=args.num_examples, desc='Writing'):
|
44 |
+
f.write(json.dumps({'query': query, 'document': document}) + '\n')
|
data/msmarco.json
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:1562c5128f6cf73a089023a57a323b2ca58585a25e2d4b9621701091a3be1895
|
3 |
+
size 207348070
|