hestenet-qa / generate_dataset.py
kardosdrur's picture
Added script for generating the dataset from Danish Gigaword
b2eb1e8
raw
history blame
1.19 kB
from functools import partial
from glob import glob
from typing import Iterable, Optional
from datasets import Dataset
def is_query(text: str):
n_question = text.count("?")
n_statement = text.count(".")
if n_question + n_statement < 2:
return False
if n_question > (n_statement * 0.5):
return True
def first_answer(answers: list[str]) -> Optional[str]:
for answer in answers:
if answer and (not is_query(answer)) and ("tak" not in answer.lower()):
return answer
return None
def generate_dataset(threads: list[list[str]]) -> Iterable[dict]:
for thread in threads:
question = thread[0]
if is_query(question):
answer = first_answer(thread[1:])
if answer:
yield {"question": question, "answer": answer}
def main():
files = glob("dagw/sektioner/hest/hest_forum*")
threads = []
for file in files:
with open(file) as in_file:
threads.append(in_file.read().split("\n"))
dataset = Dataset.from_generator(partial(generate_dataset, threads))
dataset.push_to_hub("kardosdrur/hestenet-qa")
if __name__ == "__main__":
main()