rlyapin/OpenTriviaQA

Unofficial port of https://github.com/uberspot/OpenTriviaQA

List of all available dataset splits:

['animals', 'brain-teasers', 'celebrities', 'entertainment', 'for-kids', 'general', 'geography', 'history', 'hobbies', 'humanities', 'literature', 'movies', 'music', 'newest', 'people', 'rated', 'religion-faith', 'science-technology', 'sports', 'television', 'video-games', 'world']

Code to reproduce the dataset:

import os
import pandas as pd
import pyarrow as pa
import pyarrow.parquet as pq

source_dir = "OpenTriviaQA/categories"
topics = sorted(os.listdir(source_dir))
target_dir = "target/"

def process_question(text, topic, i):
    try:
        question, options = text.strip().split("\n^ ")
    
        assert question[:3] == "#Q "
        question = question[3:]
      
        lines = options.split("\n")
        lines = [x for x in lines if x]
    
        answer = lines[0]
        choices = [x[2:] for x in lines[1:]]
        assert answer in choices
    
        return {"id": f"{topic[:3]}-{i}", "question": question, "choices": choices, "answer": answer}
    except:
        return {}

for topic in topics:
    filename = os.path.join(source_dir, topic)
    data = open(filename, encoding = "ISO-8859-1").read()

    questions = data.split("\n\n")
    questions = [x for x in questions if x]

    questions = [process_question(x, topic, i) for (i, x) in enumerate(questions)]
    questions = [x for x in questions if x]

    data = pa.Table.from_pandas(pd.DataFrame(questions))

    os.makedirs(os.path.join(target_dir, topic), exist_ok=True) 
    pq.write_table(data, os.path.join(target_dir, topic, "data.parquet"))