create_examples_dataset.py

import ftfy
import json
import random
random.seed(1)
import argparse
import ir_datasets
from tqdm.auto import tqdm

def load_examples(base_dataset):
    dataset = ir_datasets.load(base_dataset)

    documents = {}
    for doc in tqdm(
        dataset.docs_iter(), total=dataset.docs_count(), desc="Loading documents"
    ):
        text = (
            f"{doc.title} {doc.text}"
            if "title" in dataset.docs_cls()._fields
            else doc.text
        )
        documents[doc.doc_id] = ftfy.fix_text(text)

    queries = {}
    for query in dataset.queries_iter():
        queries[query.query_id] = ftfy.fix_text(query.text)

    return [
        (queries[qrel.query_id], documents[qrel.doc_id])
        for qrel in dataset.qrels_iter()
    ]

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument('--num_examples', default=500_000)
    parser.add_argument('--dataset', default='beir/msmarco/train')
    parser.add_argument('--output', required=True)
    args = parser.parse_args()

    examples = load_examples(args.dataset)
    random.shuffle(examples)

    with open(args.output, 'w') as f:
        for (query, document) in tqdm(examples[:args.num_examples], total=args.num_examples, desc='Writing'):
            f.write(json.dumps({'query': query, 'document': document}) + '\n')