make_ds_gutenberg.py

from glob import glob
import json
import os

from datasets import Dataset


def main():
    workdir = "./gutenberg"
    books = glob(os.path.join(workdir, "books", "*.json"))
    
    def gen():
        for book in books:
            with open(book, "r", encoding="utf-8") as reader:
                data = json.load(reader)
            data["Bookshelf"] = str(data["Bookshelf"])
            data["Author"] = str(data["Author"])
            yield data

    ds = Dataset.from_generator(gen)
    column_names = list(ds.features.keys())
    new_names = {k: k.lower() for k in column_names}
    ds = ds.rename_columns(new_names)

    ds.save_to_disk(os.path.join(workdir, "datasets", "data", "train"), max_shard_size="400MB")
    return


if __name__ == "__main__":
    main()