Files
mlm-scoring/examples/lingacc-blimp/convert_to_dataset.py
2020-05-15 01:18:56 -07:00

28 lines
1.0 KiB
Python
Executable File

#!/usr/bin/env python3
import argparse
import logging
import json
from pathlib import Path
if __name__ == '__main__':
parser = argparse.ArgumentParser(description="Convert BLiMP files into text")
parser.add_argument('--input-dir', type=str, default='blimp', help="BLiMP directory")
parser.add_argument('--output-dir', type=str, default='data-concat', help="Output directory for .txt files")
args = parser.parse_args()
input_dir = Path(args.input_dir) / 'data'
output_dir = Path(args.output_dir)
output_dir.mkdir(parents=True, exist_ok=True)
good_file = output_dir / 'good.txt'
bad_file = output_dir / 'bad.txt'
with good_file.open('wt') as f_good, bad_file.open('wt') as f_bad:
for jsonl in input_dir.glob('*.jsonl'):
logging.warn("{}".format(jsonl))
lines = [json.loads(line) for line in jsonl.read_text().split('\n') if len(line.strip())]
for line in lines:
f_good.write(line['sentence_good'] + '\n')
f_bad.write(line['sentence_bad'] + '\n')