generate_reference_files.py

import os
from glob import glob
from datasets import load_dataset

os.makedirs('./reference_files', exist_ok=True)


for split in ['validation', 'test']:
    for domain in ["default", "books", "electronics", "grocery", "movies", "restaurants", "tripadvisor"]:
        dataset = load_dataset('asahi417/qg_subjqa', domain, split=split)
        for data in ['question', 'answer', 'sentence', 'paragraph']:
            with open('./reference_files/{}-{}.{}txt'.format(data, split, "" if domain == 'default' else f"{domain}."), 'w') as f:
                f.write('\n'.join(dataset[data]))
        if domain != 'default':
            length = [len(open(i).read().split('\n')) for i in glob(f'reference_files/*{split}*{domain}*.txt')]
            assert len(list(set(length))) == 1, length