create_nqii.py

# /// script
# requires-python = ">=3.10,<4.0"
# dependencies = [
#     "datasets==3.5.0",
#     "huggingface-hub==0.24.0",
#     "pandas==2.2.0",
#     "requests==2.32.3",
# ]
# ///

"""Create the NQiI-mini dataset and upload them to the HF Hub."""

import pandas as pd
from datasets.arrow_dataset import Dataset
from datasets.combine import concatenate_datasets
from datasets.dataset_dict import DatasetDict
from datasets.load import load_dataset
from datasets.splits import Split
from huggingface_hub.hf_api import HfApi

from .constants import (
    MAX_NUM_CHARS_IN_CONTEXT,
    MAX_NUM_CHARS_IN_QUESTION,
    MIN_NUM_CHARS_IN_CONTEXT,
    MIN_NUM_CHARS_IN_QUESTION,
)


def main() -> None:
    """Create the NQiI-mini dataset and upload them to the HF Hub."""
    dataset_id = "vesteinn/icelandic-qa-NQiI"

    # Load the datasets from the `alexandrainst` organisation
    train = load_dataset(dataset_id, split="train", token=True)
    val = load_dataset(dataset_id, split="validation", token=True)
    test = load_dataset(dataset_id, split="test", token=True)

    # Ensure that the datasets are indeed datasets
    assert isinstance(train, Dataset)
    assert isinstance(val, Dataset)
    assert isinstance(test, Dataset)

    # Merge the splits
    df = concatenate_datasets([train, val, test]).to_pandas()

    # Ensure that `df` is indeed a Pandas DataFrame
    assert isinstance(df, pd.DataFrame)

    # Only work with samples where the context is not very large or small
    lengths = df.context.str.len()
    df = df[lengths.between(MIN_NUM_CHARS_IN_CONTEXT, MAX_NUM_CHARS_IN_CONTEXT)]

    # Only work with samples where the question is not very large or small
    lengths = df.question.str.len()
    df = df[lengths.between(MIN_NUM_CHARS_IN_QUESTION, MAX_NUM_CHARS_IN_QUESTION)]
    assert isinstance(df, pd.DataFrame)

    # Extract information on which examples contain an answer
    has_answer: pd.Series = df.answers.map(
        lambda dct: len(dct["text"]) > 0 and dct["text"][0] != ""
    )

    # Only work with the questions having answers in the context
    df_with_answer: pd.DataFrame = df.loc[has_answer]

    # Create validation split
    val_size = 256
    val_df = df_with_answer.sample(n=val_size, random_state=4242)

    # Create test split
    test_size = 1024
    df_with_answer_filtered: pd.DataFrame = df_with_answer.loc[
        ~df_with_answer.index.isin(val_df.index)
    ]
    test_df = df_with_answer_filtered.sample(n=test_size, random_state=4242)

    # Create train split
    train_size = 1024
    full_train_df_with_answer = df_with_answer_filtered.loc[
        ~df_with_answer_filtered.index.isin(test_df.index)
    ]
    train_df = full_train_df_with_answer.sample(n=train_size, random_state=4242)

    val_df = val_df.reset_index(drop=True)
    test_df = test_df.reset_index(drop=True)
    train_df = train_df.reset_index(drop=True)

    assert isinstance(train_df, pd.DataFrame)
    assert isinstance(val_df, pd.DataFrame)
    assert isinstance(test_df, pd.DataFrame)

    # Collect datasets in a dataset dictionary
    dataset = DatasetDict(
        {
            "train": Dataset.from_pandas(train_df, split=Split.TRAIN),
            "val": Dataset.from_pandas(val_df, split=Split.VALIDATION),
            "test": Dataset.from_pandas(test_df, split=Split.TEST),
        }
    )

    # Push the dataset to the Hugging Face Hub
    dataset_id = "EuroEval/nqii-mini"
    HfApi().delete_repo(dataset_id, repo_type="dataset", missing_ok=True)
    dataset.push_to_hub(dataset_id, private=True)


if __name__ == "__main__":
    main()