-
Notifications
You must be signed in to change notification settings - Fork 0
/
figure_1_runs.py
43 lines (33 loc) · 1.16 KB
/
figure_1_runs.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
import pyterrier as pt
pt.init()
from pyterrier_t5 import MonoT5ReRanker
import argparse
import torch
BATCH_SIZE = 16
def main():
torch.manual_seed(0)
parser = argparse.ArgumentParser()
parser.add_argument('dataset')
parser.add_argument('model')
parser.add_argument('out')
parser.add_argument('--initial')
parser.add_argument('--query_field')
parser.add_argument('--passage', action='store_true')
args = parser.parse_args()
dataset = pt.get_dataset(args.dataset)
reranker = MonoT5ReRanker(model=args.model, batch_size=16)
if args.passage:
# apply passaging
reranker = pt.text.sliding(text_attr='text', prepend_attr=None) >> reranker >> pt.text.max_passage()
pipeline = pt.text.get_text(dataset, 'text') >> reranker
if args.initial == None:
# initial ranking provided by dataset
inp = dataset.get_results(args.query_field).sort_values(by=['qid'])
else:
# read initial ranking from file
inp = pt.io.read_results(args.initial, topics=dataset.get_topics(args.query_field))
res = pipeline(inp)
#pt.io.write_results(res, args.model+'.dev-small.res')
pt.io.write_results(res, args.out)
if __name__ == '__main__':
main()